DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim 20 is canceled.
Claim 21 is newly added.
Claims 1-19 and 21 are presented for examination.
The claims and only the claims form the metes and bounds of the invention. “Office personnel are to give claims their broadest reasonable interpretation in light of the supporting disclosure. In re Morris, 127 F.3d 1048, 1054-55, 44 USPQ2d 1023, 1027-28 (Fed. Cir. 1997). Limitations appearing in the specification but not recited in the claim are not read into the claim. In re Prater, 415 F.2d 1393, 1404-05, 162 USPQ 541, 550-551 (CCPA 1969)” (MPEP p 2100-8, c 2, I 45-48; p 2100-9, c 1, l 1-4). The Examiner has full latitude to interpret each claim in the broadest reasonable sense. The Examiner will reference prior art using terminology familiar to one of ordinary skill in the art. Such an approach is broad in concept and can be either explicit or implicit in meaning.
Response to Arguments
Applicant’s remarks/amendment was filed on 29 May 2026.
NPL “LLMs Generate Structurally Realistic Social Networks but Overestimate Political Homophily” by Chang et al. and NPL “ChatHaruhi: Reviving Anime Character in Reality via Large Language Model” by Li et al. are newly introduced to address the latest claim amendment.
Applicant's arguments have been considered but they are moot in view of new
ground(s) of rejection. However, the Examiner welcomes any suggestion(s) Applicant
may have on moving prosecution forward.
Claim Objections
The objections raised in the Office Action mailed on 03/04/2026 have been overcome by applicant’s amendment received on 05/29/2026.
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, 5, 7-8, 12, 14-15, 19 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over NPL “LLMs Generate Structurally Realistic Social Networks but Overestimate Political Homophily” (29 Aug 2024) by Chang et al. (“Chang”) in view of NPL “ChatHaruhi: Reviving Anime Character in Reality via Large Language Model” (18 Aug 2023) by Li et al. (“Li”).
As to Claim 1, Chang teaches a method of retrieval augmented generation, the method comprising:
generating, via a language model, a social graph including a plurality of synthetic personas represented by nodes of the social graph (Chang: at least 1 Introduction; “generating social networks with large language models (LLMs) has the potential to address these limitations. LLMs possess zero-shot capabilities, enabling network generation without training. LLMs can also generate networks in a flexible manner, based on natural language descriptions of each person in the network”; “Figure 1: Our three prompting methods to generate social networks with LLMs”; 4 Comparison to Real Networks further disclose “N is the number of nodes and E is the number of edges in the network” and Figure 2 shows “Generated social networks from different prompting methods: Global (top), Local (middle), Sequential (bottom)”; Model costs and comprehension further discloses “N (the number of nodes in the network), E (the number of edges), D (the number of tokens to describe a persona’s demographics), and I (the number of tokens to describe a persona’s interests)”) and relationships between the synthetic personas represented by edges of the social graph (Chang: at least 4 Comparison to Real Networks; “N is the number of nodes and E is the number of edges in the network” and “for a node i, its clustering coefficient is 2Ei/Ni(Ni−1) , where Ni is its number of neighbors and Ei is the number of edges between its neighbors”; C.3 Evaluating network structure further discloses “… friendship ratings between university freshmen. The edge weights range from -1 (risk of getting into conflict) to +3 (best friend)”), based on a first preconfigured prompt that causes the language model to generate the edges based on the plurality of synthetic personas (Chang: at least 3.2 Network Generation; “Global. In our first method, which we call “Global”, we provide the LLM with the entire list of personas, and prompt it to construct the network between them, in the form of edge pairs” and “in our second method, which we call “Local”, we have the LLM take on the identity of one persona at a time, e.g., by saying, “You are a Woman, Asian, age 54, Hindu, Independent”);
generating, via the language model, a profile for at least one of the plurality of synthetic personas based on a second preconfigured prompt that causes the language model to generate (Chang: at least Figure 1; “our three prompting methods to generate social networks with LLMs”; 4 Comparison to Real Networks further disclose “Figure 2: Generated social networks from different prompting methods: Global (top), Local (middle), Sequential (bottom)”; 3.2 Network Generation further discloses “Global. In our first method, which we call “Global”, we provide the LLM with the entire list of personas, and prompt it to construct the network between them, in the form of edge pairs” and “in our second method, which we call “Local”, we have the LLM take on the identity of one persona at a time, e.g., by saying, “You are a Woman, Asian, age 54, Hindu, Independent”; note: entry of personas as a prompt), random characteristics based on initial description and positions of the plurality of synthetic personas in the social graph (Chang: at least Figure 1; “1. Woman, Asian, 54, Hindu, Independent”, “2. Man, White, 18, Protestant, Republican”, 3. Woman, Black, 39, Unreligious, Democrat”; 3.2 Network generation further discloses “we provide the LLM with the list of all other personas (in the same format) and prompt it to pick friends for the persona it is currently assigned”; B. Additional Results further discloses “LLM can match many structural characteristics of real social networks, including density, clustering, connectivity, and degree distribution (Figures 3 and 4)”; 4 Comparison to Real Networks further discloses “Social networks are known to exhibit clustering, where one’s friends are likely to be friends with each other”; Model costs and comprehension further discloses “N (the number of nodes in the network), E (the number of edges), D (the number of tokens to describe a persona’s demographics), and I (the number of tokens to describe a persona’s interests)”).
