DETAILED ACTION
Continued Examination under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 05/10/2026 has been entered.
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 .
Examiner’s Comment
This Action is in response to the Request for Continued Examination filed on 05/10/2026 with Amended Claims and Applicant's Remarks filed on 04/10/2026.
Applicant has amended claims 1, 11, and 13, canceled claims 3, 4, 14, 16, and 17 according to Amendments filed on 04/10/2026. Claims 1, 5-11, 13, and 15 are pending and currently under consideration for patentability.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1, 5-11, 13, and 15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claims are directed to a judicial exception (i.e., a law of nature, natural phenomenon, or abstract idea) without significantly more.
Step 1: In a test for patent subject matter eligibility, claims 1, 5-11, 13, and 15 are found to be in accordance with Step 1 (see 2019 Revised Patent Subject Matter Eligibility), as they are related to a process, machine, manufacture, or composition of matter. Claims 1 and 5-9 recite a method, claim 10 recites a computer-readable medium, and claims 11, 13, and 15 recite a system. When assessed under Step 2A, Prong I, they are found to be directed towards an abstract idea. The rationale for this finding is explained below:
Step 2A, Prong I: Under Step 2A, Prong I, claims 1, 10, and 11 are directed to an abstract idea without significantly more, as they all recite a judicial exception. Claims 1, 10, and 11 recite limitations directed to the abstract idea including “creating LLM results from a prompt of a user; selecting an information provider from a plurality of information providers when the prompt of the user is a prompt for providing content of the information provider or when content of the information provider may be provided in associated with the prompt by the user; creating a prompt for creation of a question related to the LLM results and the content of the information provider and creating a plurality of question candidates by inputting the created prompt; selecting at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results; providing the created selected at least one question as a response to the prompt of the user with the created LLM results; and providing additional content to the user based on an answer to the selected at least one question provided by the user, wherein the content of the information provider is dynamically generated by using one or more assets pre-registered by the information provider, and wherein the correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates.” These further limitations are not seen as any more than the judicial exception. Creating results based on a model (i.e. LLM results based on a large language model), creating a question based on the results, and providing the created question alongside results is considered to be an abstract idea under mental processes because the claims are directed to concepts performed in the human mind (including an observation, evaluation, judgment, opinion). Creating results based on a model (i.e. LLM results based on a large language model), creating a question based on the results, and providing the created question alongside results is also considered to be an abstract idea under certain methods of organizing human activity because the claims are directed to commercial or legal interactions (including agreements in the form of contracts; legal obligations; advertising, marketing or sales activities or behaviors; business relations) and managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions). Claims 1, 10, and 11 recite additional limitations including “by the at least one processor, and based on a / to the large language model (LLM).” Therefore, under Step 2A, Prong I, claims 1, 10, and 11 are directed towards an abstract idea.
Step 2A, Prong II: Step 2A, Prong II is to determine whether any claim recites any additional element that integrate the judicial exception (abstract idea) into a practical application. Claims 1, 10, and 11 recite additional limitations including “by the at least one processor, and based on a / to the large language model (LLM).” These limitations are seen as adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f). Accordingly, alone, and in combination, these additional elements are seen as using a computer or tool to perform an abstract idea, adding insignificant-extra-solution activity to the judicial exception. They do no more than link the judicial exception to a particular technological environment or field of use, i.e. processors/large language model, and therefore do not integrate the abstract idea into a practical application. The courts decided that although the additional elements did limit the use of the abstract idea, the court explained that this type of limitation merely confines the use of the abstract idea to a particular technological environment and this fails to add an inventive concept to the claims (See Affinity Labs of Texas v. DirecTV, LLC,). Under Step 2A, Prong II, these claims remain directed towards an abstract idea.
Step 2B: Claims 1, 10, and 11 recite additional limitations including “by the at least one processor, and based on a / to the large language model (LLM).” These additional limitations do not integrate the judicial exception (abstract idea) into a practical application because of the analysis provided in Step 2A, Prong II. Claims 1, 10, and 11 do not include additional elements or a combination of elements that result in the claims amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements listed amount to no more than mere instructions to apply an exception using a generic computer component. In addition, the applicant’s specifications disclose “any other device capable of responding to and executing instructions in a defined manner”, ¶ [00110], for implementing the processor, which do not amount to significantly more than the abstract idea of itself, which is not enough to transform an abstract idea into eligible subject matter. Furthermore, there is no improvement in the functioning of the computer or technological field, and there is no transformation of subject matter into a different state. Under Step 2B in a test for patent subject matter eligibility, these claims are not patent eligible.
Dependent claims 5-9 and 13, 15 further recite the method and system of claims 1 and 11, respectively. Dependent claims 5-9, 13, and 15 when analyzed as a whole are held to be patent ineligible under 35 U.S.C. 101 because the additional recited limitation fail to establish that the claims are not directed to an abstract idea:
Under Step 2A, Prong I, these additional claims only further narrow the abstract idea set forth in claims 1, 10, and 11. For example, claims 5-9, 13, and 15 describe the limitations for creating LLM results based on a large language model, creating a question based on the LLM results, and providing the created question alongside the LLM results – which only further narrows the scope of the abstract idea recited in the independent claims.
Under Step 2A, Prong II, for dependent claims 5-9, 13, and 15, there are no additional elements introduced. Thus, they do not present integration into a practical application, or amount to significantly more.
