DETAILED ACTION
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 § 112
Claims 1-18 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 1, the language “cause the at least one processor to implement a user reception unit configured to ………” “ ….. unit configured to ….” is confusing. It is not clear what is meant by a processor implementing a unit. Note that the unit, according to the claim’s language is already “configured”. This applies to all the claimed units.
Is the user “input” (line 5) the same as the user “request” (line 15)?
The language “…provide the user terminal with at least one of ….. automatically …… and ….human consultant” means that one response may be provided. This makes the following limitations “store two or more responses” and “ranking” them confusing since only one response may be provided.
Claim 10 has a similar issue with respect to “automatically generated response or a response of a human consultant. Meaning only one response may be provided (e.g., from a human consultant). Then the terminology ”two or more responses” and “rankable” becomes confusing. Note, according to the claim’s language, the “two or more response” may possibly be from the human consultant. ”. Is the user “input” the same as the user “request”?
In both claims 1 and 10, the use of the alternative language automatically generated vs. provided by human creates confusion as the response provided to the user may be only provided by the human consultant. The use of the “intelligent consultation unit” to provide a response to the user terminal appears to be optional according to the claims’ language.
The dependent claims 1-9 and 11-18 are rejected for the same reasons.
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, 7, 8, 10, 16, and 17 are rejected under 35 U.S.C. § 103 as being unpatentable over Banerjee US (2023/0351352 A1) in view of Hiraoka (US 2020/0050669 A1).
Regarding Claim 1, Banerjee teaches an intelligent call center consultation system based on artificial intelligence, comprising:
at least one processor; and a memory storing instructions which, when executed by the at least one processor, cause the at least one processor to implement: Banerjee Abstract: “An artificial intelligence (AI) system for guiding a user interaction in a phone call or chat session. The system includes a computer running an AI algorithm, such as a machine learning algorithm …” Banerjee [0119], see processing device/processor 220 (Fig. 1).
a user reception unit configured to receive a voice or text input from a user terminal; Banerjee [0107] (FIG. 7, box 702): “a user contacts the system via a phone call or by initiating a chat session.” Banerjee [0103]: “the system asks an initial question and receives a response from the user.”
an intelligent consultation unit configured to automatically generate a consultation response to the voice or text input; Banerjee [0103]: “based on the answer to the initial question, the system determines a next question to ask or a next action to perform.” Banerjee [0006]: “the AI algorithm adaptively guides the dialog … including identifying a next question to ask the user …, identifying an information item to provide the user …, and identifying an action to take.”
a consultant transceiver unit configured to perform transmission and reception of information to and from a consultant terminal of a human consultant; Banerjee [0105]: “it is advantageous to connect a live agent into the call or chat. When this determination is made, the live agent is connected at box 708.” Banerjee [0107]: “when the live agent is connected into the call or chat, the human agent presents the question or action to the user.”
a response provision unit configured to provide the user terminal with at least one of a response that is automatically generated by the intelligent consultation unit and a response of the human consultant received from the consultant transceiver unit; Banerjee [0107]: “the human agent presents the question or action to the user … When no live agent is connected into the call or chat, the system presents the question or action to the user, in a known manner.”
a ranking response storage unit configured to store two or more responses that are rankable based on user acceptance when the two or more responses are present with respect to a user request; Banerjee [0104]: “if the matter is not resolved by the most recent system action, the process returns to the box 706 to determine what to do next.” Banerjee [0112] (FIG. 7, box 716, “store data”): “the entire user interaction is stored in a database. This includes storing the questions and actions/information presented by the system, and the user’s responses. It also includes storing a final outcome—e.g., a likelihood that the user’s matter was resolved, and a satisfaction level of the user (“user acceptance.”)
a reinforced learning unit configured to train the intelligent consultation unit by using the two or more responses. Banerjee [0118] (FIG. 8, box 814, Perform update training of machine learning use interaction system): “the update training is performed at box 814 … includes the user interaction data from the database 810—that is, the calls and chats which were handled by the production system.” Banerjee [0007]: “Both satisfactory and unsatisfactory outcome examples are included in the training database, as the AI algorithm learns from both types of scenarios.”