Chang does not explicitly disclose, but Li discloses generating, via the language model, a plurality of synthetic documents relating to the plurality of synthetic personas based on a third preconfigured prompt that causes the language model to generate a text conversation (Li: at least Introduction; “we designed a system to automatically generate dialogues fitting the character’s personality, even for characters with fewer original dialogues” and “allowing language models to mimic the tone and knowledge of specific anime/TV characters during a conversation” and “we additionally simulated over 31,000 dialogues for these characters”; 2. Related Works further discloses our approach produces the subsequent dialogues responding to each question in-context”) between two synthetic personas of the plurality of synthetic personas based at least on one of the social graph or the profile (Li: at least Introduction; “I want you to act like {character} from {series}. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use” and “… play real characters from anime, TV or other works during a conversation” and “each virtual character exists in their own background” and “the personality of a character is also a very important part of anime, TV and even game works. The personality must remain consistent throughout the work” and “we attempt to better mimic the character by controlling the language model”; 2. Related Works further discloses “our work similarly leverages large conversational models like ChatGPT to automatically generate dialogues between agents”), wherein the text conversation indicates a first characteristic of the random characteristics (Li: at least Introduction; “we designed a system to automatically generate dialogues fitting the character’s personality, even for characters with fewer original dialogues”; 2. Related Works further discloses “our system incorporates substantial prompts about the character’s background, personality and prior conversations, in order to produce in-character dialogues”);
generating a training dataset including the text conversation labeled with the first characteristic (Li: at least Introduction; “using our proposed algorithm, aided by GPT-3 and GPT-4, we additionally simulated
over 31,000 dialogues for these characters. Combined, this forms the ChatHaruhi-54k dataset.” 5.1 Generate Dialogue from Question Introduction; “the language model can simulate generating the corresponding dialogue d’. The resulting d’ becomes fine-tuning data for the language model”; 6.3 Language Model Tuning futher discloses “with the complete 54K ChatHaruhi dataset, we can
fine-tune local language models”);
training a retrieval model to retrieve the text conversation in response to a user query regarding the first characteristic, using the training dataset (Li: at least 3 ChatBot Design; “I want you to act like {character} from {series}. You are now cosplay {character} If others’ questions are related with the novel, please try to reuse the original lines from the novel. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use. You must know all of the knowledge of {character}”); and
generating, by an AI agent, a response to a second user query based on one or more documents retrieved via the trained retrieval model (Li: at least 3 ChatBot Design; “I want you to act like {character} from {series}. You are now cosplay {character} If others’ questions are related with the novel, please try to reuse the original lines from the novel. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use. You must know all of the knowledge of {character}” and “in order to make the content of D(q,R) more relevant to the content of q, we use sentence embeddings here to search for the M most relevant Q&As from a larger memory bank U. Here U is the set of all sentences where other characters interact with R throughout the novel/ movie”; 3.3 Original Dialogue Searching further discloses “for a query q, we will use a sentence embedding model f() to extract embeddings f(d) for all d ∈ D. After similarly extracting f(q) for the query q, we extract the M samples from D that are closest (in cosine similarity) to f(q)”; 2 Related Work also discloses “given a character’s background and prior memories, our approach produces the subsequent dialogues responding to each question in-context”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Lis features of generating, via the language model, a plurality of synthetic documents relating to the plurality of synthetic personas based on a third preconfigured prompt that causes the language model to generate a text conversation (Li: at least Introduction; 2. Related Works) between two synthetic personas of the plurality of synthetic personas based at least on one of the social graph or the profile (Li: at least Introduction; 2. Related Works), wherein the text conversation indicates a first characteristic of the random characteristics (Li: at least Introduction; 2. Related Works);
generating a training dataset including the text conversation labeled with the first characteristic (Li: at least Introduction; 5.1 Generate Dialogue from Question; 6.3 Language Model Tuning);
training a retrieval model to retrieve the text conversation in response to a user query regarding the first characteristic, using the training dataset (Li: at least 3 ChatBot Design); and
generating, by an AI agent, a response to a second user query based on one or more documents retrieved via the trained retrieval model (Li: at least 3 ChatBot Design; 3.3 Original Dialogue Searching; 2 Related Work) with Chang’s method.