Under Step 2B, the dependent claims do not include any additional elements that are sufficient to amount to significantly more than the judicial exception. Additionally, there is no improvement in the functioning of the computer or technological field, and there is no transformation of subject matter into a different state. As discussed above with respect to integration of the abstract idea into a practical application, the additional claims do not provide any additional elements that would amount to significantly more than the judicial exception. Under Step 2B, these claims are not patent eligible.
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.
Claim(s) 1, 5-11, 13, and 15 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Publication 2024/0289361 to Batina in view of U.S. Publication 2023/0274086 to Tunstall and in further view of U.S. Publication 2024/0046318 to Muriqi.
Claims 1, 5-9; 10; and 11, 13, and 15 are method, computer-readable medium, and system claims, respectively, with substantially indistinguishable features between each group. For purposes of compact prosecution, the Office has grouped the common method, system and non-transitory computer readable storage medium claims in applying applicable prior art.
With respect to Claim 1:
Batina teaches:
A question recommendation method of a computer device comprising at least one processor, the question recommendation method comprising (Batina: ¶ [0024]):
creating, by the at least one processor, LLM results based on a large language model (LLM) from a prompt of a user (i.e. creating results of cowboy hats based on prompt from user) (Batina: ¶ [0035] “For example, the generated enhancement data may include the search query and/or additional keywords or synonyms. Additionally, or alternatively, the generated enhancement data may be in a format associated with the embedding space. For example, if the embedding space is a sentence embedding space, the generated enhancement data may include full sentences based on the search query. Additionally, or alternatively, the generated enhancement data may be based on a desired level of specificity or generalization for a search of the embedding space. For example, the generated enhancement data may comprise text that results in a vector embedding to be used in a search that is, e.g., spatially equidistant to many types of "hats" (more general), rather than, e.g., spatially closer to a single color of cowboy hat (more specific). The original user input, the text generated by the LLM, or a combination thereof may be transformed into a vector embedding to be used in search.”);
creating, by the at least one processor, a prompt for creation of a question related to the LLM results and the content of the information provider and creating a plurality of question candidates by inputting the created prompt to the large language model (i.e. creating a prompt to generate a plurality of questions from several rounds of chat) (Batina: ¶ [0040] “In some embodiments, additional context data received from a user may be processed to generate follow-up questions for the user to provide a chat-guided search experience. The additional context may comprise, for example, a user selection of one or more objects (e.g., products in an e-commerce context), indicators thereof, and/or images or media thereof. The system may process the user selection (or several such selections, e.g., from several rounds of a chat involving a user) to determine one or more object attributes that may be relevant to the search. The object attributes and/or the additional context may then be: inputted into an LLM to generate further follow-up prompts/questions for the user; used to generate collections (e.g., of one or more objects, indicators, images, and/or media) for display and/or for prompting further user selections; and/or otherwise used as input for a further search ( e.g., as further enhancement text for a vector search). For clarity, the generated collections may themselves be the result of a further search (e.g., a vector search carried out using enhancement text that is based on the object attributes).” Furthermore, as cited in ¶ [0043] “When a search is initiated by a user, the chatbot presents questions to the user in order to progressively gather relevant context for the search. An LLM may be employed for generating the questions and processing user responses thereto as part of gathering context. Upon receiving a user response to a question, the set of images presented in the image canvas may be updated. The user can select an image of a product to indicate their interest and/or preference for the product. The user selection serves to refine the product search. In this way, both the user's chat data and image selections are used to reduce the set of search results for a user's search query.”);
providing, by the at least one processor, the selected at least one question as a response to the prompt of the user with the created LLM results (i.e. providing follow up questions as a response to the prompt of cowboy hats) (Batina: ¶ [0038] “The described technique of enhancing user input using an LLM prior to a vector search may find application in various different contexts. In an example implementation, a chatbot application that is configured for performing various types of searches may use an LLM for enhancing user input to the chatbot. The chatbot may initially ask a user for text input (e.g., the user's product of interest). Upon receiving input of the user's initial response (e.g., "cowboy hat"), the chatbot may present follow-up questions for the user ( e.g., when does the user want to wear the cowboy hat). The user may provide further responses to the questions (e.g., "in the summer'). The LLM may infer additional properties of a product that the user is searching for ( e.g., lightweight, summer-suitable, etc.) based on the user's responses to the chatbot prompts. The additional properties are captured using vector embeddings and then used in a vector search. The vector search may identify products that are similar to, or otherwise related to, the user's desired product(s). The LLM may be trained on a plurality of different search spaces, including the vector space. In particular, the enhancement text may be generated based on text data available in multiple search spaces.”); and
providing additional content to the user based on an answer to the selected at least one question provided by the user (i.e. providing additional content based on user’s answer to question) (Batina: ¶¶ [0043] [0044] “When a search is initiated by a user, the chatbot presents questions to the user in order to progressively gather relevant context for the search. An LLM may be employed for generating the questions and processing user responses thereto as part of gathering context. Upon receiving a user response to a question, the set of images presented in the image canvas may be updated. The user can select an image of a product to indicate their interest and/or preference for the product. The user selection serves to refine the product search. In this way, both the user's chat data and image selections are used to reduce the set of search results for a user's search query…The image canvas may be dynamically updated based on user interaction with the chatbot and the product images. In particular, the chat data and image selections may facilitate identification of relevant product attributes for the search. When a user selects an image, a search may be conducted to identify similar items.”),
