DETAILED ACTION
Remarks
Claims 1-20 have been examined and rejected. This Office action is responsive to the amendment filed on 08/26/2026, which has been entered in the above identified application.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-3 and 5-19 are rejected under 35 U.S.C. 103 as being unpatentable over Ramchandran et al. (US 20190146647 A1, published 05/16/2019), hereinafter Ramchandran, in view of Ghoche et al. (US 20240386214 A1, published 11/21/2024), hereinafter Ghoche.
Regarding claim 1, Ramchandran teaches the claim comprising:
A method for facilitating electronic safety alert communications by a safety alert management system, the method comprising: electronically receiving a safety alert for a specific safety event from a user electronic device, the safety alert received at a safety alert management application and the safety alert comprising at least one user message from a user associated with the user electronic device and the safety alert comprising additional user data associated with the user (Ramchandran Figs. 1-7; [0021], FIG. 1 is an example representation 100 of a human agent 102 engaged in a chat interaction 104 with a customer 106 of an enterprise, in accordance with an embodiment of the invention. The customer 106 is shown to be accessing an enterprise Website 108 using an electronic device (exemplarily depicted to be a desktop computer). It is noted that the Website 108 is depicted to be devoid of content for illustration purposes and that the Website 108 may display content related to enterprise products or services, promotional offers, new launches from the enterprise, and the like. Further, the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302 between a human agent and a customer of an enterprise; [0057], a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day);
initiating a chat session between the user and a safety agent attending the safety alert management application, wherein the chat session is associated with a identifier, the chat session permitting exchange of text-based messages between the user and the safety agent and the chat session permitting display of the text-based messages in a chat window of a graphical user interface, wherein the graphical user interface is accessed via the safety alert management application (Ramchandran Figs. 1-7; [0021], Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’. The customer 106 may click on the widget or the pop-up to seek agent assistance. Upon receiving an input corresponding to the widget or the pop-up, a Web server hosting the Website may be configured to cause display of a chat console such as the chat console 110 on the display screen of the customer's electronic device. The customer 106 may use the chat console 110 to engage in a textual chat conversation (i.e. the chat interaction 104) with the human agent 102, for receiving desired assistance’; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302 between a human agent and a customer of an enterprise; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day);
for at least one user message received at the safety alert management application, determining a reply message to send to the user electronic device in response to the at least one user message, wherein determining a reply message includes: via an artificial intelligence engine associated with the safety alert management application, using a machine learning model trained on historical chat sessions associated with historical safety alerts and trained on historical data associated with historical safety alerts to analyze the at least one user message and determine one or more recommended reply messages addressing a possible safety issue experienced by the user, at least one of the one or more recommended reply messages determined by selecting at least one relevant pre-determined reply message from a stored library of pre-determined reply messages related to various kinds of safety issues, based at least on the at least one user message (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0040], database 250 is any computer-operated hardware suitable for storing and/or retrieving data, such as, but not limited to, repository of tagged responses (responses tagged with intents by human agents), a list of intents (both programmed and learnt), a registry of human agents and virtual agents; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0051], the first agent may wish to tag the response 310; [0052], the selection input may also cause display of a drop-down menu of intents; the first agent chooses to create a custom intent by providing a selection of the customization option; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions; [0056], the processor 202 is configured to predict possible customer intents for ongoing agent interactions and provide the agents with a respective list of responses; [0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0065], [0067]),
displaying the one or more recommended reply messages on the graphical user interface; and receiving a selection indicating an agent-selected message from the safety agent, the agent-selected message including one of the one or more recommended reply messages (Ramchandran Figs. 1-7; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0076], The second agent may choose an appropriate response from among the trending responses 432-438. In FIG. 4B, the second agent is exemplarily depicted to have selected the response 432 using a touch input; [0065], a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day)
associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together in a database (Ramchandran Figs. 1-7; [0021], FIG. 1 is an example representation 100 of a human agent 102 engaged in a chat interaction 104 with a customer 106 of an enterprise, in accordance with an embodiment of the invention. The customer 106 is shown to be accessing an enterprise Website 108 using an electronic device (exemplarily depicted to be a desktop computer). It is noted that the Website 108 is depicted to be devoid of content for illustration purposes and that the Website 108 may display content related to enterprise products or services, promotional offers, new launches from the enterprise, and the like. Further, the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions. Tracking the count and the time of use of a response in other agent interactions may cause a trending of the respective response; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input; [0082], a track of a count of a number of times an agent response is used in agent interactions with the customers. Such tracked information may facilitate identifying trending responses and trending agents, such as the agents 1, 2 and 3; [0093-0098], At operation 704 of the method 700, at least one trending response relevant to the predicted intent is identified. As explained with reference to FIG. 2, the repeated selection of some tagged responses in agent interactions causes those responses to trend and be shown on the agent consoles for possible inclusion in their interactions)
However, Ramchandran fails to expressly disclose wherein the chat session is associated with a unique identifier, associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database. In the same field of endeavor Ghoche teaches:
wherein the chat session is associated with a unique identifier, associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database (Ghoche Figs. 1-56; [0003], Customer support issues may be assigned a ticket that is served by available human agents over the lifecycle of the ticket; [0069], a ticket dealing with a customer support issue has at least one question, leading to at least one answer being generated in response during the lifecycle of a ticket; [0070], A database 120 stores customer support data. This may include an archive of historical tickets, that includes the Question/Answer pairs as well as other information associated with the lifecycle of a ticket; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0146], A classifier training engine 1410 may be provided to train/retrain the granular taxonomy classifier; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; [0156], Customer support tickets may include emails, chats; [0158], FIG. 16 is a flow chart of an example method of training the classifier according to an implementation. In block 1605, support tickets are ingested)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated wherein the chat session is associated with a unique identifier, associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 2, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the step of selecting at least one relevant pre-determined reply message is based at least on an entire message history of the chat session and the user data (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0040], [0044], [0051-0054], [0065], [0067])
Regarding claim 3, Ramchandran in view Ghoche teaches all the limitations of claim 1, further comprising:
wherein the entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message associated together with the identifier are used to train the machine learning model (Ramchandran Figs. 1-7; [0021], FIG. 1 is an example representation 100 of a human agent 102 engaged in a chat interaction 104 with a customer 106 of an enterprise, in accordance with an embodiment of the invention. The customer 106 is shown to be accessing an enterprise Website 108 using an electronic device (exemplarily depicted to be a desktop computer). It is noted that the Website 108 is depicted to be devoid of content for illustration purposes and that the Website 108 may display content related to enterprise products or services, promotional offers, new launches from the enterprise, and the like. Further, the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions. Tracking the count and the time of use of a response in other agent interactions may cause a trending of the respective response; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input; [0082], a track of a count of a number of times an agent response is used in agent interactions with the customers. Such tracked information may facilitate identifying trending responses and trending agents, such as the agents 1, 2 and 3; [0093-0098], At operation 704 of the method 700, at least one trending response relevant to the predicted intent is identified. As explained with reference to FIG. 2, the repeated selection of some tagged responses in agent interactions causes those responses to trend and be shown on the agent consoles for possible inclusion in their interactions.)