Banerjee does not expressly teach that the two or more responses are “rankable” nor a “reinforced learning unit configured to train the intelligent consultation unit.”
Hiraoka teaches setting a score to each of a set of response candidates so they are rankable (Hiraoka Abstract: “a policy unit configured to set a score to each of response candidates included in a set of response candidates … and, referring to the set scores, to select one of the response candidates as a dialog act of the apparatus”), and a reinforced learning unit that obtains a reward and updates a policy (compensation) model (Hiraoka Abstract/[0014]: “a policy parameter updating unit configured to obtain a reward … using a reward function that … returns an evaluation of a behavior … and to update the policy parameter based on the obtained reward”; [0076]: “a Deep Q-Network as an algorithm for the reinforcement learning processing”).
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date to modify Banerjee so the two or more stored responses to an identical user’s request are rankable based on user acceptance and are used by a reinforced learning unit to train the intelligent consultation unit, as taught by Hiraoka, because Banerjee already stores acceptance/outcome-labeled responses and trains its model therefrom (Banerjee [0112], [0117]) with the goal that “the machine learning algorithm can be trained to recognize patterns in user interaction dialog which lead to satisfactory outcomes for the user” (Banerjee [0089]), and Hiraoka teaches this predictably enables “creation of a policy corresponding to operating circumstances” (Hiraoka [0011]) — a combination of known techniques yielding a predictable result, with a reasonable expectation of success. See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007).
Regarding Claim 7, Banerjee teaches the two rankable-response scenarios: (a) not-accepted automated response → subsequent automated response accepted (Banerjee [0110]: “the process returns to the box 706 where the system determines the next question or action”; [0104]); and (b) not-accepted automated response → dialogue continues to a human consultant response accepted (Banerjee [0105]: “the live agent is connected at box 708”; [0105]: “ambiguity in user responses, complexity of a transaction …”).
Hiraoka teaches candidate scoring/ranking (Hiraoka Abstract).
Thus, it would have been obvious to treat Banerjee’s earlier non-accepted response and the subsequent accepted response (automated or human) as the recited rankable responses, in view of Hiraoka’s candidate scoring, for the reasons stated above.
Regarding Claim 8, Banerjee teaches collecting/storing responses including, within the interaction, an unresolved (denial) response followed by a resolving (consent) response, with the stored dialogue history (Banerjee [0110]; [0112]).
Banerjee does not expressly teach a ranking response determination unit that determines responses to be rankable if a user input in the dialogue history comprises a consent expression after an expression of denial intent.
Hiraoka teaches a reward function that evaluates dialog behaviors against a stored speech history (Hiraoka Abstract; [0053]: policy parameter updated “such that the numerical value determined by the reward function is maximized”; [0066], dialog state “SPEECH HISTORY”).
Thus, it would have been obvious to configure a ranking response determination unit that determines rankability based on consent-after-denial, by applying Hiraoka’s reward/outcome evaluation (against the stored speech history) to Banerjee’s denial-then-consent loop and stored dialogue history, to identify the preferable response for training — a predictable result with a reasonable expectation of success.
Regarding Claim 10, Banerjee teaches the computer-implemented consultation method (See rejection of claim 1.
Banerjee does not expressly teach “rankable” responses or a “reinforced learning step.”
Hiraoka teaches scoring a set of candidates so they are rankable (Hiraoka Abstract) and a reinforced learning step updating a policy via a reward (as discussed above).
Thus, it would have been obvious to modify Banerjee’s method accordingly, for the reasons stated for Claim 1, with a reasonable expectation of success.
Claims 16 and 17 are rejected for the same reasons as claims 7 and 8, respectively.
Claims 2, 3, 4, 11, 12, and 13 are rejected under 35 U.S.C. § 103 as being unpatentable over Banerjee (US 2023/0351352 A1) in view of Hiraoka (US 2020/0050669 A1), and further in view of Mielke (US 2023/0135179 A1).