The suggestion/motivation for doing so would have been to get language models to implement role-playing that have “potential applications in many areas like games, creative industries, etc.” (Li: at least Introduction).
Claim 8 (a system claim) corresponds in scope to claim 1 and is similarly rejected.
Claim 15 (a medium claim) corresponds in scope to claim 1 and is similarly rejected.
As to Claim 5, Chang and Li teach the method of claim 1, wherein the plurality of synthetic documents includes documents of multiple types including at least two of: conversations between the synthetic personas (Li: at least Introduction; “we designed a system to automatically generate dialogues fitting the character’s personality, even for characters with fewer original dialogues” and “allowing language models to mimic the tone and knowledge of specific anime/TV characters during a conversation” and “we additionally simulated over 31,000 dialogues for these characters”; 2. Related Works further discloses “our work similarly leverages large conversational models like ChatGPT to automatically generate dialogues between agents”); persona-Al interactions (Li: at least 3 ChatBot Design; “I want you to act like {character} from {series}. You are now cosplay {character} If others’ questions are related with the novel, please try to reuse the original lines from the novel. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use. You must know all of the knowledge of {character}”; ); and interaction logs for a website.
Claim 12 (a system claim) corresponds in scope to claim 5 and is similarly rejected.
Claim 19 (a medium claim) corresponds in scope to claim 5 and is similarly rejected.
As to Claim 7, Chang and Li teach the method of claim 1, wherein the random characteristics are generated via the language model (Li: at least Introduction; “I want you to act like {character} from {series}. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use” and “… play real characters from anime, TV or other works during a conversation” and “each virtual character exists in their own background” and “the personality of a character is also a very important part of anime, TV and even game works. The personality must remain consistent throughout the work” and “we attempt to better mimic the character by controlling the language model” and “this algorithm can effectively organize a character’s memories, allowing language models to mimic the tone and knowledge of specific anime/TV characters during a conversation”).
Claim 14 (a system claim) corresponds in scope to claim 7 and is similarly rejected.
As to Claim 21, Chang and Li teach the method of claim 1, further comprising, generating, by the language model, the plurality of synthetic personas, wherein each of the synthetic personas includes a combination of a randomly sampled persona description and a name (Li: at least Introduction; “I want you to act like {character} from {series}. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use” and “… play real characters from anime, TV or other works during a conversation” and “each virtual character exists in their own background” and “the personality of a character is also a very important part of anime, TV and even game works. The personality must remain consistent throughout the work” and “we attempt to better mimic the character by controlling the language model” and “this algorithm can effectively organize a character’s memories, allowing language models to mimic the tone and knowledge of specific anime/TV characters during a conversation”; 4 Character Dataset Building further discloses “Haruhi Suzumiya”, “Li Yunlong”, “Sheldon, Penny and Raj”).
Claims 2, 9 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over NPL “LLMs Generate Structurally Realistic Social Networks but Overestimate Political Homophily” (29 Aug 2024) by Chang et al. (“Chang”) in view of NPL “ChatHaruhi: Reviving Anime Character in Reality via Large Language Model” (18 Aug 2023) by Li et al. (“Li”) as applied to claims 1, 5, 7-8, 12, 14-15, 19 and 21 above, and further in view of US PGPUB 2010/0146010 by Muller et al. (“Muller”).