Ghoche further teaches:
the entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message associated together with the unique identifier are used to train the machine learning model (Ghoche Figs. 1-56; [0003], Customer support issues may be assigned a ticket that is served by available human agents over the lifecycle of the ticket; [0069], a ticket dealing with a customer support issue has at least one question, leading to at least one answer being generated in response during the lifecycle of a ticket; [0070], A database 120 stores customer support data. This may include an archive of historical tickets, that includes the Question/Answer pairs as well as other information associated with the lifecycle of a ticket; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0146], A classifier training engine 1410 may be provided to train/retrain the granular taxonomy classifier; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; [0156], Customer support tickets may include emails, chats; [0158], FIG. 16 is a flow chart of an example method of training the classifier according to an implementation. In block 1605, support tickets are ingested)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message associated together with the unique identifier are used to train the machine learning model as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 15, claim 15 contains substantially similar limitations to those found in claim 3, the only difference being the GPT model (Ghoche Figs. 1-56; 0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; see also [0003], [0069], [0070], [0098], [0108], [0128], [0131], [0146], [0148], [0156], [0158]). Consequently, claim 15 is rejected for the same reasons.
Regarding claim 5, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the machine learning model determines whether to select a pre-determined reply message pertaining to on-site assistance (Ramchandran Figs. 1-7; [0021], Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’. The customer 106 may click on the widget or the pop-up to seek agent assistance. Upon receiving an input corresponding to the widget or the pop-up, a Web server hosting the Website may be configured to cause display of a chat console such as the chat console 110 on the display screen of the customer's electronic device. The customer 106 may use the chat console 110 to engage in a textual chat conversation (i.e. the chat interaction 104) with the human agent 102, for receiving desired assistance; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day; an agent response such as ‘SATELLITE SIGNAL RECEPTION HAS BEEN AFFECTED ON ACCOUNT OF INCLEMENT WEATHER. PLEASE REBOOT YOUR TV AFTER 5 PM, WHEN THE WEATHER IS EXPECTED TO BE BETTER’ may be tagged with intent ‘#SIGNAL ERROR’ and such a response may trend on agent consoles; see also [0040], [0044], [0051-0054])
Regarding claim 6, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the safety alert management application receives location information generated by the user electronic device and the machine learning model determines the one or more recommended reply messages based at least on the location information (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input; a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0059], information related to the customer's activity on the enterprise interaction channel, the captured customer data may also include information such as the device used for accessing the Website, the browser and the operating system associated with the device, the type of Internet connection, whether cellular or Wi-Fi, the IP address, the location co-ordinates, and the like; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0021], [0040], [0044-0047], [0051-0054])
Regarding claim 7, Ramchandran in view of Ghoche teaches all the limitations of claim 6, further comprising:
wherein the safety alert management application further receives environmental data obtained by one or more environmental sensors and the machine learning model determines the one or more recommended reply messages based at least on the environmental data (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input; All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day; an agent response such as ‘SATELLITE SIGNAL RECEPTION HAS BEEN AFFECTED ON ACCOUNT OF INCLEMENT WEATHER. PLEASE REBOOT YOUR TV AFTER 5 PM, WHEN THE WEATHER IS EXPECTED TO BE BETTER’ may be tagged with intent ‘#SIGNAL ERROR’ and such a response may trend on agent consoles; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0021], [0040], [0044-0047], [0051-0054])
Regarding claim 8, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the machine learning model is a language processing model selected from (Ramchandran Figs. 1-7; [0021], the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions. Tracking the count and the time of use of a response in other agent interactions may cause a trending of the respective response; [0082], a track of a count of a number of times an agent response is used in agent interactions with the customers. Such tracked information may facilitate identifying trending responses and trending agents, such as the agents 1, 2 and 3; [0093-0098], At operation 704 of the method 700, at least one trending response relevant to the predicted intent is identified. As explained with reference to FIG. 2, the repeated selection of some tagged responses in agent interactions causes those responses to trend and be shown on the agent consoles for possible inclusion in their interactions; see also [0044-0047], [0065], [0067])
Ghoche further teaches:
wherein the machine learning model is a language processing model selected from recurrent neural networks, long short-term memory networks, and transformer models (Ghoche Figs. 1-56; [0088], transformer-based machine learning techniques for natural language processing pre-training; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; see also [0003], [0069-0070], [0146], [0156])
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated wherein the machine learning model is a language processing model selected from recurrent neural networks, long short-term memory networks, and transformer models as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 9, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the machine learning model is (Ramchandran Figs. 1-7; [0021], the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions. Tracking the count and the time of use of a response in other agent interactions may cause a trending of the respective response; [0082], a track of a count of a number of times an agent response is used in agent interactions with the customers. Such tracked information may facilitate identifying trending responses and trending agents, such as the agents 1, 2 and 3; [0093-0098], At operation 704 of the method 700, at least one trending response relevant to the predicted intent is identified. As explained with reference to FIG. 2, the repeated selection of some tagged responses in agent interactions causes those responses to trend and be shown on the agent consoles for possible inclusion in their interactions; see also [0044-0047], [0065], [0067])
Ghoche further teaches:
wherein the machine learning model is generative pretrained transformer (GPT) (Ghoche Figs. 1-56; [0088], transformer-based machine learning techniques for natural language processing pre-training; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; see also [0003], [0069-0070], [0146], [0156])
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated wherein the machine learning model is generative pretrained transformer (GPT) as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 10, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the machine learning model determines the one or more recommended reply messages based on contextual information extracted from one or more of sensor data, camera data, emergency call data, law enforcement data, weather data, geolocation data (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input; a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0059], information related to the customer's activity on the enterprise interaction channel, the captured customer data may also include information such as the device used for accessing the Website, the browser and the operating system associated with the device, the type of Internet connection, whether cellular or Wi-Fi, the IP address, the location co-ordinates, and the like; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day; an agent response such as ‘SATELLITE SIGNAL RECEPTION HAS BEEN AFFECTED ON ACCOUNT OF INCLEMENT WEATHER. PLEASE REBOOT YOUR TV AFTER 5 PM, WHEN THE WEATHER IS EXPECTED TO BE BETTER’ may be tagged with intent ‘#SIGNAL ERROR’ and such a response may trend on agent consoles; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0021], [0040], [0044-0047], [0051-0054])
Regarding claim 11, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
wherein the graphical user interface includes a first selectable tab that displays a pool of the pre-determined reply messages and a second selectable tab that displays the one or more recommended reply messages (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels; [0072], Because the customer intent is not clear during the initial stage of the interaction, the processor 202 may be configured to display a plurality of trending intents on a portion 420 of the agent console 400. The portion 420 is exemplarily depicted to display a header 422 showing a label ‘RELEVANT INTENTS’. Initially, the portion 420 is depicted to display intents 424, 426 and 428 associated with text “#PAYMENT”, ‘#SIGNAL ERROR’ AND ‘#BILL HIGH’, respectively. In some embodiments, these intents may be determined to be relevant to the customer-agent interaction based on a current or past activity of the customer on one or more enterprise interaction channels. For example, if a monthly bill has been recently generated for the second customer, then the interaction 402 may be related to the bill. Similarly, if the second customer has recently tried to make a purchase transaction and was unsuccessful in completing the transaction, then the second customer may have initiated the interaction to query the cause of payment failure; [0073], As the interaction progresses, the intents displayed in the portion 420 may constantly be refined so as to match the relevance of the current conversation. Moreover, each trending intent may be associated with one or more trending responses; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention; ; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; [0076], The second agent may choose an appropriate response from among the trending responses 432-438. In FIG. 4B, the second agent is exemplarily depicted to have selected the response 432 using a touch input. In some embodiments, the second agent may be allowed to drag and drop the appropriate response in the chat interaction display section from the portion 420; [0078], the second agent may proactively select an intent of the interaction from among the relevant intents displayed in the portion 420; see also [0040], [0044-0047], [0051-0054], [0065], [0067])
Regarding claim 12, Ramchandran teaches the claim comprising:
A method for facilitating electronic safety communications by a safety alert management system, the method comprising: electronically receiving a safety alert for a specific safety event from a user electronic device, the safety alert received at a safety alert management application and the safety alert comprising at least one user message from a user associated with the user electronic device and the safety alert comprising additional user data associated with the user (Ramchandran Figs. 1-7; [0021], FIG. 1 is an example representation 100 of a human agent 102 engaged in a chat interaction 104 with a customer 106 of an enterprise, in accordance with an embodiment of the invention. The customer 106 is shown to be accessing an enterprise Website 108 using an electronic device (exemplarily depicted to be a desktop computer). It is noted that the Website 108 is depicted to be devoid of content for illustration purposes and that the Website 108 may display content related to enterprise products or services, promotional offers, new launches from the enterprise, and the like. Further, the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302 between a human agent and a customer of an enterprise; [0057], a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day);
initiating a chat session between the user and a safety agent attending the safety alert management application, wherein the chat session is associated with a identifier, the chat session permitting exchange of text-based messages between the user and the safety agent and the chat session permitting display of the text-based messages in a chat window of a graphical user interface, wherein the graphical user interface is accessed via the safety alert management application (Ramchandran Figs. 1-7; [0021], Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’. The customer 106 may click on the widget or the pop-up to seek agent assistance. Upon receiving an input corresponding to the widget or the pop-up, a Web server hosting the Website may be configured to cause display of a chat console such as the chat console 110 on the display screen of the customer's electronic device. The customer 106 may use the chat console 110 to engage in a textual chat conversation (i.e. the chat interaction 104) with the human agent 102, for receiving desired assistance’; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302 between a human agent and a customer of an enterprise);
for at least one user message received at the safety alert management application, determining a reply message to send to the user electronic device in response to the at least one user message, wherein determining a reply message includes: via an artificial intelligence engine associated with the safety alert management application, using a AI model trained on historical chat sessions associated with historical safety alerts and historical data associated with historical safety alerts to analyze the at least one user message and determine one or more recommended reply messages addressing a possible safety issue experienced by the user, wherein at least one of the recommended reply messages is a generated message generated by the AI model based at least on the at least one user message (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0040], database 250 is any computer-operated hardware suitable for storing and/or retrieving data, such as, but not limited to, repository of tagged responses (responses tagged with intents by human agents), a list of intents (both programmed and learnt), a registry of human agents and virtual agents; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0051], the first agent may wish to tag the response 310; [0052], the selection input may also cause display of a drop-down menu of intents; the first agent chooses to create a custom intent by providing a selection of the customization option; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions; [0056], the processor 202 is configured to predict possible customer intents for ongoing agent interactions and provide the agents with a respective list of responses; [0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0065], [0067]),
displaying the one or more recommended reply messages on the graphical user interface; and receiving a selection indicating an agent-selected message from the safety agent, the agent-selected message including one of the one or more recommended reply messages (Ramchandran Figs. 1-7; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0076], The second agent may choose an appropriate response from among the trending responses 432-438. In FIG. 4B, the second agent is exemplarily depicted to have selected the response 432 using a touch input; [0065], a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day)
associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together in a database (Ramchandran Figs. 1-7; [0021], FIG. 1 is an example representation 100 of a human agent 102 engaged in a chat interaction 104 with a customer 106 of an enterprise, in accordance with an embodiment of the invention. The customer 106 is shown to be accessing an enterprise Website 108 using an electronic device (exemplarily depicted to be a desktop computer). It is noted that the Website 108 is depicted to be devoid of content for illustration purposes and that the Website 108 may display content related to enterprise products or services, promotional offers, new launches from the enterprise, and the like. Further, the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions. Tracking the count and the time of use of a response in other agent interactions may cause a trending of the respective response; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input; [0082], a track of a count of a number of times an agent response is used in agent interactions with the customers. Such tracked information may facilitate identifying trending responses and trending agents, such as the agents 1, 2 and 3; [0093-0098], At operation 704 of the method 700, at least one trending response relevant to the predicted intent is identified. As explained with reference to FIG. 2, the repeated selection of some tagged responses in agent interactions causes those responses to trend and be shown on the agent consoles for possible inclusion in their interactions)
However, Ramchandran fails to expressly disclose wherein the chat session is associated with a unique identifier; a generative pretrained transformer (GPT) model trained on historical chat sessions; a generated message generated by the GPT model; associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database. In the same field of endeavor, Ghoche teaches:
wherein the chat session is associated with a unique identifier; a generative pretrained transformer (GPT) model trained on historical chat sessions; a generated message generated by the GPT model; associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database (Ghoche 1-56; [0003], Customer support issues may be assigned a ticket that is served by available human agents over the lifecycle of the ticket; [0069], a ticket dealing with a customer support issue has at least one question, leading to at least one answer being generated in response during the lifecycle of a ticket; [0070], A database 120 stores customer support data. This may include an archive of historical tickets, that includes the Question/Answer pairs as well as other information associated with the lifecycle of a ticket; [0080], This includes a history of tickets and chats and whatever else a company may potentially have regarding CRMs/helpdesks like Zendesk® or Salesforce®; [0081], The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0146], A classifier training engine 1410 may be provided to train/retrain the granular taxonomy classifier; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; [0156], Customer support tickets may include emails, chats; [0158], FIG. 16 is a flow chart of an example method of training the classifier according to an implementation. In block 1605, support tickets are ingested; [0182], a generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers. A generative AI empathy model training/fine tuning module 2750 may be provided to train/fine tune a an empathy model; [0230], The large language model may be implemented using generative AI models such as ChatGPT or GPT4. Additionally, the large language model may be a large language model customized for customer support; the large language model may be trained on a set of customer support examples)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated wherein the chat session is associated with a unique identifier; a generative pretrained transformer (GPT) model trained on historical chat sessions; a generated message generated by the GPT model; associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 13, Ramchandran in view of Ghoche teaches all the limitations of claim 12, further comprising:
wherein at least one of the one or more recommended reply messages determined by the AI model is a selected pre-determined reply message determined by selecting at least one relevant pre-determined reply message from a stored library of pre-determined reply messages, based at least on the at least one user message (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0040], [0044], [0051-0054], [0065], [0067])
Ghoche further teaches:
reply messages determined by the GPT model (Ghoche 1-56; [0080], This includes a history of tickets and chats and whatever else a company may potentially have regarding CRMs/helpdesks like Zendesk® or Salesforce®; [0081], The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0182], a generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers. A generative AI empathy model training/fine tuning module 2750 may be provided to train/fine tune a an empathy model; [0230], The large language model may be implemented using generative AI models such as ChatGPT or GPT4. Additionally, the large language model may be a large language model customized for customer support; the large language model may be trained on a set of customer support examples)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated reply messages determined by the GPT model model as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 14, Ramchandran in view of Ghoche teaches all the limitations of claim 13, further comprising:
wherein displaying the one or more recommended reply messages on the graphical user interface includes indicating via at least one of color, text, or placement whether a displayed recommended reply message is a selected pre-determined reply message or a generated message (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0040], [0044], [0051-0054], [0065], [0067])
Regarding claim 16, Ramchandran in view of Ghoche teaches all the limitations of claim 12, further comprising:
wherein the model determines the one or more recommended reply messages based at least on an entire message history of the chat session (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0056-0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0040], [0044], [0051-0054], [0065], [0067])
Ghoche further teaches:
wherein the GPT model determines the one or more recommended reply messages based at least on an entire message history of the chat session (Ghoche 1-56; [0080], This includes a history of tickets and chats and whatever else a company may potentially have regarding CRMs/helpdesks like Zendesk® or Salesforce®; [0081], The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0182], a generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers. A generative AI empathy model training/fine tuning module 2750 may be provided to train/fine tune a an empathy model; [0230], The large language model may be implemented using generative AI models such as ChatGPT or GPT4. Additionally, the large language model may be a large language model customized for customer support; the large language model may be trained on a set of customer support examples)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated wherein the GPT model determines the one or more recommended reply messages based at least on an entire message history of the chat session as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 17, Ramchandran teaches the claim comprising:
A method for training a safety chat language model in a safety alert management system, the method comprising: inputting at least one first set of training data including a plurality of text messages related to safety events into a safety chat language model and generating a trained safety chat language model via training the safety chat language model on the at least one first set of training data to determine relevant reply messages addressing possible safety issues described in the plurality of text messages in response to the plurality of text messages (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0040], database 250 is any computer-operated hardware suitable for storing and/or retrieving data, such as, but not limited to, repository of tagged responses (responses tagged with intents by human agents), a list of intents (both programmed and learnt), a registry of human agents and virtual agents; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0051], the first agent may wish to tag the response 310; [0052], the selection input may also cause display of a drop-down menu of intents; the first agent chooses to create a custom intent by providing a selection of the customization option; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions; [0056], the processor 202 is configured to predict possible customer intents for ongoing agent interactions and provide the agents with a respective list of responses; [0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day);
electronically receiving a safety alert for a specific safety event from a user electronic device, the safety alert received at a safety alert management application and the safety alert comprising at least one user message from a user associated with the user electronic device and the safety alert comprising additional user data associated with the user (Ramchandran Figs. 1-7; [0021], FIG. 1 is an example representation 100 of a human agent 102 engaged in a chat interaction 104 with a customer 106 of an enterprise, in accordance with an embodiment of the invention. The customer 106 is shown to be accessing an enterprise Website 108 using an electronic device (exemplarily depicted to be a desktop computer). It is noted that the Website 108 is depicted to be devoid of content for illustration purposes and that the Website 108 may display content related to enterprise products or services, promotional offers, new launches from the enterprise, and the like. Further, the Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302 between a human agent and a customer of an enterprise; [0057], a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day);
initiating an electronic chat session for the safety alert between the user and a safety agent attending the safety alert management application (Ramchandran Figs. 1-7; [0021], Website 108 may display a widget or a pop-up, which is associated with text such as ‘Let's Chat’ or ‘Need Assistance, Click Here!’. The customer 106 may click on the widget or the pop-up to seek agent assistance. Upon receiving an input corresponding to the widget or the pop-up, a Web server hosting the Website may be configured to cause display of a chat console such as the chat console 110 on the display screen of the customer's electronic device. The customer 106 may use the chat console 110 to engage in a textual chat conversation (i.e. the chat interaction 104) with the human agent 102, for receiving desired assistance’; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302 between a human agent and a customer of an enterprise; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries);
analyzing the at least one user message via the trained safety chat language model to determine one or more recommended reply messages, wherein the one or more recommended reply messages are determined based at least on an entire message history of the electronic chat session and the additional user data (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0040], database 250 is any computer-operated hardware suitable for storing and/or retrieving data, such as, but not limited to, repository of tagged responses (responses tagged with intents by human agents), a list of intents (both programmed and learnt), a registry of human agents and virtual agents; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0051], the first agent may wish to tag the response 310; [0052], the selection input may also cause display of a drop-down menu of intents; the first agent chooses to create a custom intent by providing a selection of the customization option; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions; [0056], the processor 202 is configured to predict possible customer intents for ongoing agent interactions and provide the agents with a respective list of responses; [0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; see also [0065], [0067]);
displaying the one or more recommended reply messages to the safety agent in a graphical user interface of the safety alert management application; receiving a selection indicating an agent-selected message selected by the safety agent, the agent-selected message including one of the one or more recommended reply messages; transmitting the agent-selected message to the user electronic device via the chat session (Ramchandran Figs. 1-7; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0076], The second agent may choose an appropriate response from among the trending responses 432-438. In FIG. 4B, the second agent is exemplarily depicted to have selected the response 432 using a touch input; The selected response may be displayed as an answer to the customer's reply 412 in the chat interaction display section; [0065], a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day; [0077], The second agent may then need to only click the button 470 labeled ‘SEND’ to send the response to second customer);
associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data (Ramchandran Figs. 1-7; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions. Tracking the count and the time of use of a response in other agent interactions may cause a trending of the respective response; [0082], a track of a count of a number of times an agent response is used in agent interactions with the customers. Such tracked information may facilitate identifying trending responses and trending agents, such as the agents 1, 2 and 3; [0093-0098], At operation 704 of the method 700, at least one trending response relevant to the predicted intent is identified. As explained with reference to FIG. 2, the repeated selection of some tagged responses in agent interactions causes those responses to trend and be shown on the agent consoles for possible inclusion in their interactions)
However, Ramchandran fails to expressly disclose associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert ID to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data. In the same field of endeavor Ghoche teaches:
associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert ID to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data (Ghoche Figs. 1-56; [0003], Customer support issues may be assigned a ticket that is served by available human agents over the lifecycle of the ticket; [0069], a ticket dealing with a customer support issue has at least one question, leading to at least one answer being generated in response during the lifecycle of a ticket; [0070], A database 120 stores customer support data. This may include an archive of historical tickets, that includes the Question/Answer pairs as well as other information associated with the lifecycle of a ticket; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; [0146], A classifier training engine 1410 may be provided to train/retrain the granular taxonomy classifier; [0156], Customer support tickets may include emails, chats; [0158], FIG. 16 is a flow chart of an example method of training the classifier according to an implementation. In block 1605, support tickets are ingested)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert ID to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Regarding claim 18, Ramchandran in view of Ghoche teaches all the limitations of claim 17, further comprising:
wherein the at least one first set of training data comprises historical chat sessions and historical data associated with historical safety alerts (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0040], database 250 is any computer-operated hardware suitable for storing and/or retrieving data, such as, but not limited to, repository of tagged responses (responses tagged with intents by human agents), a list of intents (both programmed and learnt), a registry of human agents and virtual agents; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0051], the first agent may wish to tag the response 310; [0052], the selection input may also cause display of a drop-down menu of intents; the first agent chooses to create a custom intent by providing a selection of the customization option; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions; [0056], the processor 202 is configured to predict possible customer intents for ongoing agent interactions and provide the agents with a respective list of responses; [0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day)
Regarding claim 19, Ramchandran in view of Ghoche teaches all the limitations of claim 17, further comprising:
wherein the safety chat language model determines the one or more recommended reply messages based at least in part on contextual information extracted from one or more of sensor data, camera data, emergency call data, law enforcement data, weather data, or geolocation data, and the contextual information is associated with the alert identifier and included in the associated second set of training data (Ramchandran Figs. 1-7; [0032], memory 204 is configured to store a list of predefined intents (both programmed and learnt). Further, the memory 204 stores Natural Language Processing (NLP) algorithms and other machine learning algorithms for interpreting customer inputs and predicting customer intents based at least in part on the customer inputs; [0040], database 250 is any computer-operated hardware suitable for storing and/or retrieving data, such as, but not limited to, repository of tagged responses (responses tagged with intents by human agents), a list of intents (both programmed and learnt), a registry of human agents and virtual agents; [0044], the agent may tag a response if the agent feels that the customer has responded favorably to a response or has liked the response. In another illustrative example, the agent may tag a response if the response resulted in a preferred outcome; [0045], FIG. 3A shows a simplified representation of an agent console 300 displaying an ongoing chat interaction 302; [0047], The inputs provided by the first customer during the interaction 302 are depicted to be associated with label ‘JOHN’; [0051], the first agent may wish to tag the response 310; [0052], the selection input may also cause display of a drop-down menu of intents; the first agent chooses to create a custom intent by providing a selection of the customization option; [0053], the response 310 including text: ‘PLEASE LET ME KNOW WHAT ERROR IS BEING DISPLAYED ON THE TV WHILE THE TV KEEPS FREEZING.’ may be tagged with the intent ‘#SIGNAL ERROR’ and stored in the database 250 as exemplarily depicted in FIG. 3C; [0054], a tag ID, a response and a tagged intent, respectively, for illustration purposes. It is noted that the tabular representation 380 may also be configured to store information (not shown in the tabular representation 380) such as a name of the agent, i.e. the name of the first agent, who has tagged the response with the corresponding intent, the time stamp of the tagging of the response, a count of a number of times the response is used in other agent interactions and a time of use of the response in other agent interactions; [0056], the processor 202 is configured to predict possible customer intents for ongoing agent interactions and provide the agents with a respective list of responses; [0057], the processor 202 is configured to use the NLP algorithms and other machine learning algorithms stored in the memory 204 to interpret each customer input and predict one or more intents of the customer corresponding to each customer input. In some embodiments, the customer's intent is predicted solely based on the customer's input. For example, the customer may provide the following input ‘THE DELIVERY OF MY SHIPMENT HAS BEEN DELAYED BY TWO DAYS NOW. THIS IS UNACCEPTABLE!!’ to an agent. Based on such an input, the processor 202 may be configured to predict the intent as ‘#DELIVERY DELAY’. In some embodiments, the customer intent may be predicted based on past interactions of the customer on enterprise interaction channels. For example, if a customer has recently purchased an airline ticket, then the intent for requesting a chat interaction may most likely be related to confirmation of the flight time, rescheduling the journey or cancellation of the ticket. In some embodiments, the customer intent may be predicted based on current interaction of the customer on an enterprise interaction channel. For example, a customer having visited the enterprise Website may browse through a number of Web pages and may have viewed a number of products on the Website prior to requesting a chat interaction with an agent. All such activity of the customer during the current journey of the customer on the enterprise Website may be captured and used for intent prediction purposes; [0059], information related to the customer's activity on the enterprise interaction channel, the captured customer data may also include information such as the device used for accessing the Website, the browser and the operating system associated with the device, the type of Internet connection, whether cellular or Wi-Fi, the IP address, the location co-ordinates, and the like; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0075], The processor 202 monitoring the interaction 402 may receive the customer input, i.e. reply 412, and determine the intent as ‘#BILL HIGH’. Further, the processor 202 may be configured to fetch the top trending responses tagged to that intent; [0065], Some examples of recent event may be a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc. In case of occurrence of such events, the agent responses may have wider applicability as the fellow agents may also face similar queries; [0067], a particularly severe rainy day; an agent response such as ‘SATELLITE SIGNAL RECEPTION HAS BEEN AFFECTED ON ACCOUNT OF INCLEMENT WEATHER. PLEASE REBOOT YOUR TV AFTER 5 PM, WHEN THE WEATHER IS EXPECTED TO BE BETTER’ may be tagged with intent ‘#SIGNAL ERROR’ and such a response may trend on agent consoles)
Ghoche further teaches:
via the unique alert identifier (Ghoche Figs. 1-56; [0003], Customer support issues may be assigned a ticket that is served by available human agents over the lifecycle of the ticket; [0069], a ticket dealing with a customer support issue has at least one question, leading to at least one answer being generated in response during the lifecycle of a ticket; [0070], A database 120 stores customer support data. This may include an archive of historical tickets, that includes the Question/Answer pairs as well as other information associated with the lifecycle of a ticket; [0079], FIG. 3 illustrates an example of a portion of a system 306 in which an incoming ticket 302 is received that has a customer question. The incoming question can be analyzed for question document features 310, document pair features 312, answer document features 314, and can be used to identify answers with scores 320 according to a ranking model 316. For example, an incoming ticket 302 can be analyzed to determine if a solution to a customer question can be automatically responded to using a pre-approved answer within a desired threshold level of accuracy; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; [0146], A classifier training engine 1410 may be provided to train/retrain the granular taxonomy classifier; [0156], Customer support tickets may include emails, chats; [0158], FIG. 16 is a flow chart of an example method of training the classifier according to an implementation. In block 1605, support tickets are ingested)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated via the unique alert identifier as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Ramchandran in view of Ghoche in further view of Da Jornada et al. (US 20230370393 A1, published 11/16/2023), hereinafter Da Jornada.
Regarding claim 4, Ramchandran in view of Ghoche teaches all the limitations of claim 1, further comprising:
further comprising displaying the agent-selected message in a text input portion of the chat window and permitting the safety agent to select the agent- selected message before sending the agent-selected message to the user electronic device (Ramchandran Figs. 1-7; [0074], FIG. 4B shows a simplified representation of the agent console 400 of FIG. 4A displaying a plurality of trending responses tagged to a relevant intent, in accordance with an embodiment of the invention. The agent console 400 shows the interaction 402 between an agent, for example the second agent, and a customer, for example the second customer, of the enterprise. The agent console 400 includes new messages exchanged by the second agent and the customer, i.e. Lara. For example, after receiving the query 404 from the second agent, the second customer, i.e. Lara, is depicted to have answered with a reply 412 displaying text ‘MY MONTHLY PHONE BILL SEEMS TO BE UNUSUALLY HIGH, CAN YOU HELP ME WITH THE DETAILS?; [0076], The second agent may choose an appropriate response from among the trending responses 432-438. In FIG. 4B, the second agent is exemplarily depicted to have selected the response 432 using a touch input)
However, Ramchandran in view of Ghoche fails to expressly disclose further comprising displaying the agent-selected message in a text input portion of the chat window and permitting the safety agent to edit the agent-selected message before sending the agent-selected message to the user electronic device. In the same field of endeavor Da Jornada teaches:
further comprising displaying the agent-selected message in a text input portion of the chat window and permitting the safety agent to edit the agent-selected message before sending the agent-selected message to the user electronic device (Da Jornada Figs. 1-7; [0061], as noted above, chat messages within an existing message platform can be intercepted and provided to the agent response prediction server 110 of FIG. 1 to predict one or more customer service agent responses. The interface provided by the existing message platform to a customer service agent may be modified to present the predicted agent responses (e.g., a configurable and/or learned number of alternative response options) to the customer service agent, for example, ranked by a confidence score, with any placeholder fields automatically populated. The alternative response options may be displayed to a customer service agent, for example, in the form of one button for each alternative response option, sorted by relevance; [0062], a customer service agent can select any of the buttons to insert the content associated with the alternative response option in the chat message composition portion of the user interface. The customer service agent can then send, modify, delete and/or typeover the inserted content, as required or desired. In addition, the customer service agent may include additional text or completely ignore the prediction and manually type a full response (e.g., by not selecting one of the populated predicted response buttons))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated further comprising displaying the agent-selected message in a text input portion of the chat window and permitting the safety agent to edit the agent-selected message before sending the agent-selected message to the user electronic device as suggested in Da Jornada into Ramchandran in view of Ghoche. Doing so would be desirable because despite the increased use of automated chat agents by numerous customer service organizations, human agents are still needed for a number of interactions, such as for interactions involving more complex topics and for highly valued customers. A number of customer service organizations provide human agents with access to messages templates that are often based on frequently used messages. Nonetheless, it is often difficult for a human agent to compose a message when responding to a customer (see Da Jornada [0002]). One or more embodiments of the disclosure provide improved methods, apparatus and computer program products for machine learning-based prediction of message responses. The disclosed machine learning-based response prediction techniques assist a customer service agent to understand what the user is looking for and makes it more efficient (e.g., easier and/or faster) for the customer service agent to respond to the user (see Da Jornada [0064]). Additionally, the system of Da Jornada would improve the system of Ramchandran by enabling the user to provide enhanced responses that are customized the customers’ needs, thereby increasing the usefulness and desirability of the system.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Ramchandran in view of Ghoche in further view of Thiyagarajan et al. (US 20210377391 A1, published 12/02/2021), hereinafter Thiyagarajan.