Regarding Claim 2, the combination of Banerjee and Hiraoka teaches all limitations of Claim 1. Claim 2 further recites that the intelligent consultation unit comprises:
a user intent prediction unit ….. , a task generation unit ….. , a response generation unit …
wherein the dialogue information comprises a dialogue history that is a dialogue record between the user and the intelligent call center consultation system and a dialogue state comprising information of the intent of the user …..
Banerjee teaches, as discussed above, generating the response based on the state/history of the dialog (Banerjee [0103]: “based on the answer to the initial question, the system determines a next question to ask or a next action to perform”; [0006]: “the AI algorithm adaptively guides the dialog to achieve the most favorable outcome based on the current status of the dialog”) and storing a dialogue history of the questions/actions presented and the user’s responses (Banerjee [0112]: “storing the questions and actions/information presented by the system, and the user’s responses”).
Banerjee and Hiraoka do not expressly teach the recited user intent prediction unit, task generation unit, and response generation unit generating a response from task information and a dialogue state that comprises intent and task information.
Mielke teaches these limitations:
User intent prediction unit: Mielke [0091]: “the NLU [natural Language Understanding] module 210 may process the domain classification/selection results using an intent classifier 336b. The intent classifier 336b may determine the user’s intent associated with the user input.”
Task generation unit: Mielke [0122]: “the task tracker 410 may track the task state associated with an assistant task”; [0123]: “the task tracker 410 may comprise … task candidate generation module 416” which generates task candidates for the assistant system to perform.
Response generation unit based on task and dialogue information: Mielke [0113]: “the CU composer 370 may generate a communication content for the user using a natural-language generation (NLG) component 372.”
Dialogue state (intent + task) and dialogue history: Mielke [0072]: “the dialog state tracker 218a may track state changes over time as a user interacts with the world and the assistant system.” Mielke [0122]: “a task state may be a data structure persistent cross interaction turns and updates in real time to capture the state of the task.”
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date to structure the intelligent consultation unit of Banerjee/Hiraoka to include Mielke’s user intent prediction unit, task generation unit, and response generation unit that generates a system response from the task information and a tracked dialogue state (comprising the intent and the task) and dialogue history. One of ordinary skill would have been motivated to do so because Mielke teaches that determining user intent, generating corresponding tasks, and tracking dialogue state overturns enables the system to “understand user intention, interact with the user and the world to complete the right task for the user” (Mielke [0117]), predictably improving the accuracy and relevance of the automatically generated consultation responses in Banerjee. This is the application of a known conversational-AI architecture (intent → task → NLG response, with dialogue-state tracking) to a known AI consultation system to yield a predictable result, with a reasonable expectation of success. See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007).
Regarding Claim 3, the combination teaches all limitations of Claim 2. Claim 3 further recites that the user intent prediction unit comprises:
a slot tagging prediction unit …… an intent analysis prediction unit ……
wherein the user intent prediction unit operates in a dynamic cyclic pipeline way to predict and generate the slot value that is not explicitly expressed in the voice or text input ……
Mielke teaches the slot tagging prediction unit and the intent analysis prediction unit:
Slot tagging: Mielke [0091]: “the NLU module 210 may then use a slot tagger 338b to annotate one or more slots associated with the user input. In particular embodiments, the slot tagger 338b may annotate the one or more slots for the n-grams of the user input.”
Intent analysis / classification: Mielke [0091]: “The intent classifier 336b may determine the user’s intent associated with the user input.”
Handling of unclear intent (disambiguation): Mielke [0104]: “a disambiguation intent handler to handle the intents when the assistant system 140 asks for disambiguation.” Mielke [0111]: “This capability may allow the assistant system 140 to interact with the user not only to collect missing slot values but also to reduce ambiguity of more complex/ambiguous utterances to complete the task.”
