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 .
Claims 1-20 are presented for examination.
Claim Rejections - 35 USC § 103
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 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Liu et al (US 12531056 A1) hereafter Liu, and further in view of Zhang et al (US 20250200356 A1) hereafter Zhang.
With respect to claim 1, Liu teaches the system comprising:
one or more processors programmed to:
retrieve client interaction data comprising text communications between clients and a large language model (LLM) chatbot (Language modeling (LM) is used in determining the probability of a given sequence of words in a sentence and to perform various generative tasks. Language models are generative models, wherein language models may be large language models (LLMs). The system uses interaction history information (past user inputs, past system actions, past interactions, visual content, context data related to past interaction), user data, dialog data to prompt the LM to generate words. A system component that controls what actions the system takes in response to user inputs of a dialog may be referred as a chatbot [col. 2, line 10 - col. 3, line 25; col. 6, lines 35-55; col. 8, lines 40-55]);
input the client interaction data into a first generative artificial intelligence model trained to identify topics associated with the text communications, wherein the first generative artificial intelligence model is configured to output, for each of the topics, a label representing the topic (a prompt generation component may process the data to generate the prompt including words and/or topic. A topic herein may be a label that represents a particular topic. For example, a topic may represent “medical information”, “cold remedies” or “over the counter medication”, “congestion”. A data selector component may determine rare and/or unique words from the data, such that the component may filter certain words from the data that may be common words. The data selector component may keep words representing entities, and such component is configured to determine a topic from the data (a dialog data). The data selector component may determine one or more topics of the current dialog session [col. 9, line 55 – col. 10, line 45]);
rank the topics to determine a set of labels representing a subset of the topics that are most frequently identified within the text communications (a word may appear more than once (frequently) within the data. The data selector component may use ranking techniques to determine which words from the data to be included in the prompt. The ranking techniques may use factors such as recency of the words, and frequency of the words (words that appear frequently in the data may be ranked higher than words that appear once) [col. 9, line 55 – col. 10, line 45]);
generate first training data comprising the set of labels and one or more of the text communications classified by the first generative artificial intelligence model into each of the set of labels (the data selector component may use topic classification models to determine the topic. The prompt generation component may generate the prompt to include the words and/or the topic. The LM may process the prompt and generate the LM output that includes words [col. 10, line 45 – col. 11, line 20]);
train, using the first training data, a second generative artificial intelligence model to generate plain text descriptions for each label of the set of labels (if a topic is medical information, the words may be words related to medical information, such as symptoms or medical professional [col. 10, line 45 – col. 11, line 20]).
However, Liu does not disclose a system for using generative artificial intelligence to automatically label training prompts for training a classification model; retrieve unlabeled training data comprising a plurality of unlabeled sample prompts; responsive to the second generative artificial intelligence model being trained, generate second training data comprising the unlabeled training data, the set of labels, and the plain text descriptions generated for each label of the set of labels; input the second training data into a third generative artificial intelligence model to train the third generative artificial intelligence model to classify each of the plurality of unlabeled sample prompts into one or more of the subset of the topics, wherein the third generative artificial intelligence model outputs a plurality of labeled sample prompts, the plurality of labeled sample prompts comprising the plurality of unlabeled sample prompts each labeled with one or more labels from the set of labels; responsive to the third generative artificial intelligence model being trained, generate third training data comprising the plurality of labeled sample prompts; and train a classification model to classify input prompts into one or more of the subset of the topics based on the third training data.