wherein the correlation is computed based on [[a number of]] the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates (i.e. embedding or correlation is computed based on products or assets pre-registered by information provider, a presence of content that contain product, and suitability of questions/prompts are based on correlation/embeddings of text such as hats and other products in the same category and on correlation/embedding of analyzing sentence structure of the inputs using NLP) (Batina: ¶¶ [0035] [0036] “For example, the generated enhancement data may include the search query and/or additional keywords or synonyms. Additionally, or alternatively, the generated enhancement data may be in a format associated with the embedding space. For example, if the embedding space is a sentence embedding space, the generated enhancement data may include full sentences based on the search query. Additionally, or alternatively, the generated enhancement data may be based on a desired level of specificity or generalization for a search of the embedding space. For example, the generated enhancement data may comprise text that results in a vector embedding to be used in a search that is, e.g., spatially equidistant to many types of "hats" (more general), rather than, e.g., spatially closer to a single color of cowboy hat (more specific). The original user input, the text generated by the LLM, or a combination thereof may be transformed into a vector embedding to be used in search…A general search for a product or product category (e.g., pants) that is initiated by the user may result in creation of a query embedding that corresponds to a specific product/category (e.g., women's jeans), as opposed to a general vector.” Furthermore, as cited in ¶ [0045] “In some implementations, the chatbot may present prompts for the user to identify attributes that the user likes about an initial set of similar products to the user's selection, where the attributes are determined based on what may be unique about the selection for the particular product category or the particular product selected. For example, the system may determine features of the selection in the vector space that are different/sufficiently distant from features of other objects/products (e.g., other embeddings) in the same vector space. To determine which product attributes may be most relevant to a buyer, the system may identify the most common attributes of products in the product category of the selected product.” Furthermore, as cited in ¶ [0082] “In at least some implementations, the search engine 114 is configured to perform vector searches. As previously described, a vector search uses vector embeddings for representing and searching content. The search engine 114 indexes queries and searchable objects (e.g., text, image, documents, data records, etc.) of a library with vector embeddings. In particular, the query set and searchable objects are each mapped to a vector in a common embedding space. An embeddings module 116 creates the vector representations of data.” Furthermore, as cited in ¶ [0063] “An example of how the transformer 50 may process textual input data is now described. Input to a language model (whether transformer-based or otherwise) typically is in the form of natural language as may be parsed into tokens. It should be appreciated that the term "token" in the context of language models and NLP has a different meaning from the use of the same term in other contexts such as data security. Tokenization, in the context of language models and NLP, refers to the process of parsing textual input ( e.g., a character, a word, a phrase, a sentence, a paragraph, etc.) into a sequence of shorter segments that are converted to numerical representations referred to as tokens ( or "compute tokens"). Typically, a token may be an integer that corresponds to the index of a text segment ( e.g., a word) in a vocabulary dataset. Often, the vocabulary dataset is arranged by frequency of use. Commonly occurring text, such as punctuation, may have a lower vocabulary index in the dataset and thus be represented by a token having a smaller integer value than less commonly occurring text.”), and
Batina does not explicitly disclose selecting an information provider from a plurality of information providers when the prompt of the user is a prompt for providing content of the information provider or when content of the information provider may be provided in associated with the prompt by the user; selecting at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results; wherein the content of the information provider is dynamically generated by using one or more assets pre-registered by the information provider.
However, Tunstall further discloses:
selecting an information provider from a plurality of information providers when the prompt of the user is a prompt for providing content of the information provider or when content of the information provider may be provided in associated with the prompt by the user (i.e. matching information providers to the user’s prompt) (Tunstall: ¶ [0554] “Once the semantics of the statements in the documents are captured, reasoning can be effected to produce an optimal mutual match. LLMs can be used as a substitute or alternative method for doing this matching, as described herein. For the recruitment application the agents are job seekers and hirers. For other applications different agents are juxtaposed to detect an optimal match grounded in knowledge. The aim might be to produce a mutual match or for one agent to find a set of best matched agents. Some examples include people matching ( dating, friendship, professional relationships, companions, employment, personal services), people-product matching (e.g. advertisement recommendation, product selection support, product recommendation), people-business matching (recruitment, evaluating applicants for government services, credit scoring, matching insurance criteria, matching tax criteria, test group selection, university application) and business to business matching ( e.g. partnership, customers, merger or acquisition potential).”);
selecting at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results (i.e. selecting question to be provided based on the plurality of question and the solutions/results generated) (Tunstall: ¶¶ [0396] [0397] “In a preferred example, an oflline process runs questions using a very high reasoning effort and stores the resulting thinking results in the manner already described. The metadata is stored in a distributed in-memory cache; we store that the question has been processed, along with when, and the number of solutions generated. Subqueries generated while reasoning are also added to the distributed cache…The oflline process may run continuously, choosing questions to process based on their hit count ( see below) and how long ago we last processed that question. Questions with low hit counts that were processed some time ago are removed from the cache--or where there is evidence that the results may have expired.”); and
wherein the content of the information provider is dynamically generated by using one or more assets pre-registered by the information provider (i.e. content includes results based on selected prompt/question and a pre-registered asset or registered product of the provider) (Tunstall: ¶ [1036] “LLMs have been trained on extremely large amounts of data and have often partially or fully memorized much of that data being able to faithfully reproduce significant sections of it given the right prompt. This data often contains creative works where the copyright owner may object if the LLM were to reproduce that work without credit or payment. Products based on LLMs that can generate such material thus risk legal liability, litigation and the requirement to pay damages and royalties to possibly many hundreds of thousands of copyright owners.” Furthermore, as cited in ¶ [1056] “Source of the text could include bibliographic references for papers and books along with potentially page numbers; URLs for web pages and other appropriate references for other sources.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Tunstall’s selecting an information provider from a plurality of information providers when the prompt of the user is a prompt for providing content of the information provider or when content of the information provider may be provided in associated with the prompt by the user; selecting at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results; wherein the content of the information provider is dynamically generated by using one or more assets pre-registered by the information provider to Batina’s providing, by the at least one processor, the created question as a response to the prompt of the user with the created LLM results. One of ordinary skill in the art would have been motivated to do so in order to “return almost instantly with high quality results.” (Tunstall: ¶ [0399]).