Banerjee, Hiraoka, and Mielke do not expressly recite re-processing in the specific “dynamic cyclic pipeline” manner (transmitting the intent-analysis results back to slot tagging to predict a slot value not explicitly expressed). However, it would have been obvious to one of ordinary skill to re-process the input in this cyclic manner — feeding intent-analysis results back to slot tagging to predict a slot value not explicitly expressed when the intent is not clearly revealed — as a predictable application of Mielke’s disambiguation/ambiguity-reduction mechanism (Mielke [0104], [0111]) to resolve intents that are not clearly revealed by the initial slot tagging. Doing so would predictably improve intent understanding for inputs in which the intent is not explicitly stated, with a reasonable expectation of success.
Regarding Claim 4, the combination of references teaches all limitations of Claim 3. Claim 4 further recites:
the task comprises a clarification question task …..
Banerjee, as discussed above, teaches determining and asking a next question of the user (Banerjee “the system determines a next question to ask”).
Mielke teaches generating clarification/disambiguation to resolve unclear intent and to collect missing/ambiguous information (Mielke [0104]: disambiguation intent handler; [0111]: “interact with the user not only to collect missing slot values but also to reduce ambiguity of more complex/ambiguous utterances”).
Thus, it would have been obvious to generate a clarification question task — including the predicted results and the part that needs to be checked — when the intent is not explicitly expressed, by applying Mielke’s disambiguation/clarification mechanism (Mielke [0104], [0111]) together with Banerjee’s next-question determination (Banerjee [0103]). Doing so would predictably improve intent resolution for ambiguous inputs by proactively confirming the uncertain content with the user, with a reasonable expectation of success.
Claim 11 is rejected for the same reason as claim 2.
Claim 12 is rejected for the same reason as claim 3.
Claim 13 is rejected for the same reason as claim 4.
Claims 5 and 14 are rejected under 35 U.S.C. § 103 as being unpatentable over Banerjee, Hiraoka and Miekle, as applied above, and further in view of Huang (US 11,989,524 B2).
Regarding Claim 5, the combination of Banerjee, Hiraoka and Mielke, as discussed above, teaches all limitations of Claim 2. Claim 5 further recites:
the user intent prediction unit further comprises a knowledge search unit …….
Banerjee teaches generating the next task/action based on the dialog state and a task generation function (as combined with Mielke in the above rejections).
Banerjee, Hiraoka, and Mielke do not expressly teach a knowledge search unit that searches for knowledge necessary for the consultation dialogues with reference to the dialogue information and provides the retrieved knowledge to task generation.
However, the use of knowledge-grounded dialogue system that searches for supporting knowledge with reference to dialogue information and uses the retrieved knowledge to generate the system’s turn-specific task/response is old and well-known in the art. As an example:
Huang teaches a knowledge-grounded dialogue system that searches for supporting knowledge with reference to dialogue information and uses the retrieved knowledge to generate the system’s turn-specific task/response. Huang claim 1: “a supporting sentence searcher configured, to support the sentence to the correct answer in the passage by referring to the question and the correct answer of the reading comprehension data.” Huang (FIGS. 6–7, and corresponding description staring on col. 8, line 26: “a supporting sentence searching unit 202 configured to search for supporting sentences from a passage … using the turn-specific ground information … as a query, and output supporting knowledge”; the retrieved supporting knowledge is combined with the dialogue history and passage-related exercises to generate the system utterance for the turn.
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date to add the well-known (as also shown by Huang) knowledge search (supporting-sentence search) to the intent prediction unit of the Banerjee/Hiraoka/Mielke system, such that knowledge necessary for the consultation is retrieved with reference to the dialogue information and provided to the task generation unit for generating a detailed task. One of ordinary skill would have been motivated to do so because Huang teaches that grounding dialogue generation in retrieved supporting knowledge enables “more accurate search for supporting sentences and generation of system utterances” (Huang [0047]), predictably improving the accuracy and relevance of the tasks/responses generated by Banerjee’s consultation system. This is the application of a known knowledge-grounding technique to a known AI consultation system to yield a predictable result, with a reasonable expectation of success. See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007).