As to Claim 2, Chang and Li teach the method of claim 1.
Chang and Li do not explicitly disclose, but Muller discloses the method further comprising: performing a heuristic check of the social graph to ensure consistency with one or more predetermined rules (Muller: at least ¶0026; “it can be determined whether or not the relationship specified by the tag is symmetrical or asymmetrical in nature”), wherein the one or more predetermined rules include at least a requirement that relationships be symmetric (Muller: at least ¶0026; “if the relationship is determined to be symmetrical in nature, in block 350 a tag can be constructed for the reciprocal resource that specifies the same relationship as the tag for the selected resource”); and updating the social graph in response to the heuristic check (Muller: at least ¶0026; “if the relationship is determined to be symmetrical in nature, in block 350 a tag can be constructed for the reciprocal resource that specifies the same relationship as the tag for the selected resource” and “if the relationship is determined to be asymmetrical in nature, in block 360 a tag can be constructed for the reciprocal resource that specifies the inverse of the relationship for the tag for the selected resource”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Muller’s features of the method further comprising: performing a heuristic check of the social graph to ensure consistency with one or more predetermined rules (Muller: at least ¶0026), wherein the one or more predetermined rules include at least a requirement that relationships be symmetric (Muller: at least ¶0026); and updating the social graph in response to the heuristic check (Muller: at least ¶0026) with the method taught by Chang and Li.
The suggestion/motivation for doing so would have been to implement “reciprocal tagging of resources in a computer communications network” (Muller: at least ¶0009).
Claim 9 (a system claim) corresponds in scope to claim 2 and is similarly rejected.
Claim 16 (a medium claim) corresponds in scope to claim 2 and is similarly rejected.
Claims 3-4, 10-11 and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over NPL “LLMs Generate Structurally Realistic Social Networks but Overestimate Political Homophily” (29 Aug 2024) by Chang et al. (“Chang”) in view of NPL “ChatHaruhi: Reviving Anime Character in Reality via Large Language Model” (18 Aug 2023) by Li et al. (“Li”) as applied to claims 1, 5, 7-8, 12, 14-15, 19 and 21 above, and further in view of US PGPUB 2021/0103626 by Jolly et al. (“Jolly”).
As to Claim 3, Chang and Li teach the method of claim 1.
Chang and Li do not explicitly disclose, but Jolly discloses the method further comprising: generating a second plurality of synthetic documents relating to the plurality of synthetic personas, wherein the second plurality of synthetic documents does not indicate any characteristic of the random characteristics (Jolly: at least ¶¶0002 & 0037; “collections of news source items” and “information associated with the first collection of news source items can be configured as the newsfeed item configured for display on the user device and the displayed newsfeed item information can include: an automatically generated text summary for at least some of the news source items in the first collection; and news source names and any assigned news source ratings for each of the news sources having one or more news source items in the first collection”; ¶0161 also discloses “when the news event or topic timeline is presented to a user on a user's screen, the user can scroll through the timeline to review the generated headlines associated with each grouped collection. The user can also select a grouped collection appearing on the timeline to allow review of one or more news source items in the collection in a report”; ¶0093 further discloses “user can incorporate her own rating of bias, skew, or viewpoint for a news source that can be presented to her”), wherein the training dataset further includes the second plurality of synthetic documents (Jolly: at least ¶0176; “extraction of information that can aggregate a large number of news source items describing with a news event or topic of interest where such news source items incorporate appropriate ratings can be useful in the generation of training sets that can be applied in such machine learning methods”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Jolly’s feature of the method further comprising: generating a second plurality of synthetic documents relating to the plurality of personas, wherein the second plurality of synthetic documents does not indicate any characteristic of the random characteristics (Jolly: at least ¶¶0002 & 0037, 0093, 0161), wherein the training dataset further includes the second plurality of synthetic documents (Jolly: at least ¶0176) with the method taught by Chang and Li.
