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 .
Status of Claims
This action is in reply to the amendment and remarks filed on September 23, 2026 in response to the non-final Office action mailed July 14, 2026.
Claims 1, 2, 4, and 5 are amended. Claim 3 is cancelled. Claim 6 is new. Claims 1, 2, and 4-6 are pending and are examined. Claims 1, 4, and 5 are independent.
The amendments to claims 1, 2, 4, and 5 and new claim 6 have been entered. Applicant identifies support at ¶¶[0030]-[0036], [0040]-[0042], and [0045]-[0050] of the specification as published (US 2026/0187742 A1). The amended limitations are supported at least at ¶¶[0033]-[0034] (first instruction text that includes the independent claim and instructs extraction of the characteristic part), ¶¶[0035]-[0038] and [0053] (second instruction text and the two investigation alternatives), ¶[0056] and original claim 2 (dependent-claim descriptions), and ¶[0039] (vectorization, similarity, and threshold of claim 6). No new matter is found.
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119(a)-(d) to Japanese Patent Application No. 2024-231079, filed December 26, 2024. Receipt is acknowledged of a certified copy of the foreign priority document, retrieved electronically from the Japan Patent Office through the priority document exchange program on January 8, 2026 as required by 37 CFR 1.55. The request in applicant’s remarks for acknowledgment of the priority claim is granted. This action uses December 26, 2024, as the effective filing date for purposes of qualifying prior art. Each reference applied below qualifies as prior art under that date and under the actual U.S. filing date of December 4, 2025.
Information Disclosure Statement
The information disclosure statement filed on December 4, 2025, was considered in the prior action.
Response to Amendment and Arguments
Foreign priority (Remarks, p. 6)
Applicant requests acknowledgment of the claim for foreign priority to JP 2024-231079 and of receipt of the certified copy. The request is granted. See the Priority section above. The prior action did not acknowledge the claim, and this action corrects the record.
Rejection under 35 U.S.C. 101 (Remarks, pp. 7-8)
Applicant’s arguments have been fully considered but are not persuasive. The rejection under 35 U.S.C. 101 is maintained for claims 1, 2, 4, and 5, as amended, and is applied to new claim 6. The rejection is restated below in full for the amended claims, and that statement supersedes the § 101 analysis of the prior action. The rejection of claim 3 is moot in view of its cancellation.
Applicant argues that the amended claims recite additional elements, namely the first and second generative-model processes, and that extracting the characteristic part first and then using it in the investigation “reduces the influence of such non-characteristic information and improves the accuracy of the investigation,” citing ¶¶[0022]-[0024] and [0049]-[0050]. Applicant concludes that the elements integrate the abstract idea into a practical application at Step 2A, Prong Two.
The examiner agrees that the amended claims recite additional elements beyond the abstract idea: the processor, the generative model, the steps of inputting text to the generative model, the steps of acquiring the model’s output, and the step of outputting the result. The examiner does not treat the use of the generative model as part of the mental process. See the August 4, 2025, memorandum from Deputy Commissioner for Patents Charles Kim, “Reminders on evaluating subject matter eligibility of claims under 35 U.S.C. 101” (cautioning against treating limitations that cannot practically be performed in the mind as mental processes). Those elements are evaluated as additional elements at Prong Two and Step 2B below.
The additional elements do not integrate the exception into a practical application, for three reasons.
First, the asserted improvement is an improvement in the accuracy of the analysis, not an improvement in the functioning of a computer or of the generative model. The specification identifies the problem as “noise” in the investigation: when the whole claim is input, the model may search for the preamble instead of the point of the disclosure, so “the accuracy of the investigation result is reduced” (¶¶[0006], [0026]). The solution is to give the model the characteristic part instead, so that “the noise included in the investigation result is also reduced, so that the accuracy of the investigation result is improved” (¶[0027]; see also ¶¶[0052]-[0053]). Focusing a claim comparison on the point of novelty rather than the preamble is how a person skilled in patent analysis performs the same comparison mentally. The improvement lies in the abstract idea itself, namely a better-focused claim-to-document comparison. “[A]n improvement in the abstract idea itself … is not an improvement in technology.” MPEP 2106.05(a)(II) (citing Trading Technologies Int’l, Inc. v. IBG LLC, 921 F.3d 1084, 1093-94 (Fed. Cir. 2019)); see also SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161, 1163 (Fed. Cir. 2018) (a claim to a new and better abstract idea is still a claim to an abstract idea). The claims do not change how the generative model is built, trained, or executed. The model is “a known generative AI” (¶[0022]), for example “any of the GPT series (e.g., GPT-3, GPT-4), BERT, or RoBERTa” (¶[0031]), and the claims use it as supplied, by giving it text prompts and reading its output.