Claim 14 is rejected for the same reason as claim 5.
Claims 6 and 15 are rejected under 35 U.S.C. § 103 as being unpatentable over Banerjee, Hiraoka and Mielke, as discussed above, and further in view of Rosen (US 2025/0350683 A1).
Regarding Claim 6, the combination of Banerjee, Hiraoka and Mielke, as discussed above, teaches all limitations of Claim 2. Claim 6 further recites:
the intelligent consultation unit further comprises a summarization unit
Banerjee, as discussed above, teaches determining that intervention of a human consultant is required and connecting the live agent : “the live agent is connected at box 708”; reasons a live agent is needed).
Banerjee, Hiraoka, and Mielke do not expressly teach generating a consultation dialogue summary of the dialogue up to the point of intervention and transmitting that summary to the human consultant’s terminal.
Rosen teaches generating, upon a monitored conversation condition, a summary of the interaction and communicating that summary to a user such as a supervisor or agent for use in the conversation. Rosen [0029, 0075]: “a prompt for summary creates a short summary of the interaction.” Rosen claim 5 / [0134]: the response “comprises: a summary of an interaction between a customer and an agent during the monitored conversation.” Rosen [0120]/[0080] (coaching/summary communicated to an external agent/supervisor application): the summary and insights are forwarded to a supervisor/coaching application for use by the human.
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date to add Rosen’s conversation summarization to the Banerjee/Hiraoka/Mielke system, such that when human-consultant intervention is required, a summarization task is generated, a consultation dialogue summary of the dialogue up to that point is created, and the summary is transmitted through the consultant transceiver unit to the consultant terminal. One of ordinary skill would have been motivated to do so because Rosen teaches that providing a summary of the interaction to a human (supervisor/agent) helps the human quickly understand the situation and reduces response time as taught by Rosen, and because Banerjee already hands off the interaction to a live agent — so providing the agent a summary of the dialogue up to that point predictably enables the agent to continue the consultation efficiently without restarting. This is the application of a known summarization/handoff technique to a known AI consultation system to yield a predictable result, with a reasonable expectation of success. See KSR, 550 U.S. 398.
Claim 15 is rejected for the same reason as claim 6.
Claims 9 and 18 are rejected under 35 U.S.C. § 103 as being unpatentable over Banerjee and Hiraoka, as discussed above, in view of and Rosen (US 2025/0350683 A1).
Regarding Claim 9, the combination of Banerjee and Hiraoka teaches all limitations of Claim 1. Claim 9 further recites:
the intelligent consultation unit is implemented with one unified intelligent consultation artificial intelligence model capable of executing multi-tasking …….
Banerjee teaches the AI consultation model trained on user-interaction (dialogue) data.
Banerjee and Hiraoka do not expressly teach a single unified multi-tasking model obtained by instruction tuning a pre-trained LLM on the recited mixture of data (intent prediction, task generation, tagged dialogue, tagged summarization, untagged dialogue pairs, and untagged summarization data).
Mielke teaches a unified model that performs multiple dialogue subtasks (NLU/intent, dialogue-state, response generation) trained jointly rather than as isolated components. Mielke [0286]: “the sub-components that can be viewed as a single hybrid E2E model”; Mielke [0281]: “the overall function of NLU 210, DST 218, and DP 360 may be combined into a larger model, which may be optimized for its overall output … enabling end-to-end fine-tuning of sub-components.” Mielke [0262] (large pre-trained language models, e.g., GPT-2/BART/GPT-3, and prompting/fine-tuning to construct TOD systems). Mielke thus supplies the “unified … multi-tasking” model built on a pre-trained LLM and trained on intent-prediction, task-generation, and tagged dialogue data.
Rosen teaches training/prompting a large language model with a mixture that includes summarization data — i.e., using summary prompts alongside insight/recommendation prompts for a generative LLM (Rosen [0029]: “a prompt for summary 222”; Rosen [0131]: the executable instruction prompts “an output from a large language model (LLM) for a response” covering summary, insight, and recommendation). Rosen thus supplies the summarization-data component (tagged and untagged summarization data mapping dialogues to summaries) of the recited training mixture.