Batina and Tunstall do not explicitly disclose wherein the correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets […].
However, Muriqi further discloses wherein the correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets […] (i.e. correlation is based on a number of keywords in the presence of ads, wherein the keywords correspond to the number of products offered by a provider over other competitors, correlation is also based on suitability of the conversation/query context with recommended ) (Muriqi: ¶¶ [0148] [0149] “A correlation index indicating the likelihood that users interested in one of the keywords will also be interested in the other keyword, is determined. The computed degree, the determined ratio and the correlation index can be processed to determine a percentage of cooccurrence for each keyword. The percentage of co-occurrence for each keyword is used to determine a correlation ratio, which indicates how often a co-occurring keyword is present when another co-occurring keyword is present, as compared to how often it occurs on its own. This information is used in processing keywords in queries to identify matching keywords. The matching keywords can be used to search products, services or Internet sites to generate recommendations…Term frequency-inverse-document frequency (tfidf) weighing measures can be used to determine how important an identified keyword is to a subject user profile in a collection or corpus of profiles. The importance of the identified keyword can increase proportionally to the number of times it appears in the document, offset by the frequency the identified keyword occurs in the corpus.” Furthermore, as cited in ¶ [0146] “A low valued user, i.e., one who is anticipated to produce low returns for advertisers may therefore see a greater number of ads, or be provided with lesser valued content. On the other hand, a higher valued user, may see a fewer number of highly targeted, high value ads, and have access to premium content. For example, a user who is anticipated to purchase a car or jewelry may receive significant subsidies from competing providers of those products, to the exclusion of other advertising of lesser anticipated value. Similarly, after a transaction, the seller may continue to subsidize use of the network by a valued user, with only after-the-transaction appropriate communications. In the market for attention, one ad sponsor may even bid to displace the competitor's ad, even if the sponsor's ad is not itself displayed.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Muriqi’s correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets to Batina’s correlation is computed based on the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates. One of ordinary skill in the art would have been motivated to do so in order “to "optimize" utility of the experience of use of the network, and part of that optimization is selection of ads and other sponsored content for users based on both the objective value of the ad placement, e.g., in terms of driving sales, transactions, or sentiment according to the sponsor's goals, and subjective value to the recipient according to that recipients need or desire for information or content.” (Muriqi: ¶ [0146]).
With respect to Claims 10 and 11:
All limitations as recited have been analyzed and rejected to claim 1. Claim 10 recites “A non-transitory computer-readable recording medium storing a computer program for executing the question recommendation method of claim 1 on a computer device” (Batina: ¶ [0025]). Claim 11 recites “A computer device for providing question recommendations, comprising: at least one processor configured to execute computer instructions stored in a memory, wherein the at least one processor is configured to,” (Batina: ¶ [0024]) perform the steps of method claim 1. Claims 10 and 11 do not teach or define any new limitations beyond claim 1. Therefore they are rejected under the same rationale.
With respect to Claim 5:
Batina teaches:
The question recommendation method of claim 1, wherein the providing of the selected at least one question as the response to the prompt of the user comprises providing search results that include the LLM results and the selected at least one question through a search service that receives the prompt of the user as input (i.e. providing search results which include the LLM results and the created question as a prompt that receives an input/selection from user) (Batina: ¶ [0040] “In some embodiments, additional context data received from a user may be processed to generate follow-up questions for the user to provide a chat-guided search experience. The additional context may comprise, for example, a user selection of one or more objects (e.g., products in an e-commerce context), indicators thereof, and/or images or media thereof. The system may process the user selection (or several such selections, e.g., from several rounds of a chat involving a user) to determine one or more object attributes that may be relevant to the search. The object attributes and/or the additional context may then be: inputted into an LLM to generate further follow-up prompts/questions for the user; used to generate collections (e.g., of one or more objects, indicators, images, and/or media) for display and/or for prompting further user selections; and/or otherwise used as input for a further search ( e.g., as further enhancement text for a vector search). For clarity, the generated collections may themselves be the result of a further search (e.g., a vector search carried out using enhancement text that is based on the object attributes).” Furthermore, as cited in ¶ [0043] “When a search is initiated by a user, the chatbot presents questions to the user in order to progressively gather relevant context for the search. An LLM may be employed for generating the questions and processing user responses thereto as part of gathering context. Upon receiving a user response to a question, the set of images presented in the image canvas may be updated. The user can select an image of a product to indicate their interest and/or preference for the product. The user selection serves to refine the product search. In this way, both the user's chat data and image selections are used to reduce the set of search results for a user's search query.”).