In the same field of endeavor, Zhang teaches a system for using generative artificial intelligence to automatically label training prompts for training a classification model (an online system configured to provide an instruction prompt to follow and automatically generate labels based on the data provided to a machine-learned language model. The instruction prompt may include an instruction to generate an evaluation label of a training sample of a classification model. The language models are large language models [par. 0002, 0003, 0034, 0035]);
retrieve unlabeled training data comprising a plurality of unlabeled sample prompts (The training samples used for an unsupervised model may not be labeled. The training may be semi-supervised with a training set having a mix of labeled samples and unlabeled samples [par. 0158-0161]);
responsive to the second generative artificial intelligence model being trained, generate second training data comprising the unlabeled training data, the set of labels, and the plain text descriptions generated for each label of the set of labels (a model serving system may receive a request including input data and encodes the input data into a set of input tokens. The model serving system applies the machine-learned model to generate a set of output tokens. Each token in the set may correspond to a text unit. For example, a token may correspond to a word. The machine learning (ML) techniques may be used for training a classification model using training samples that have labels automatically generated. The ML models may be trained with a set of training samples that are labeled. The training samples may include, for example, query-item pairs, attributes, images, descriptions, and other metadata [par. 0032, 0075, 0081, 0158-0161]);
input the second training data into a third generative artificial intelligence model to train the third generative artificial intelligence model to classify each of the plurality of unlabeled sample prompts into one or more of the subset of the topics, wherein the third generative artificial intelligence model outputs a plurality of labeled sample prompts, the plurality of labeled sample prompts comprising the plurality of unlabeled sample prompts each labeled with one or more labels from the set of labels (an online system may receive plurality of responses from the LM, and each response includes an evaluation label corresponding to each evaluation request prompt. The online system may store evaluation labels and the data in the evaluation request prompts as training samples for the classification model. An online concierge system performs automatic label assignments for data using a machine-learned language model that may be used to automatically generate evaluation labels for outputs of the online concierge system. The online concierge system provides training samples to be labeled to the interface system [par. 0003, 0038-0041, 0063]);
responsive to the third generative artificial intelligence model being trained, generate third training data comprising the plurality of labeled sample prompts (The online system may store evaluation labels and the data in the evaluation request prompts as training samples for the classification model. Each response from the online system includes an evaluation label corresponding to the data in an evaluation request prompt [par. 0003, 0080, 0084]); and
train a classification model to classify input prompts into one or more of the subset of the topics based on the third training data (the online system may create a batch of evaluation request prompts by retrieving stored output of the classification model. The online system may also retrieve multi-modal data and/or out-of-band information to complement the stored inputs and outputs of the classification model. The online system may retrieve metadata, descriptions and other attributes with any items or concepts in the input or output [par. 0081, 0146]).
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 concept of generating an evaluation label of a training sample of a classification model as suggested by Zhang into the concept of using LLM to generate words that are relevant to a future user input as suggested by Liu because both of these systems addressing the process of using machine-learned model (LLMs) to generate outputs using classification model(s). Doing so would be desirable because the concept of Liu would be more efficient by automatically generating evaluation labels of training samples of a classification model based on the data provided to the language models (LLMs) (Zhang, [par. 0002, 0003]).
With respect to claim 2, Liu teaches the method comprising:
retrieving interaction data comprising communications between clients and a chatbot (Language modeling (LM) is used in determining the probability of a given sequence of words in a sentence and to perform various generative tasks. Language models are generative models, wherein language models may be large language models (LLMs). The system uses interaction history information (past user inputs, past system actions, past interactions, visual content, context data related to past interaction), user data, dialog data to prompt the LM to generate words. A system component that controls what actions the system takes in response to user inputs of a dialog may be referred as a chatbot [col. 2, line 10 - col. 3, line 25; col. 6, lines 35-55; col. 8, lines 40-55]);
identifying, based on the interaction data, a set of topics represented by the communications, wherein each communication is assigned one or more labels from a set of labels, and each label from the set of labels is associated with a topic from the set of topics (a prompt generation component may process the data to generate the prompt including words and/or topic. A topic herein may be a label that represents a particular topic. For example, a topic may represent “medical information”, “cold remedies” or “over the counter medication”, “congestion”. A data selector component may determine rare and/or unique words from the data, such that the component may filter certain words from the data that may be common words. The data selector component may keep words representing entities, and such component is configured to determine a topic from the data (a dialog data). The data selector component may determine one or more topics of the current dialog session [col. 9, line 55 – col. 10, line 45]);
generating first training data comprising the set of labels and one or more of the communications assigned to the label (the data selector component may use topic classification models to determine the topic. The prompt generation component may generate the prompt to include the words and/or the topic. The LM may process the prompt and generate the LM output that includes words [col. 10, line 45 – col. 11, line 20]);
training, using the first training data, a generative artificial intelligence model to generate metadata comprising a description of each label from the set of labels (if a topic is medical information, the words may be words related to medical information, such as symptoms or medical professional [col. 10, line 45 – col. 11, line 20]);
However, Liu does not particularly disclose a method for using generative artificial intelligence to automatically label training prompts for training a classification model, the method being executed by one or more processors of a computing system; generating second training data comprising a plurality of sample prompts, the set of labels, and the metadata; and training, using the second training data, a classification model to autonomously label each of the plurality of sample prompts with one or more labels from the set of labels.