Regarding claim 20, Ramchandran in view of Ghoche teaches all the limitations of claim 17. Ghoche further teaches:
via the unique alert ID (Ghoche Figs. 1-56; [0003], Customer support issues may be assigned a ticket that is served by available human agents over the lifecycle of the ticket; [0069], a ticket dealing with a customer support issue has at least one question, leading to at least one answer being generated in response during the lifecycle of a ticket; [0070], A database 120 stores customer support data. This may include an archive of historical tickets, that includes the Question/Answer pairs as well as other information associated with the lifecycle of a ticket; [0079], FIG. 3 illustrates an example of a portion of a system 306 in which an incoming ticket 302 is received that has a customer question. The incoming question can be analyzed for question document features 310, document pair features 312, answer document features 314, and can be used to identify answers with scores 320 according to a ranking model 316. For example, an incoming ticket 302 can be analyzed to determine if a solution to a customer question can be automatically responded to using a pre-approved answer within a desired threshold level of accuracy; [0098], A ticket covers the entire lifecycle of an issue; [0108], an answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer and/or revise the recommended answer. In some implementations, a one-click answer functionality is supported for an agent to select a recommended answer; [0114], an agent is provided with auto-suggestions for at least partially completing a response answer. For example, typing their response, the system suggests a selection of words having a threshold confidence level for a specific completion. This may correspond to a text for the next X words, where X might be between 10, 15, or 20 words, as an example, with the total number of words being selected may be limited to maintain a high confidence level. The auto-suggestion may be based on the history of tickets as an agent is typing their answer. It may also be customized for individual agents. The ML model may, for example, be based on a GPT2 model; [0115], historical tickets are tokenized to put in markers at the beginning of the question, the beginning of the subject of the description, the beginning of the answer, or at any other location where a token may help to identify portions of questions and corresponding portions of an answer. At the end of the whole ticket, additional special markers are placed. The marked tickets are fed into GPT2. The model is trained to generate word prompts based on the entire question as well as anything that the agent has typed so far in their answer; [0128], where the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers; [0131], a ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response; [0148], previous answers given by agents for a topic in the granular taxonomy may be used to generate a recommended answer when an incoming customer support ticket is handled by an agent; [0146], A classifier training engine 1410 may be provided to train/retrain the granular taxonomy classifier; [0156], Customer support tickets may include emails, chats; [0158], FIG. 16 is a flow chart of an example method of training the classifier according to an implementation. In block 1605, support tickets are ingested)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated via the unique alert ID as suggested in Ghoche into Ramchandran. Doing so would be desirable because customer support service is an important aspect of many businesses. For example, there are a variety of customer support applications to address customer service support issues (see Ghoche [0002]). While an institution may have a lot of institutional knowledge to aid human agents, there may be practical difficulties in training agents to use all the institutional knowledge that is potentially available to aid in responding to tickets. For example, conventionally, a human agent may end up doing a manual search of the institutional knowledge (see Ghoche [0004]). There are substantial training and labor costs to have a large pool of highly trained human agents available to service customer issues. There are also labor costs associated with having human experts making decisions about how to label and route tickets. But in addition to labor costs, there are other issues in terms of the frustration customers experience if there is a long delay in responding to their queries (see Ghoche [0006]). The history of tickets is a valuable resource for training an AI engine to mimic the way human agents respond to common questions. Historical tickets track the lifecycle of responding to a support question. As a result they include a history of the initial question, answers by agents, and chat information associated with the ticket (see Ghoche [0081]). Large language models (LLMs) can be used to aid in providing customer support. Generative AI models, such as ChatGPT may be used (see Ghoche [0175]). A generative AI model such as T5, GPT-Neo, or OPT may be trained and finetunes on the task of generating a template answer given a prompt of many similar (real) answers (see Ghoche [0182]).
However, Ramchandran in view of Ghoche fails to expressly disclose further comprising measuring one or more time periods associated with the safety alert wherein the one or more time periods associated with the safety alert are selected from a total length of time of the chat session or a length of time between receipt of the at least one user message and transmission of the agent-selected message in reply to the at least one user message, associating the one or more time periods with the second set of training data, and training the safety language model to determine subsequent recommended reply messages based on time efficiency. In the same field of endeavor Thiyagarajan teaches:
further comprising measuring one or more time periods associated with the safety alert wherein the one or more time periods associated with the safety alert are selected from a total length of time of the chat session or a length of time between receipt of the at least one user message and transmission of the agent-selected message in reply to the at least one user message, associating the one or more time periods with the second set of training data, and training the safety language model to determine subsequent recommended reply messages based on time efficiency (Thiyagarajan Figs. 1-7; [0019], The agent-guided chatbot assistant may implement machine learning to help improve chat session accuracy in responding to customer queries, significantly shorten average customer handling times, and allow an agent to drive multiple customer chat sessions concurrently; [0043], AI/ML engine 440 may also include a bot training improvement component that may modify intent-flow matching data based on input received from an agent relative to matching determinations; [0050], If no input has been received from the service agent via the GUI, the CSC chat system may determine whether an amount of time configured for an auto-send feature has expired (block 570). If the time has not expired (block 570—NO), process 500 may return to block 560. When the CSC system determines that the time has expired (block 570—YES), the CSC system may send the suggested chat message response to the customer via chatbot access channel 410, without any action performed by the service agent (i.e., passive monitoring) (block 580); [0055], The agent assist tool may include interactive GUI objects 740 that allow a service agent to send the suggested response message, skip the suggested response message, edit the suggested response message, etc. For example, the service agent may alter the text of the suggested response message and/or one or more of GUI objects 735, may add other GUI objects, etc. In one implementation, predictive feed app field 710 may also include an “auto send” timer object 745 that indicates a predetermined amount of time after which the suggested response message will be sent to the customer (and post to the chat session) without the service agent acting upon the suggested response message. For example, the service agent may be given a particular number of seconds to review, approve, and/or alter the suggested response message before the message is automatically sent to the customer; [0056], the amount of time preceding the auto send operation is configurable. In other implementations, the service agent may incrementally extend and/or temporarily pause the predetermined amount of time before the time has elapsed. In one or more implementations, the predetermined amount of time may be a value that is based on a relative confidence level that is associated with the suggested response. For example, if the suggested response is identical to a response that the service agent has previously composed and/or approved for the identified customer intent and/or chat flow subject matter, the auto send time may be proportionally shortened. In scenarios in which the service agent at agent device 230 approves of the suggested response message, the auto send time may be bypassed at any time by the service agent's input via agent device 230, such as activating the “send” button)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated further comprising measuring one or more time periods associated with the safety alert wherein the one or more time periods associated with the safety alert are selected from a total length of time of the chat session or a length of time between receipt of the at least one user message and transmission of the agent-selected message in reply to the at least one user message, associating the one or more time periods with the second set of training data, and training the safety language model to determine subsequent recommended reply messages based on time efficiency as suggested in Thiyagarajan into Ramchandran in view of Ghoche. Doing so would be desirable because Currently, the artificial intelligence (AI) resources available to the agents present various technological challenges that affect the customer experience, for instance, relative to minimizing the average customer handling times. For example, in the absence of an integrated system of assistive AI tools, agents must consult multiple applications and navigate their user interfaces (e.g., windows, browsers, etc.) during a typical transaction, which limits potential economies of concurrency (i.e., increased agent to customer ratio) (see Thiyagarajan [0002]). Chatbot systems may present an end user with a convenient conversational interface for prompt resolution of a submitted inquiry (e.g., technical support), or efficient provision of a requested service (e.g., a service/device upgrade), irrespective of live agent staffing availability. However, optimizing a chatbot to simulate human conversation during a customer chat session—start to finish—is technically complex (see Thiyagarajan [0011]). Systems and methods disclosed herein can also provide capability of improved chatbot training, through agent monitoring of and dynamic intervention upon chatbot fallout, and repurposing of customized agent-customer interchanges, to continually improve chatbot performance—including speed and efficiency of the chatbot in task accomplishment, and customer experience (see Thiyagarajan [0018]). The agent-guided chatbot assistant may implement machine learning to help improve chat session accuracy in responding to customer queries, significantly shorten average customer handling times, and allow an agent to drive multiple customer chat sessions concurrently (see Thiyagarajan [0019]).