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date to implement the intelligent consultation unit of Banerjee/Hiraoka as a single unified multi-tasking model obtained by instruction tuning a pre-trained LLM on the recited mixture of intent-prediction, task-generation, tagged dialogue, and summarization data (tagged and untagged), as taught by Mielke and Rosen. One of ordinary skill would have been motivated to do so because Mielke teaches that unifying the dialogue subcomponents into a single model “which may be optimized for its overall output” reduces cascading errors and simplifies the architecture as taught by Mielke, and Rosen teaches leveraging a pre-trained LLM prompted on summary/insight/recommendation data to generate responses. Combining these known LLM-based multi-task training techniques with Banerjee’s consultation model predictably yields a single model capable of performing intent prediction, task generation, response generation, and summarization — the application of known techniques to a known system to yield a predictable result, with a reasonable expectation of success. See KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007).
Claim 18 is rejected for the same reason as claim 9.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 10, 11, and 14 are rejected under 35 U.S.C. § 102(a)(2) as being anticipated by Mielke et al. (US 2023/0135179 A1) (hereinafter "MIELKE").
Regarding claim 10, MIELKE discloses the limitation of "A computer-implemented consultation method in an artificial intelligence-based consultation system, comprising". MIELKE teaches an assistant system 140 (Fig. 1) that processes user inputs and provides responses (Abstract). It utilizes computer-implemented modules such as ASR, NLU, and a dialog manager to manage user interactions, see "the processors operable when executing the instructions to receive a user input, process the user input using the ASR module, the NLU module, the dialog manager, one or more of the agents, the arbitrator, and the delivery system, and provide a response to the user input", Abstract and [0057].
MIELKE discloses the limitation of "receiving a voice or text input from a user terminal". MIELKE teaches receiving user-generated inputs via various modalities ([0006], [0041]) such as receiving speech and text inputs from a client system. See "the user may interact with the assistant system 140 by providing user input to the assistant application 136 via various modalities (e.g., audio, voice, text...)" [0041], [0118]).
MIELKE discloses the limitation of "automatically generating a consultation response to the voice or text input". MIELKE teaches generating responses using natural-language generation ([0056]). That is, it automatically constructs natural-language responses to user queries. See "The assistant system 140 may generate a response for the user regarding the information or services by using natural-language generation (NLG)", [0056], [0057].
MIELKE discloses the limitation of "providing the automatically generated response or a response of a human consultant to the user terminal". MIELKE teaches presenting generated responses to the user at the client system (para. [0041]). It teaches transmitting the generated response back to the user's client device. See “The assistant application 136 may then present the responses to the user at the client system 130 via various modalities", [0041] .
MIELKE discloses the limitation of "storing two or more responses that are rankable based on user acceptance when the two or more responses are present with respect to a user request". MIELKE teaches ranking generated communication content (para. [0114]). It teaches utilizing a response ranker to evaluate and prioritize multiple (two or more) generated response candidates. See "Since the generated communication content may be considered as a response to the user input, the CU composer 370 may additionally rank the generated communication content using a response ranker 376", [0114] and [0081]. Ranking involves two or more.
MIELKE discloses the limitation of "training the consultation system by using the two or more responses". MIELKE teaches implementing reinforcement learning frameworks, [0101]. It teaches using feedback (responses)and reinforcement learning to optimize dialog policies and response generation. See "the dialog manager 216 may implement reinforcement learning frameworks to improve the dialog optimization", [0101].
Regarding claim 11, MIELKE further discloses the limitation of "analyzing and predicting an intent of a user from a user the voice or text input". MIELKE teaches an intent classifier that determines user intent,[0091]. It calculates probabilities of user input being associated with predefined intents. See "The intent classifier 336b may determine the user's intent associated with the user input", [0091] and [0121].