With respect to Claim 6:
Batina teaches:
The question recommendation method of claim 1, wherein the providing of the selected at least one question as the response to the prompt of the user comprises providing an answer that includes the LLM results and the selected at least one question as an answer to artificial intelligence through a conversation session between the user and an artificial intelligence module based on the LLM (i.e. providing the follow up question as a response to the prompt and providing the follow up question and the results through a conversation session between the user and chatbot) (Batina: ¶ [0038] “The described technique of enhancing user input using an LLM prior to a vector search may find application in various different contexts. In an example implementation, a chatbot application that is configured for performing various types of searches may use an LLM for enhancing user input to the chatbot. The chatbot may initially ask a user for text input (e.g., the user's product of interest). Upon receiving input of the user's initial response (e.g., "cowboy hat"), the chatbot may present follow-up questions for the user ( e.g., when does the user want to wear the cowboy hat). The user may provide further responses to the questions (e.g., "in the summer'). The LLM may infer additional properties of a product that the user is searching for ( e.g., lightweight, summer-suitable, etc.) based on the user's responses to the chatbot prompts. The additional properties are captured using vector embeddings and then used in a vector search. The vector search may identify products that are similar to, or otherwise related to, the user's desired product(s). The LLM may be trained on a plurality of different search spaces, including the vector space. In particular, the enhancement text may be generated based on text data available in multiple search spaces.” Furthermore, as cited in ¶ [0043] “When a search is initiated by a user, the chatbot presents questions to the user in order to progressively gather relevant context for the search. An LLM may be employed for generating the questions and processing user responses thereto as part of gathering context. Upon receiving a user response to a question, the set of images presented in the image canvas may be updated. The user can select an image of a product to indicate their interest and/or preference for the product. The user selection serves to refine the product search. In this way, both the user's chat data and image selections are used to reduce the set of search results for a user's search query.”).
With respect to Claim 7:
Batina teaches:
The question recommendation method of claim 1, further comprising: dynamically creating, by the at least one processor, content of the information provider related to the selected at least one question and providing the same to the user when the selected at least one question provided as the response is selected by the user (i.e. dynamically creating content related to the question and providing the content to user according to answers to the follow up question) (Batina: ¶¶ [0043] [0044] “When a search is initiated by a user, the chatbot presents questions to the user in order to progressively gather relevant context for the search. An LLM may be employed for generating the questions and processing user responses thereto as part of gathering context. Upon receiving a user response to a question, the set of images presented in the image canvas may be updated. The user can select an image of a product to indicate their interest and/or preference for the product. The user selection serves to refine the product search. In this way, both the user's chat data and image selections are used to reduce the set of search results for a user's search query…The image canvas may be dynamically updated based on user interaction with the chatbot and the product images. In particular, the chat data and image selections may facilitate identification of relevant product attributes for the search. When a user selects an image, a search may be conducted to identify similar items. The system may determine which attributes are unique, uncommon, or otherwise pertinent ( e.g., based on the similar items, and/or for the relevant product category) about the user's selection. Additionally, or alternatively, the system may determine which attributes are common (e.g., based on the similar items and/or for the relevant product category) or otherwise of low discriminating power, in order to, e.g., remove such attributes from further consideration. Additionally, or alternatively, the system may determine attributes to include or remove from consideration based on the output of, e.g., a trained ML model, a rules-based system involving natural language processing, an LLM, and/or a system otherwise capable of determining a relevant subset of object attributes based on an object. The attributes are then used for dynamically generating questions to ask the user ( or otherwise prompt the user with the possibility of making a further selection) and/or to update the image canvas.”).
With respect to Claim 15:
All limitations as recited have been analyzed and rejected to claim 7. Claim 15 does not teach or define any new limitations beyond claim 7. Therefore it is rejected under the same rationale.
With respect to Claim 8:
Batina does not explicitly disclose the question recommendation method of claim 7, wherein an instance for the content is dynamically created by using LLM results created based on the large language model for the selected at least one question and a pre-registered asset of the information provider.
However, Tunstall further discloses wherein an instance for the content is dynamically created by using LLM results created based on the large language model for the selected at least one question and a pre-registered asset of the information provider (i.e. content includes results based on selected prompt/question and a pre-registered asset or registered product of the provider) (Tunstall: ¶ [1036] “LLMs have been trained on extremely large amounts of data and have often partially or fully memorized much of that data being able to faithfully reproduce significant sections of it given the right prompt. This data often contains creative works where the copyright owner may object if the LLM were to reproduce that work without credit or payment. Products based on LLMs that can generate such material thus risk legal liability, litigation and the requirement to pay damages and royalties to possibly many hundreds of thousands of copyright owners.” Furthermore, as cited in ¶ [1056] “Source of the text could include bibliographic references for papers and books along with potentially page numbers; URLs for web pages and other appropriate references for other sources.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Tunstall’s content is dynamically created by using LLM results created based on the large language model for the selected question and a pre-registered asset of the information provider to Batina’s providing, by the at least one processor, the created question as a response to the prompt of the user with the created LLM results. One of ordinary skill in the art would have been motivated to do so in order to “return almost instantly with high quality results.” (Tunstall: ¶ [0399]).
With respect to Claim 9:
Batina does not explicitly disclose the question recommendation method of claim 8, wherein the instance for the content is dynamically created by further using at least one of the question, a pre-registered prompt of the information provider, and information on the user.