In the same field of endeavor, Zhang teaches a method for using generative artificial intelligence to automatically label training prompts for training a classification model, the method being executed by one or more processors of a computing system (an online system configured to provide an instruction prompt to follow and automatically generate labels based on the data provided to a machine-learned language model. The instruction prompt may include an instruction to generate an evaluation label of a training sample of a classification model. The language models are large language models [par. 0002, 0003, 0034, 0035]);
generating second training data comprising a plurality of sample prompts, the set of labels, and the metadata (a model serving system may receive a request including input data and encodes the input data into a set of input tokens. The model serving system applies the machine-learned model to generate a set of output tokens. Each token in the set may correspond to a text unit. For example, a token may correspond to a word. The machine learning (ML) techniques may be used for training a classification model using training samples that have labels automatically generated. The ML models may be trained with a set of training samples that are labeled. The training samples may include, for example, query-item pairs, attributes, images, descriptions, and other metadata [par. 0032, 0075, 0081, 0158-0161]); and
training, using the second training data, a classification model to autonomously label each of the plurality of sample prompts with one or more labels from the set of labels (an online system may receive plurality of responses from the LM, and each response includes an evaluation label corresponding to each evaluation request prompt. The online system may store evaluation labels and the data in the evaluation request prompts as training samples for the classification model. An online concierge system performs automatic label assignments for data using a machine-learned language model that may be used to automatically generate evaluation labels for outputs of the online concierge system. The online concierge system provides training samples to be labeled to the interface system [par. 0003, 0038-0041, 0063]).
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 concept of generating an evaluation label of a training sample of a classification model as suggested by Zhang into the concept of using LLM to generate words that are relevant to a future user input as suggested by Liu because both of these systems addressing the process of using machine-learned model (LLMs) to generate outputs using classification model(s). Doing so would be desirable because the concept of Liu would be more efficient by automatically generating evaluation labels of training samples of a classification model based on the data provided to the language models (LLMs) (Zhang, [par. 0002, 0003]).
With respect to claim 3, the combination of Liu and Zhang teaches further comprising: generating, using the classification model, labeled training data comprising the plurality of sample prompts and the one or more labels assigned to each of the plurality of sample prompts (Zhang, a model serving system may receive a request including input data and encodes the input data into a set of input tokens. The model serving system applies the machine-learned model to generate a set of output tokens. Each token in the set may correspond to a text unit. For example, a token may correspond to a word. The machine learning (ML) techniques may be used for training a classification model using training samples that have labels automatically generated. The ML models may be trained with a set of training samples that are labeled. The training samples may include, for example, query-item pairs, attributes, images, descriptions, and other metadata [par. 0032, 0075, 0081, 0158-0161]).
With respect to claim 4, the combination of Liu and Zhang teaches further comprising:
training a machine learning model using the labeled training data; or monitoring streaming interaction data for prompts associated with one or more of the set of topics based on the labeled training data (Zhang, the online concierge system provides training samples to be labeled. The machine-learned language model can automatically mass produce labeled data such as training samples for a classification model for supervised learning [par. 0041, 0080]).
With respect to claim 5, the combination of Liu and Zhang teaches wherein the classification model comprises a first classification model, identifying the set of topics comprises:
inputting the interaction data into a second classification model to obtain a plurality of topics represented by the communications (Liu, Language models are generative models, wherein language models may be large language models (LLMs). The system uses interaction history information (past user inputs, past system actions, past interactions, visual content, context data related to past interaction), user data, dialog data to prompt the LM to generate words. A system component that controls what actions the system takes in response to user inputs of a dialog may be referred as a chatbot [col. 2, line 10 - col. 3, line 25; col. 6, lines 35-55; col. 8, lines 40-55]); and
ranking the plurality of topics based on a frequency with which the communications relate to each of the plurality of topics, wherein the set of topics is selected from the plurality of topics based on the ranking of the plurality of topics (Liu, a word may appear more than once (frequently) within the data. The data selector component may use ranking techniques to determine which words from the data to be included in the prompt. The ranking techniques may use factors such as recency of the words, and frequency of the words (words that appear frequently in the data may be ranked higher than words that appear once) [col. 9, line 55 – col. 10, line 45]).