Response to Arguments
The Examiner acknowledges the Applicant’s amendments to claims 1, 3-7, 12- 15, 17, 18, and 20. The objection to claim 5 is respectfully withdrawn. The rejections of claims 1-20 under 35 U.S.C. 112(b) are respectfully withdrawn.
Regarding independent claim 1, the Applicant alleges that Ramchandran as described in the previous Office action, does not explicitly teach amended claim 1 (see remarks pp. 14-15). Examiner has therefore rejected independent claim 1 under 35 U.S.C § 103 as unpatentable over Ramchandran in view of Ghoche.
As discussed in the rejection above, Ramchandran is considered to teach a method for facilitating electronic safety alert communications by a safety alert management system, the method comprising: electronically receiving a safety alert for a specific safety event from a user electronic device, the safety alert received at a safety alert management application and the safety alert comprising at least one user message from a user associated with the user electronic device and the safety alert comprising additional user data associated with the user (Ramchandran Figs. 1-7; [0021], [0045], [0057], [0065], [0067]); initiating a chat session between the user and a safety agent attending the safety alert management application, wherein the chat session is associated with a unique identifier, the chat session permitting exchange of text-based messages between the user and the safety agent and the chat session permitting display of the text-based messages in a chat window of a graphical user interface, wherein the graphical user interface is accessed via the safety alert management application (Ramchandran Figs. 1-7; [0021], [0045], [0065], [0067]); for at least one user message received at the safety alert management application, determining a reply message to send to the user electronic device in response to the at least one user message, wherein determining a reply message includes: via an artificial intelligence engine associated with the safety alert management application, using a machine learning model trained on historical chat sessions associated with historical safety alerts and trained on historical data associated with historical safety alerts to analyze the at least one user message and determine one or more recommended reply messages addressing a possible safety issue experienced by the user, at least one of the one or more recommended reply messages determined by selecting at least one relevant pre-determined reply message from a stored library of pre-determined reply messages related to various kinds of safety issues, based at least on the at least one user message (Ramchandran Figs. 1-7; [0032], [0040], [0044-0045], [0047], [0052-0054], [0057], [0074-0075], [0065], [0067]), displaying the one or more recommended reply messages on the graphical user interface; and receiving a selection indicating an agent-selected message from the safety agent, the agent-selected message including one of the one or more recommended reply messages (Ramchandran Figs. 1-7; [0074], [0076], [0065], [0067]); associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together in a database (Ramchandran Figs. 1-7; [0021], [0044-0045], [0047], [0053], [0054], [0056-0057], [0082], [0093-0098]). Ghoche is cited to clarify wherein the chat session is associated with a unique identifier, associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database (Ghoche Figs. 1-56; [0003], [0069-0070], [0098], [0108], [0114-0115], [0128], [0131], [0146], [0148], [0156], [0158]). Thus, Ramchandran in view of Ghoche is considered to teach the claim.
Regarding independent claim 12, the Applicant alleges that Ramchandran in view of Ghoche as described in the previous Office action, does not explicitly teach amended claim 12 (see remarks pp. 16-17). Examiner respectfully disagrees.
As discussed in the rejection above, Ramchandran is considered to teach a method for facilitating electronic safety communications by a safety alert management system, the method comprising: electronically receiving a safety alert for a specific safety event from a user electronic device, the safety alert received at a safety alert management application and the safety alert comprising at least one user message from a user associated with the user electronic device and the safety alert comprising additional user data associated with the user (Ramchandran Figs. 1-7; [0021], [0045], [0057], [0065], [0067]); initiating a chat session between the user and a safety agent attending the safety alert management application, wherein the chat session is associated with a identifier, the chat session permitting exchange of text-based messages between the user and the safety agent and the chat session permitting display of the text-based messages in a chat window of a graphical user interface, wherein the graphical user interface is accessed via the safety alert management application (Ramchandran Figs. 1-7; [0021], [0065], [0045]); for at least one user message received at the safety alert management application, determining a reply message to send to the user electronic device in response to the at least one user message, wherein determining a reply message includes: via an artificial intelligence engine associated with the safety alert management application, using a AI model trained on historical chat sessions associated with historical safety alerts and historical data associated with historical safety alerts to analyze the at least one user message and determine one or more recommended reply messages addressing a possible safety issue experienced by the user, wherein at least one of the recommended reply messages is a generated message generated by the AI model based at least on the at least one user message (Ramchandran Figs. 1-7; [0032], [0040], [0044-0045], [0047], [0052-0054], [0056-0057], [0074-0075], [0065], [0067]), displaying the one or more recommended reply messages on the graphical user interface; and receiving a selection indicating an agent-selected message from the safety agent, the agent-selected message including one of the one or more recommended reply messages (Ramchandran Figs. 1-7; [0074], [0076], [0065], [0067]); associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together in a database (Ramchandran Figs. 1-7; [0021], [0044-0045], [0047], [0053-0054], [0056-0057], [0082], [0093-0098]). Ghoche is cited to clarify wherein the chat session is associated with a unique identifier; a generative pretrained transformer (GPT) model trained on historical chat sessions; a generated message generated by the GPT model; associating an entire message history of the chat session, the one or more recommended reply messages, the user data, and the agent-selected message together with the unique identifier in a database (Ghoche 1-56; [0003], [0069-0070], [0080-0081], [0098], [0108], [0114-0115], [0128], [0131], [0146], [0148], [0156], [0158], [0182], [0230]). Thus, Ramchandran in view of Ghoche is considered to teach the claim.
Regarding independent claim 17, the Applicant alleges that Ramchandran in view of Ghoche as described in the previous Office action, does not explicitly teach amended claim 17 (see remarks p. 18). Examiner respectfully disagrees.