For "generating a task that needs to be performed by the consultation system based on the intent of the user", MIELKE teaches generating candidate tasks based on the current state, [0102]. It converts upstream NLU results into candidate tasks for execution. See "the dialog state tracker 218 may convert the upstream results into candidate tasks using task specifications... Given the current state, the dialog state tracker 218 may generate candidate tasks", [0102].
For "generating a system response based on information of the task and dialogue information", MIELKE teaches generating responses using a response generation component [0115]. It utilizes the active dialog state and task specifications to construct natural-language responses. See "These tasks may include writing (i.e., storing/updating) the dialog state into the data store 330 using the dialog state writing component 382 and generating responses using the response generation component 380", [0115].
For "wherein the dialogue information comprises a dialogue history that is a dialogue record between the user and the consultation system …………intent of the user ……", MIELKE teaches storing previous conversations and resolving task IDs using intents and dialog state, [0123]. MIELKE's dialog manager is interpreted as disclosing the claimed limitation of "and a dialogue state comprising information of the intent of the user and the information of the task; wherein the dialogue information comprises a dialogue history that is a dialogue record between the user and the intelligent call center consultation system" because it maintains a record of past interactions and tracks active user intents and task parameters. This interpretation is supported by MIELKE's disclosure that "The dialog manager 216 may additionally store previous conversations between the user and the assistant system 140" ([0101]) and "the task ID resolution component 417 may resolve the task ID using intents, slots, dialog state, context, and user memory" ([0123]).
Regarding claim 14, MIELKE further discloses the limitation of "analyzing and predicting the intent of the user comprises searching for knowledge that is necessary for consultation dialogues with reference to the dialogue information". MIELKE teaches resolving user intent based on dialog history and communicating with knowledge graphs,[0101], [0098], [0104]. MIELKE's entity resolution is interpreted as disclosing the claimed limitation of "analyzing and predicting the intent of the user comprises a step of searching for knowledge that is necessary for consultation dialogues with reference to the dialogue information" because it queries external knowledge graphs using contextual data from the dialogue history to resolve entities and intent. This interpretation is supported by MIELKE's disclosure that "the dialog intent resolution 356 may resolve the user intent associated with the current dialog session based on dialog history" ([0101]) and "the entity resolution module 212 may communicate with different graphs 352 including one or more of the social graph, the knowledge graph, or the concept graph to extract ontology data" ([0099]), [0093].
For generating a task comprises generating one of detailed tasks that need to be performed by the intelligent consultation system based on the retrieved information and the dialogue information", MIELKE teaches generating candidate tasks based on available data from a knowledge graph and user task history ([0102]). MIELKE's task generation is interpreted as disclosing the claimed limitation of "generating a task; comprises generating one of detailed tasks that need to be performed by the intelligent consultation system based on the retrieved information and the dialogue information" because it dynamically constructs candidate tasks using both retrieved knowledge graph data and historical dialogue state. This interpretation is supported by MIELKE's disclosure that "the dialog state tracker 218 may generate candidate tasks based on available data from, for example, a knowledge graph, a user memory, and a user task history", [0102], [0127].
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Sherstinsky et al., US 10,410,626 B1 (“Progressive Classifier”), discloses a crowd-sourced customer-service platform in which customer service inquiries are received (via phone, chat, text, email, or the like) and routed to a pool of independent human experts and/or to automated response logic. The automated response logic generates at least partial or complete responses, including predetermined answers, and an arbitrator/matching component ranks and selects a best result among candidate responses, with responses logged and scored by scoring logic.
Sherstinsky further discloses a progressive machine-learning approach that combines an unsupervised machine-learning system, which identifies clusters of inquiries/answers, with a supervised machine-learning system, wherein human input is used to improve the quality of the training data before that data is used to train the supervised model — including expert confirmation and merging of clusters and human review distinguishing true-positive from false-positive assignments to refine the training set.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 AHMAD F. MATAR whose telephone number is (571)272-7488. The examiner can normally be reached M-F 9 - 5:30.
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.
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.
/AHMAD F. MATAR/ Supervisory Patent Examiner, Art Unit 2693