Second, the claims recite the result rather than a particular way of achieving it. Claim 1 instructs the model to extract “a characteristic part” and then to investigate whether, or where, a related feature appears. The claim does not say how the characteristic part is identified or how the investigation is conducted. Those determinations are left entirely to the generative model, which is the tool. That is the type of functional, result-oriented claiming that the Federal Circuit has held ineligible. Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205 (Fed. Cir. 2025), cert. denied, No. 25-505 (U.S. Dec. 8, 2025) (claims that “do no more than claim the application of generic machine learning to new data environments, without disclosing improvements to the machine learning models to be applied, are patent ineligible”); Two-Way Media Ltd. v. Comcast Cable Commc’ns, LLC, 874 F.3d 1329, 1337 (Fed. Cir. 2017); Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1356 (Fed. Cir. 2016); MPEP 2106.05(f)(1). Here, the new data environment is patent claims and technical documents, and the model is generic.
Third, the facts differ from the precedents that found an improvement. In Ex parte Desjardins, Appeal No. 2024-000567 (PTAB Sept. 26, 2025) (Appeals Review Panel) (designated precedential Nov. 4, 2025); see also MPEP 2106.04(d), 2106.05(a), as revised by the USPTO’s Advance Notice of Change to the MPEP in Light of Ex parte Desjardins (Dec. 5, 2025), the claims recited a specific way of training a machine-learning model that reduced storage requirements and preserved performance on earlier tasks. That was an improvement to how the model itself operates. Applicant’s claims do not train, adjust, or change the operation of the generative model. They select the text supplied to it. The same distinction separates claim 3 of Example 47 of the July 2024 Subject Matter Eligibility Update, where the output of the trained network caused a specific network-security action, from claim 2 of that example, where a trained network was used at a high level of generality to detect anomalies and output the result. Applicant’s claims are analogous to Example 47, claim 2: the model is used at a high level of generality, and the output is simply provided to a user (¶¶[0040]-[0042]).
Applicant’s cited ¶¶[0022]-[0024] describe the system feeding a patent document and technical information to a known generative AI and receiving a correspondence result. ¶¶[0049]-[0050] describe steps S102 and S104 of the flowchart. Neither passage describes a technical improvement in the computer or the model. Accordingly, the rejection of claims 1, 2, 4, and 5 under 35 U.S.C. 101 is maintained. New claim 6 is rejected for the reasons given in the § 101 rejection below.
Rejection under 35 U.S.C. 103 (Remarks, pp. 8-11)
Applicant’s arguments with respect to claims 1, 2, 4, and 5 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. The rejection of claim 3 is moot in view of its cancellation. Walters (US 2025/0086737 A1), Lillemo (US 2022/0180051 A1), and Perkowski (US 2016/0048936 A1) are no longer applied.
So that the record is complete, the examiner addresses the substance of the argument. Applicant argues that the art does not teach “the claimed sequence of inputting an independent claim and a first instruction text to a generative model to acquire a characteristic part extracted from the independent claim and then inputting the acquired characteristic part to a generative model together with technical information and a second instruction text to obtain an investigation result.” The new rejection relies on Lin for the second generative-model process. Lin’s prompt includes a claim element, the prior-art document, and an instruction to assess whether the element is disclosed and to find the passage that discloses it (Lin, ¶¶[0005], [0025]-[0029]). The rejection relies on Plotkin for generating, with a language model, a part or summary of a patent claim describing its features and then using that model output as the input to the next language-model pass (Plotkin, ¶¶[0297], [0317]-[0320], [0600]). The combination meets the claimed sequence for the reasons given in the § 103 rejection below.
New claim 6 is argued only through its dependency on claim 1. It is rejected below over Lin and Plotkin, further in view of Chang.
Claim Interpretation
35 U.S.C. 112(f). The amendments removed the “extraction unit,” “acquisition unit,” and “output unit” limitations. Amended claim 1 recites “a processor configured to” perform the recited processes. A processor is a structural term, so the limitations of claims 1, 2, and 6 do not invoke 35 U.S.C. 112(f). MPEP 2181(I). Claims 4 and 5 recite a method and a storage medium and do not use means-plus-function or step-plus-function language.