The suggestion/motivation for doing so would have been to “generate information about news source items from a corpus of news sources, where identified news source items describe or are associated with a news event or topic of interest” based on “various considerations, such as news source characteristics (e.g., bias/skew/viewpoint rating, age, location of news source or event, etc.), time, and presence, absence, or frequency of news sources ratings in a collection of news source items describing the news event or topic” (Jolly: at least Abstract).
Claim 10 (a system claim) corresponds in scope to claim 3 and is similarly rejected.
Claim 17 (a medium claim) corresponds in scope to claim 3 and is similarly rejected.
As to Claim 4, Chang, Li and Jolly teach the method of claim 3, wherein the second plurality of synthetic documents is generated based on retrieved information from an information source including at least one of: news, or weather (Jolly: at least ¶0037; “information associated with the first collection of news source items can be configured as the newsfeed item configured for display on the user device and the displayed newsfeed item information can include: an automatically generated text summary for at least some of the news source items in the first collection; and news source names and any assigned news source ratings for each of the news sources having one or more news source items in the first collection”; ¶0161 also discloses “when the news event or topic timeline is presented to a user on a user's screen, the user can scroll through the timeline to review the generated headlines associated with each grouped collection. The user can also select a grouped collection appearing on the timeline to allow review of one or more news source items in the collection in a report”; ¶0093 further discloses “user can incorporate her own rating of bias, skew, or viewpoint for a news source that can be presented to her”).
Claim 11 (a system claim) corresponds in scope to claim 4 and is similarly rejected.
Claim 18 (a medium claim) corresponds in scope to claim 4 and is similarly rejected.
Claims 6 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over NPL “LLMs Generate Structurally Realistic Social Networks but Overestimate Political Homophily” (29 Aug 2024) by Chang et al. (“Chang”) in view of NPL “ChatHaruhi: Reviving Anime Character in Reality via Large Language Model” (18 Aug 2023) by Li et al. (“Li”) as applied to claims 1, 5, 7-8, 12, 14-15, 19 and 21 above, and further in view of US PGPUB 2024/0263956 by Vashisht.
As to Claim 6, Chang and Li teach the method of claim 1.
Chang and Li do not explicitly disclose, but Vashisht discloses wherein the training dataset further includes an indication of a location within the text conversation indicative of the first characteristic (Vashisht: at least ¶¶0016 & 0031; “monitor a conversation in a meeting and identify a relevant destination address to which a participant attending the meeting from a car is being instructed to drive” and “receives the audio and chat information for the meeting from collaboration module and detects destination locations mentioned or otherwise provided during the meeting (e.g. in audio, in chat messages, etc.)”; ¶¶0036 & 0043 further disclose “training set of audio signals for different words as input and known words as outputs, where the neural network attempts to produce the provided output (or word)” and “text (or chat messages) may be segmented for analysis into segments of any quantity of words or phrases (e.g., corresponding to the quantity of words or phrases of the machine learning training set, etc.)”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Vashisht’s feature of wherein the training dataset further includes an indication of a location within the text conversation indicative of the first characteristic (Vashisht: at least ¶¶0016, 0031, 0036 & 0043) with the method taught by Chang and Li.
The suggestion/motivation for doing so would have been to “automatically provides the captured destination location to a vehicle navigation system” so that the vehicle navigation system” can “initiate navigation to a provided destination location (e.g., geographic or street address, geographic coordinates, name or point of interest, etc.)” (Vashisht: at least ¶¶0013).
Claim 13 (a system claim) corresponds in scope to claim 6 and is similarly rejected.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the Examiner should be directed to Huen Wong whose telephone number is (571) 270-3426. The examiner can normally be reached on Monday - Friday (10:30AM EST - 6:30PM EST). If attempts to reach the examiner by telephone are unsuccessful, the Examiner's supervisor, Charles Rones can be reached on (571) 272-4085. The fax phone number for the organization where this application or proceeding is assigned is (571) 273-8300 for regular communications and after final communications.
Information regarding the status of an application may be obtained from thePatent Application Information Retrieval (PAIR) system. Status information forpublished 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. Shouldyou have questions on access to the Private PAIR system, contact the ElectronicBusiness Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from aUSPTO Customer Service Representative or access to the automated informationsystem, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/H .W./
Examiner, AU 2168
03 August 2026
/MAHESH H DWIVEDI/Primary Examiner, Art Unit 2168
Prosecution Insights