With respect to claim 6, the combination of Liu and Zhang teaches wherein the second classification model comprises a generative artificial intelligence model (Liu, LM can be used to perform various tasks including generative tasks. The language models are generative models. The language models may be large language models. For example, the LM may be a GPT model or Alexa generative models [col. 2, lines 10-35; col. 6, lines 57-67]).
With respect to claim 7, the combination of Liu and Zhang teaches wherein the communications comprise client-input text communications and chatbot-output text communications (Liu, a dialog may be directed to the system performing a specific action requested by a user. A user may ask a system like Alexa for an answer to a question. System components that control what actions the system takes in response to user inputs may be referred as chatbots [col. 8, lines 40-55]), identifying the set of topics comprises:
determining, using the chatbot, based on the client-input text communications, the set of topics, wherein the chatbot is configured to determine an intent of each of the client-input text communications using one or more natural language processing (NLP) models and generate a corresponding chatbot-output text communication based on the intent (Liu, the LM may be used for generating context data for spoken language understanding (SLU) processing, such that SLU involves determining an intent or requests action from audio data representing a spoken input. Other examples include automatic speech recognition (ASR) component and a natural language understanding (NLU) component [col. 2, lines 10-35; col. 15, lines 25-35; col. 20, line 40 – col. 21, line 35]).
With respect to claim 8, the combination of Liu and Zhang teaches further comprising:
receiving a new sample prompt; and inputting the new sample prompt into the classification model to obtain a model-provided label for the new sample prompt (Zhang, an online system configured to provide an instruction prompt to follow and automatically generate labels based on the data provided to a machine-learned language model. The instruction prompt may include an instruction to generate an evaluation label of a training sample of a classification model. The language models are large language models. [par. 0002, 0003, 0034, 0035, 0173]).
With respect to claim 9, the combination of Liu and Zhang may not teach further comprising: determining a set of user-provided labels assigned to the new sample prompt by a plurality of authorized labelers; determining a first user-provided label of the set of user-provided labels to assign to the new sample prompt based on a number of authorized labelers that assigned the first user-provided label to the new sample prompt satisfying a threshold condition; and assigning, based on a similarity between the first user-provided label and the model-provided label satisfying a similarity condition, the model-provided label or the first user-provided label to the new sample prompt.
With respect to claim 10, the combination of Liu and Zhang may not teach further comprising: providing a user interface to each of the plurality of authorized labelers for inputting the set of user-provided labels.
With respect to claim 11, the combination of Liu and Zhang teaches wherein training the generative artificial intelligence model comprises:
obtaining reference metadata for each label from the set of labels, wherein the reference metadata comprises reference text description contextualizing the label (Zhang, The machine learning (ML) techniques may be used for training a classification model using training samples that have labels automatically generated. The ML models may be trained with a set of training samples that are labeled. The training samples may include, for example, query-item pairs, attributes, images, descriptions, and other metadata [par. 0032, 0075, 0081, 0158-0161]);
generating, using the generative artificial intelligence model, the metadata for each label from the set of labels (Zhang, each response from the online system includes an evaluation label corresponding to the data in an evaluation request prompt. Instead of storing the data as training samples for a classification model, the online system may store the evaluation label as one of the metadata fields in a data store [par. 0083-0085]); and
adjusting one or more parameters of the generative artificial intelligence model to discriminate between the reference metadata and the metadata for each label from the set of labels (Zhang, the online system adjusts, in backpropagation, one or more parameters of the machine-learned language model based on the comparison of the predicted labels and the evaluation labels [par. 0085]).
With respect to claim 12, the combination of Liu and Zhang teaches further comprising:
retrieving additional interaction data comprising additional communications between clients and the chatbot (Zhang, the customer client device may receive additional content from the online concierge system to present to a customer, such as coupons, recipes, or item suggestions [par. 0016-0020]);
identifying, based on the additional interaction data, one or more additional topics represented by the additional communications (Liu, additional context data may be incorporated in the ASR component, such as personalized data associated with a user profile. The personalized data may include words representing various user-specific/personalized information such as user’s contacts name list/address book. Other topics related to the user information includes device name, playlist name, shopping list name, etc. [col. 14, lines 15-45]); and
updating the set of labels to include one or more additional labels respectively associated with the one or more additional topics (Liu, the personalized data may be processed using the context encoder to generate corresponding personalized embedding data. Context data corresponding to a new domain or an updated domain may be used. The ASR component may not have been retrained/updated to recognize words corresponding to a new or updated domain, for example, a music domain may be updated to respond to requests related to recently released albums [col. 14, line 15 – col. 15, line 15]).