As discussed in the rejection above, Ramchandran is considered to teach A method for training a safety chat language model in a safety alert management system, the method comprising: inputting at least one first set of training data including a plurality of text messages related to safety events into a safety chat language model and generating a trained safety chat language model via training the safety chat language model on the at least one first set of training data to determine relevant reply messages addressing possible safety issues described in the plurality of text messages in response to the plurality of text messages (Ramchandran Figs. 1-7; [0032], [0040], [0044-0045], [0047], [0052-0054], [0056-0057], [0074-0075], [0065], [0067]); electronically receiving a safety alert for a specific safety event from a user electronic device, the safety alert received at a safety alert management application and the safety alert comprising at least one user message from a user associated with the user electronic device and the safety alert comprising additional user data associated with the user (Ramchandran Figs. 1-7; [0021], [0045], [0057], [0065], [0067]); initiating an electronic chat session for the safety alert between the user and a safety agent attending the safety alert management application (Ramchandran Figs. 1-7; [0021], [0045], [0065]); analyzing the at least one user message via the trained safety chat language model to determine one or more recommended reply messages, wherein the one or more recommended reply messages are determined based at least on an entire message history of the electronic chat session and additional user data (Ramchandran Figs. 1-7; [0032], [0040], [0044-0045], [0047], [0051-0054], [0057], [0074-0075], [0065], [0067]); displaying the one or more recommended reply messages to the safety agent in a graphical user interface of the safety alert management application; receiving a selection indicating an agent-selected message selected by the safety agent, the agent-selected message including one of the one or more recommended reply messages; transmitting the agent-selected message to the user electronic device via the chat session (Ramchandran Figs. 1-7; [0074], [0076], [0065], [0067], [0077],); associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data (Ramchandran Figs. 1-7; [0044-0045], [0047], [0053-0054], [0082], [0093-0098]). Ghoche is cited to clarify associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert identifier to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data (Ghoche Figs. 1-56; [0003], [0069-[0070], [0098], [0108], [0114-0115], [0128], [0131], [0148], [0146], [0156], [0158]). Thus, Ramchandran in view of Ghoche is considered to teach the claim.
Regarding claims 1, 12, and 17, applicant alleges Ramchandran fails to disclose "receiving a safety alert for a specific safety event from a user electronic device," initiating a chat session between a user and "a safety agent", and determining recommended reply messages "addressing a possible safety issue experienced by the user", and a safety alert management system. Ramchandran merely relates to enterprise customer- service interactions and do not disclose a safety alert for a specific safety event or a safety alert management application or system (see remarks pp. 14-16). Regarding claim 17, applicant further alleges Ramchandran fails to disclose “a trained safety chat language model” (see remarks p. 18) Examiner respectfully disagrees.
Ramchandran discloses an alert management system that receives alerts from users requesting a chat interaction with an agent ([0021]), [0045], [0057]). The alerts may be related to safety issues such as a power outage event, a sudden change in weather causing disruption of services, a local event such as a political rally or a union strike or a global event of local significance, etc ([0065]) and severe rain ([0067]). A chat session is initiated between the user and an agent regarding the safety issue ([0021], ([0045]). Replies are automatically determined by a trained artificial intelligence engine to address the user safety issue ([0056-0057]) The processor 202 may be configured to fetch the top trending responses and display the responses on the interface ([0074-0075]). Examiner notes the claims place no limitations on what the safety alert, specific safety event, safety agent, or possible safety issue must comprise. Thus, Ramchandran’s disclosure of alerts, events, agents, possible issues, trained language model, and an alert management system that are related to safety issues are considered within the broadest reasonable interpretation of the claimed limitations.
Regarding claim 12, Applicant further alleges Ramchandran fails to disclose associating, with a unique identifier in a database, the specific reply messages that were recommended during a particular chat session together with the entire message history of that chat session, the user data, and the agent-selected message (see remarks pp. 16-17). Ghoche fails to remedy these deficiencies of Ramchandran. Ghoche generally describes historical customer-support tickets, recommended answers using a trained model, and training machine-learning models using historical ticket information. (Ghoche, Abstract, [0072]-[0073], [0079], FIGS. 5, 8, and 14.) However, Ghoche does not disclose that the specific recommended reply messages presented during a particular chat session interaction are retained and associated, via a unique identifier, with the entire message history, user data, and agent-selected message for that interaction in a database. Instead, Ghoche only describes that historical ticket data used to train a model includes "a history of the initial question, answers by agents, and chat information associated with the ticket". (Ghoche, para. [0081].) Ghoche does not disclose the claimed database association structure in which the recommended reply messages themselves are associated together with the other claimed session data through a common identifier (see remarks p. 17).
Regarding claim 17, Applicant further alleges the references fail to teach "associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert ID to provide an associated second set of training data" And “inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data”. Ramchandran and Ghoche fail to render obvious these limitations at least because these references fail to disclose or suggest that the specific recommended reply messages presented during a particular chat session interaction are retained and associated, via a unique identifier, with the entire message history, user data, and agent-selected message for that interaction in a database. These references also fail to disclose or suggest that the specific recommended reply messages presented during a particular chat session interaction are used as training data to train a safety chat language model (see remarks p. 18). Examiner respectfully disagrees.
As discussed in the rejection above, Ramchandran discloses associating an entire message history of a chat session, reply messages, user data, and agent-selected messages together in a database (Ramchandran Figs. 1-7; [0021], [0044-0045], [0047], [0053-0054], [0056-0057], [0082], [0093-0098]). Ramchandran discloses presenting historical reply messages during particular chat sessions ([0047], [0052-0054], [0074]). Agent selections of the historical replies are tracked in the database to further train the machine learning system ([0054], [0082]). Ghoche discloses that customer support issues are assigned a unique ticket ([0003], [0069]) that covers the entire lifecycle of an issue ([0098]). Tickets associated with customer support issues lead to questions and answers being generated ([0069]), which are stored with the ticket in a database ([0070]). The stored history of tickets is used to train an AI engine to mimic the way human agents respond to common questions ([0080]). Questions and answers from historical tickets are tokenized and fed to an LLM ([0115]). Ghoche discloses the historic ticket data includes questions and answers in which at least some of the answers may be based on pre-approved (template) answers ([0131]). A ML classifier can be trained based on questions and answers in the tickets to infer (predict) the user's intention and identify template answers to automatically generate a response ([0131]). An answer from a past ticket is identified as a recommended answer to a new incoming ticket so that the support agent can use all or part of the recommended answer ([0108], [0114]).
While applicant alleges the claims require a specific association structure, examiner notes that the claims do not recite a structure and place no limitation on what the association requires. Thus, Ramchandran’s disclosure of associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message to provide an associated second set of training data (Ramchandran Figs. 1-7; [0044-0045], [0047], [0053-0054], [0082], [0093-0098]) in combination with Ghoche’s disclosure of associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert identifier to provide an associated second set of training data (Figs. 1-56; [0003], [0069-[0070], [0098], [0108], [0114-0115], [0128], [0131], [0148], [0146], [0156], [0158]) is considered within the broadest reasonable interpretation of claim 12. Ramchandran’s disclosure of associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message to provide an associated second set of training data; inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data (Ramchandran Figs. 1-7; [0044-0045], [0047], [0053-0054], [0082], [0093-0098]) in combination with Ghoche’s disclosure of associating the safety alert, the entire message history of the electronic chat session, the additional user data, the one or more recommended reply messages, and the agent-selected message with a unique alert identifier to provide an associated second set of training data and inputting the associated second set of training data into the safety chat language model and training the safety chat language model on the associated second set of training data (Ghoche Figs. 1-56; [0003], [0069-[0070], [0098], [0108], [0114-0115], [0128], [0131], [0148], [0146], [0156], [0158]) is considered within the broadest reasonable interpretation of claim 17.
Applicant states that the dependent claims recite all the limitations of the independent claims, and thus, are allowable in view of the remarks set forth regarding the independent claims. However, as discussed above, Ramchandran in view of Ghoche is considered to teach the independent claims, and consequently, the dependent claims are rejected.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Harada (US 20250104850 A1) see Figs. 1-15 and [0036-0043], [0054], [0061-0063].
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHN T REPSHER III whose telephone number is (571)272-7487. The examiner can normally be reached Monday - Friday, 8AM-5PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jennifer Welch can be reached at (571) 272-7212. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/JOHN T REPSHER III/ Primary Examiner, Art Unit 2143
Prosecution Insights