However, Tunstall further discloses wherein the instance for the content is dynamically created by further using at least one of the question, a pre-registered prompt of the information provider, and information on the user (i.e. content is created based on prompt/question, pre-registered prompt of provider such as copyright status, and information on the user) (Tunstall: ¶¶ [1036]-[1038] “LLMs have been trained on extremely large amounts of data and have often partially or fully memorized much of that data being able to faithfully reproduce significant sections of it given the right prompt. This data often contains creative works where the copyright owner may object if the LLM were to reproduce that work without credit or payment. Products based on LLMs that can generate such material thus risk legal liability, litigation and the requirement to pay damages and royalties to possibly many hundreds of thousands of copyright owners…Some prompts are more prone to this risk than others. For example, a factual question where the answer is widely available from many sources may be of lower risk than a request to write a poem or song where the answer is very similar to one that is already published…A method to reduce the chances of this happening is to maintain a separate database of text which the LLM has been trained on. In some cases this will identify text where the copyright status or source makes the risks of reproducing it especially significant.” Furthermore, as cited in ¶ [0588] “The UL representing information about the user could come from partially or fully translating information contained in the user's social media profile, postings, profile information, "likes" and similar. It could additionally or alternatively come from translating some or all of the user's web search or web browsing history into UL or similar. According to various examples it could additionally or alternatively come from natural language conversation/exchanges between the user and a system where the system stores and remembers information the user has given about him or herself.”).
Therefore, it would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Tunstall’s content is dynamically created by further using at least one of the question, a pre-registered prompt of the information provider, and information on the user to Batina’s providing, by the at least one processor, the created question as a response to the prompt of the user with the created LLM results. One of ordinary skill in the art would have been motivated to do so in order to “return almost instantly with high quality results.” (Tunstall: ¶ [0399]).
With respect to Claim 13:
Batina teaches:
The computer device of claim 11, wherein the correlation between the plurality of question candidates and the LLM results is also computed based on at least one of presence or absence or the number of content of the information provider providable by each of the plurality of question candidates, and quality or prospective charge of content of the information provider providable by each of the plurality of question candidates (i.e. correlation between further questions/prompt and results is based on number of time content has been output, quality of content or target values associated with results, suitability with respect to conversation context, and suitability with respect to user information) (Batina: ¶ [0075] “The API call may include an API key to enable the computing system to be identified by the remote system. The API call may also include an identification of the language model or LLM to be accessed and/or parameters for adjusting outputs generated by the language model or LLM, such as, for example, one or more of a temperature parameter (which may control the amount of randomness or "creativity" of the generated output) (and/or, more generally some form of random seed as serves to introduce variability or variety into the output of the LLM), a minimum length of the output (e.g., a minimum of 10 tokens) and/or a maximum length of the output (e.g., a maximum of 1000 tokens), a frequency penalty parameter (e.g., a parameter which may lower the likelihood of subsequently outputting a word based on the number of times that word has already been output), a "best of' parameter (e.g., a parameter to control the number of times the model will use to generate output after being instructed to, e.g., produce several outputs based on slightly varied inputs).” Furthermore, as cited in ¶¶ [0050] [0051] “If the training data is labeled, the desired target values may be, e.g., the ground truth labels of the training data. If the training data is unlabeled, the desired target value may be a reconstructed ( or otherwise processed) version of the corresponding ML model input ( e.g., in the case of an autoencoder), or may be a measure of some target observable effect on the environment (e.g., in the case of a reinforcement learning agent). The parameters of the ML model are updated based on a difference between the generated output value and the desired target value. For example, if the value outputted by the ML model is excessively high, the parameters may be adjusted so as to lower the output value in future training iterations. An objective function is a way to quantitatively represent how close the output value is to the target value. An objective function represents a quantity (or one or more quantities) to be optimized ( e.g., minimize a loss or maximize a reward) in order to bring the output value as close to the target value as possible. The goal of training the ML model typically is to minimize a loss function or maximize a reward function…The training data may be a subset of a larger data set. For example, a data set may be split into three mutually exclusive subsets: a training set, a validation (or cross-validation) set, and a testing set. The three subsets of data may be used sequentially during ML model training. For example, the training set may be first used to train one or more ML models, each ML model, e.g., having a particular architecture, having a particular training procedure, being describable by a set of model hyperparameters, and/or otherwise being varied from the other of the one or more ML models. The validation ( or cross-validation) set may then be used as input data into the trained ML models to, e.g., measure the performance of the trained ML models and/or compare performance between them. Where hyperparameters are used, a new set of hyperparameters may be determined based on the measured performance of one or more of the trained ML models, and the first step of training (i.e., with the training set) may begin again on a different ML model described by the new set of determined hyperparameters. In this way, these steps may be repeated to produce a more performant trained ML model. Once such a trained ML model is obtained ( e.g., after the hyperparameters have been adjusted to achieve a desired level of performance), a third step of collecting the output generated by the trained ML model applied to the third subset (the testing set) may begin. The output generated from the testing set may be compared with the corresponding desired target values to give a final assessment of the trained ML model's accuracy.” Furthermore, as cited in ¶¶ [0104] [0105] “In at least some implementations, the LLM may be instructed to generate follow-up questions that are relevant to the search query. In particular, the LLM may be prompted to use the search query and user responses to identify one or more preferred object attributes indicating attributes of interest or preference for the user. The computing system may provide the user-inputted query as input to the LLM, and the output/result of the LLM ( e.g., question prompts) may be presented as chat-outputs in a conversation between the user and the chatbot…The computing system then performs a vector search using the search query. For example, a search module, such as the search engine 114 of FIG. 1, associated with the computing system may initiate a vector search using a text (e.g., word, sentence, etc.) embedding of the user-inputted query. In at least some implementations, the preferred object attributes may be used in the vector search. For example, said attributes (or features) may be given greater weight when performing the vector search.” Furthermore, as cited in ¶ [0037] “In some implementations, the LLM may be provided additional context for generating the enhancement text. For example, when the user input comprises text input for a chatbot, the LLM may be provided with historical chat log data for the user. As another example, when the search is a product search ( or similar search on an e-commerce platform), user profile information such as purchase history, browsing history, and the like, may be provided to the LLM along with user input of a search query. The LLM may then generate the input enhancement data based on the user input and relevant context, and a vector embedding may be created using the original user input and/or the input enhancement data.”).