With respect to claim 13, the combination of Liu and Zhang teaches further comprising:
determining, based on the additional interaction data, that the additional communications include less than a threshold number of communications related to a first topic from the set of topics; and updating the set of labels by removing a first label associated with the first topic (Liu, the data selector component may determine rare or unique words from the data. The component may filter out some non-noun words, verbs, adjectives, etc. The component may keep words representing entities (a place, a person, a thing) and to remove/filter out words not representing entity. Rare or unique words may not be included in the training data of the ASR component [col. 9, line 55 – col. 10, line 30]).
With respect to claim 14, it is a non-transitory computer-readable claim that is corresponding to the method of claim 2. Therefore, it is rejected for the same as claimed in claim 2 above.
With respect to claim 15, it is a non-transitory computer-readable claim that is corresponding to the method of claim 3. Therefore, it is rejected for the same as claimed in claim 3 above.
With respect to claim 16, it is a non-transitory computer-readable claim that is corresponding to the method of claim 4. Therefore, it is rejected for the same as claimed in claim 4 above.
With respect to claim 17, it is a non-transitory computer-readable claim that is corresponding to the method of claim 5. Therefore, it is rejected for the same as claimed in claim 5 above.
With respect to claim 18, it is a non-transitory computer-readable claim that is corresponding to the method of claim 7. Therefore, it is rejected for the same as claimed in claim 7 above.
With respect to claim 19, it is a non-transitory computer-readable claim that is corresponding to the method of claim 8. Therefore, it is rejected for the same as claimed in claim 8 above.
With respect to claim 20, it is a non-transitory computer-readable claim that is corresponding to the method of claim 9. Therefore, it is rejected for the same as claimed in claim 9 above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Chiu (US 20250097169 A1) disclosed a generative chatbot system for virtual community and a method thereof are disclosed. In the system, a client-end host is linked to a virtual community to receive a chat message; when the trigger signal is detected, the client-end host generates an operation interface for inputting the customized instruction, and transmits the chat message and the customized instruction to the server-end host, which can integrate the chat message as the context message having a timing logic and transmit the context message and the customized instruction to an Al device to generate a response message. The server-end host stores the response message received from the artificial intelligence device to a response list, so that the client-end host can select one of the response messages from the response list and output the selected response message to the virtual community.
He et al (US 20240330591 A1) disclosed an invention that provides a computer-implemented method and system for analyzing texts. The method comprises the steps of identifying one or more sentences of a first text in a first language; identifying one or more sentences of a second text in a second language; translating the identified one or more sentences of the first text from the first language into the second language; processing the sentences of the first text in the second language and the sentences of the second text in the second language into a first representation of the sentences of the first text, and a second representation of the sentences of the second text; and comparing the first representation of sentences of the first text with the second representation of sentences of the second text to identify one or more sentences of similarity between the first text and the second text.
Sommerfield et al (US 12596885 B2) disclosed a computer-implemented labeling technique generates a task description that describes a labeling task to be given to a language model. The technique then sends a prompt to the language model, which includes the task description and a particular item to be labeled. The technique receives a response provided by the language model in response to the prompt, which specifies a class assigned by the language model to the item. In some implementations, the task description specifies a group of suggested classes to be used in classifying the particular item. The task description also invites the language model to specify another class upon a finding that none of the group of suggested classes applies to the item. The technique also allows a user to stop and restart a labeling run at any point in the labeling run. Other aspects of the technique include consensus processing and weight updating.
Rosenkranz et al (US 20240296425 A1) disclosed technologies receive, via a user interface, an input associated with a first user of a user connection network. The input identifies first position data related to a position capable of being filled by a hiring of a person. In response to validating the first position data, second position data different from the first position data is extracted from the user connection network, based on the first position data. A first prompt is formulated based on the first position data and the second position data. The first prompt is sent to a generative language model. A first piece of writing is received from the generative language model. The first piece of writing includes a position description output by the generative language model based on the first prompt. The position description is sent to the user interface in response to the input.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Quoc Phung whose telephone number is (703) 756 1330. The examiner can normally be reached on Monday through Friday from 9am to 5pm PT.
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) athttp://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jennifer Welch can be reached on 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 an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Q.L.P./Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
Prosecution Insights