Response to Arguments
The Applicant's arguments see pages 7-10 of the Remarks disclosed, filed on 04/10/2026, with respect to the 35 U.S.C. § 101 rejection(s) of claim(s) 1, 3-11, and 13-17 have been considered. The Applicant asserts “The claimed method therefore improves the functioning of the computer by enhancing the LLM results provided to a user based on a prompt by the user, where a processor selects at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results, where "the correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates" and "at least one of the suitability with the conversation context and the suitability with the user characteristics includes at least one of a first correlation that is computed based on a degree of overlapping between texts and a topic matching status, and a second correlation that is computed by analyzing a sentence structure, vocabulary, and a relationship between sentences using natural language processing technology." Thus, the steps of the claimed method of calculating a correlation by synthesizing the conversation through natural language processing and an objective asset data (the number of assets and/or presence of multimedia, is a concrete technological solution that improves the LLM results provided to a user, rather than a result-oriented abstraction, where the claimed method demonstrates that it changes how the computer operates at a foundational level.” The Examiner respectfully disagrees. There is no “synthesizing the conversation through natural language processing and an objective asset data” recitation being done by the claims. The claims merely recite selecting a question to be provided to the user based on a correlations between the question candidates and LLM results with the amendments further clarify that the correlations are computed based on three elements; 1) number of assets pre-registered by information provider, 2) presence of multimedia within the assets, and 3) at least one of suitability with conversation context and/or user characteristics of the question candidates. The correlations is being described at a very high level that further describes the abstract idea. Furthermore, Claims 1, 10, and 11 recite limitations directed to the abstract idea including “creating LLM results from a prompt of a user; selecting an information provider from a plurality of information providers when the prompt of the user is a prompt for providing content of the information provider or when content of the information provider may be provided in associated with the prompt by the user; creating a prompt for creation of a question related to the LLM results and the content of the information provider and creating a plurality of question candidates by inputting the created prompt; selecting at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results; providing the created selected at least one question as a response to the prompt of the user with the created LLM results; and providing additional content to the user based on an answer to the selected at least one question provided by the user, wherein the content of the information provider is dynamically generated by using one or more assets pre-registered by the information provider, and wherein the correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates.” These further limitations are not seen as any more than the judicial exception. Creating results based on a model (i.e. LLM results based on a large language model), creating a question based on the results, and providing the created question alongside results is considered to be an abstract idea under mental processes because the claims are directed to concepts performed in the human mind (including an observation, evaluation, judgment, opinion). Creating results based on a model (i.e. LLM results based on a large language model), creating a question based on the results, and providing the created question alongside results is also considered to be an abstract idea under certain methods of organizing human activity because the claims are directed to commercial or legal interactions (including agreements in the form of contracts; legal obligations; advertising, marketing or sales activities or behaviors; business relations) and managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions). Claims 1, 10, and 11 recite additional limitations including “by the at least one processor, and based on a / to the large language model (LLM).” Therefore, the rejection(s) of claim(s) 1, 5-11, 13, and 15 under 35 U.S.C. § 101 is maintained above with an updated analysis.
The Applicant's arguments see pages 10-14 of the Remarks disclosed, filed on 04/10/2026, with respect to the 35 U.S.C. § 103 rejection(s) of claim(s) 1, 3-11, and 13-15 over Batina in view of Tunstall and Muriqi with claims 16-17 being rejected in further view of Temraz have been considered. The Applicant asserts “Batina discloses a user interface for chat guided searches where a user inputs a search query to the user interface and a computer provides an initial set of images to the user based on the search query. (Paragraph [0105] to [0106]). The user selects one of the images and then the computer performs a further similarity search based on the selected image. (Paragraph [0107]). The computer the provides another set of images based on the similarity search. (Paragraph [0108]; Fig. 3). Batina does not disclose a processor configured to "select at least one question to be provided to the user based on a correlation between the plurality of question candidates and the LLM results" where "the correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates" and "at least one of the suitability with the conversation context and the suitability with the user characteristics includes at least one of a first correlation that is computed based on a degree of overlapping between texts and a topic matching status, and a second correlation that is computed by analyzing a sentence structure, vocabulary, and a relationship between sentences using natural language processing technology" as now recited in amended claims 1 and 11…On the other hand, the contents of Muriqi cited by the Examiner simply relate to a text mining technique that calculates correlation by counting 'how many times a specific word appeared (keyword frequency)' within the user's text or search history. In other words, the claimed method utilizes 'the quantity and attributes of pre-registered marketing assets' as a correlation evaluation index that cannot be deemed obvious based on Muriqi, which describes 'keyword appearance frequency.' This is clearly different from the correlation calculation of the claimed method. Furthermore, by limiting 'the suitability with the conversation context and the user characteristics (including analysis of sentence structure, vocabulary, and degree of text overlapping based on natural language processing)' together with the number of assets and the presence of multimedia as an integrated criterion for calculating the correlation, as recited in amended claims 1 and 11. Specifically, the cited references merely disclose an image similarity search ( Batina), controlling the source of generated text (Tunstall), or word frequency analysis within text (Muriqi and Temraz)." The Examiner would like to refer the Applicant to ¶ [0040] of the Batina reference; The object attributes and/or the additional context may then be: inputted into an LLM to generate further follow-up prompts/questions for the user; used to generate collections (e.g., of one or more objects, indicators, images, and/or media) for display and/or for prompting further user selections; and/or otherwise used as input for a further search ( e.g., as further enhancement text for a vector search). For clarity, the generated collections may themselves be the result of a further search (e.g., a vector search carried out using enhancement text that is based on the object attributes).” With the correlations computed based on the three elements; 1) [[number of]] assets pre-registered by information provider, 2) presence of multimedia within the assets, and 3) at least one of suitability with conversation context and/or user characteristics of the question candidates also being disclosed by the Batina reference in at least ¶¶ [0035] [0036] [0045] [0063] [0082] (See pages 10-11 above). The Examiner would like to note that the Batina reference does not explicitly disclose that the correlations are computed based on a number of assets pre-registered by information provider (emphasis added). Muriqi reference was brought in to explicitly disclose correlations computed based on a number of assets pre-registered by information provider (See ¶¶ [0146] [0148] [0149] of the Muriqi reference and pages 14-15 above). It would have been obvious to one of ordinary skill in the art, at the time the invention was made, to add Muriqi’s correlation is computed based on a number of the assets pre-registered by the information provider, a presence of multimedia content within the assets to Batina’s correlation is computed based on the assets pre-registered by the information provider, a presence of multimedia content within the assets, and at least one of suitability with a conversation context of each of the plurality of question candidates and suitability with user characteristics of each of the plurality of question candidates. One of ordinary skill in the art would have been motivated to do so in order “to "optimize" utility of the experience of use of the network, and part of that optimization is selection of ads and other sponsored content for users based on both the objective value of the ad placement, e.g., in terms of driving sales, transactions, or sentiment according to the sponsor's goals, and subjective value to the recipient according to that recipients need or desire for information or content.” (Muriqi: ¶ [0146]). Furthermore, the Examiner would also like to note that one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413,208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091,231 USPQ 375 (Fed. Cir. 1986). Therefore, the rejection(s) of claim(s) 1, 5-11, 13, and 15 under 35 U.S.C. § 103(a) is provided above with updated citations.
Conclusion
The prior art made of record and not relied upon is considered pertinent to Applicant’s disclosure. The following reference are cited to further show the state of the art:
U.S. Publication 2024/0330597 to Temraz for disclosing generating data related to chatbot performance may include receiving a plurality of conversations between a chatbot and chatbot users, providing the plurality of conversations to a machine learning model trained to identify patterns based on the plurality of conversations, identifying patterns within the plurality of conversations based on an output of the machine learning model, displaying, using a graphical user interface (GUI), the identified patterns, and retraining the chatbot based on the identified patterns.
U.S. Publication 2024/0370658 to Zhang for disclosing receiving, at an interface between a user and a large language model, an original natural language prompt including a plurality of facts from the user. The implementations further include generating a series of contextual sub-questions based on the original natural language prompt using the large language model. Additionally, the implementations further include providing the contextual sub-questions to the large language model to obtain contextual answers. Additionally, the implementations further include applying the contextual sub-questions against the original natural language prompt with the contextual answers as a refined natural language prompt to the large language model in a reverse order of the series. Additionally, the implementations further include outputting, to the user, a final answer from the large language model to a terminal state of the refined natural language prompt.
U.S. Publication 2016/0140216 to Allen for disclosing adjusting fact-based answers provided by a question/answer (QA) system. A user submits a question to the QA system, where it is categorized into a question type. The QA system then processes the question to generate an answer. The QA system then generates an answer adjustment if it is determined that the question type and answer meet a predicted undesirable outcome. The answer adjustment may include a warning, a disclaimer, a recommendation, an alternative fact-based answer, a referral to an assistance service, or any combination thereof.
U.S. Patent 11,514,330 to Ma for disclosing Methods and systems are provided for a natural language processing system comprising a chatbot adapted for dialog generation. In one example, the system may include a combination of a variational autoencoder (VAE) and a generative adversarial network (GAN) for generating natural responses to input queries. The VAE may convert queries into vector embeddings that may then be used by the GAN to continuously update and improve responses provided by the chatbot.
U.S. Publication 2022/0198154 to Liu for disclosing An intelligent question answering method includes: determining, based on received question information, a target object and a target attribute corresponding to the question information; obtaining an answer knowledge path and an external knowledge path of the target object other than the answer knowledge path from a pre-established knowledge graph based on the target object and the target attribute, the answer knowledge path including target context information for describing the target attribute, and the external knowledge path including external context information for describing another attribute; inputting the answer knowledge path and the external knowledge path into a trained neural network model to obtain a reply text, a training corpus of the neural network model during training including at least comment information of the target object; and outputting the reply text.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Azam Ansari, whose telephone number is (571) 272-7047. The examiner can normally be reached from Monday to Friday between 8 AM and 4:30 PM.
If any attempt to reach the examiner by telephone is unsuccessful, the examiner's supervisor, Waseem Ashraf, can be reached at (571) 270-3948.
Another resource that is available to applicants is the Patent Application Information Retrieval (PAIR). Information regarding the status of an application can be obtained from the (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAX. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pairdirect.uspto.gov. Should you have questions on access to the Private PAIR system, please feel free to contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
Applicants are invited to contact the Office to schedule either an in-person or a telephonic interview to discuss and resolve the issues set forth in this Office Action. Although an interview is not required, the Office believes that an interview can be of use to resolve any issues related to a patent application in an efficient and prompt manner.
/AZAM A ANSARI/
Primary Examiner, Art Unit 3621
May 30, 2026
Prosecution Insights