“Generative model.” The specification describes the generative model as “a trained model that has been machine-learned in advance to generate and output a text, an image, or the like in response to an input,” giving GPT-3, GPT-4, BERT, and RoBERTa as examples (¶[0031]). Under the broadest reasonable interpretation, a commercially available large language model (LLM) is a generative model.
“Instruction text.” The specification equates the instruction text with a prompt (“a predetermined instruction text (prompt)” (¶[0022])). The “first instruction text” and the “second instruction text” are interpreted as prompts that direct the model to perform the recited task. The specification states that “the instruction text includes the description of the independent claim” (¶[0034]). Accordingly, the independent claim and the first instruction text may be supplied in a single prompt.
“Characteristic part.” The specification describes the characteristic part as the “point” of the disclosure, as opposed to a premise portion such as the preamble (¶¶[0025]-[0026], [0034]). The specification does not define the term to exclude any particular claim language, and it states that the characteristic part may be one component or parts of several components (¶[0025]). In the claims, the characteristic part is whatever text the generative model returns in response to the first instruction text. Under the broadest reasonable interpretation, the term covers text identified from an independent claim as the portion or features of the claim to be investigated.
“Technical information.” The specification defines technical information as “at least one of textual information, image information, and video information in which the technical matter is expressed,” and gives “a patent document” as an example of textual information (¶[0023]). A prior-art patent document is technical information.
“At least one of a first investigation … and a second investigation.” The specification presents the two investigations as alternatives (¶¶[0036]-[0037], introduced with “Alternatively”) and describes “at least one of” the two instruction texts (¶¶[0011], [0053]). The phrase is therefore read in the alternative: a second instruction text that instructs either investigation meets the limitation. Lin teaches both investigations in any event, as shown below, so the result does not depend on this reading.
“The generative model.” Claims 1, 4, and 5 recite a single generative model used in both processes. A single LLM that receives successive prompts meets the limitation.
Claim Rejections - 35 U.S.C. 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, 2, and 4-6 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (an abstract idea) without significantly more. The analysis follows MPEP 2106 and the July 17, 2024, Subject Matter Eligibility Update (including its artificial-intelligence examples), applied in light of the August 4, 2025, memorandum noted above.
Step 1: Statutory category
Claim 1 recites a device having a processor (a machine). Claim 4 recites a method (a process). Claim 5 recites a non-transitory storage medium (a manufacture). Claims 2 and 6 depend from claim 1. Each claim falls within a statutory category. Because claim 5 is limited to a “non-transitory” medium, it does not cover a transitory signal per se.
Step 2A, Prong One: Independent claim 4
Claim 4 is analyzed first; claims 1 and 5 recite the same steps. Claim 4 recites the following steps, with the abstract idea shown in plain text and the additional elements identified in brackets:
“…causing a computer to execute a first generative-model process and a second generative-model process using a generative model” [additional elements: computer; generative model]
“inputting, to the generative model, an independent claim in a patent document and a first instruction text instructing extraction of a characteristic part of a disclosure described in the independent claim” and “acquiring, from the generative model, the characteristic part extracted from the independent claim”. The underlying activity is identifying the characteristic part (the point) of an independent claim. A person reads a claim and decides which language carries the inventive point and which is preamble or premise. This is an observation, evaluation, and judgment that can practically be performed in the human mind or with pen and paper. [additional elements: inputting text to, and acquiring text from, the generative model]
“inputting, to the generative model, the characteristic part acquired in the first generative-model process, technical information, and a second instruction text instructing at least one of a first investigation of whether a technical feature related to the characteristic part is present in the technical information and a second investigation of a portion of the technical information where the technical feature related to the characteristic part is present” and “acquiring, from the generative model, an investigation result output in response to the second instruction text”. The underlying activity is comparing the identified point of the claim with a technical document and deciding whether, and where, a related feature is described. This is the mental comparison a patent examiner, searcher, or attorney performs in preparing a claim chart, and it can practically be performed in the human mind or with pen and paper. [additional elements: inputting text to, and acquiring text from, the generative model]
“outputting the investigation result” [additional element: outputting data]
Identifying the characteristic part of a claim and determining whether and where that part appears in a document are concepts performed in the human mind, including observation, evaluation, and judgment. They fall within the mental processes grouping of abstract ideas. MPEP 2106.04(a)(2)(III). Courts have treated similar collection, comparison, and analysis of information as abstract. Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1353-54 (Fed. Cir. 2016) (collecting information, analyzing it, and displaying the results); Content Extraction & Transmission LLC v. Wells Fargo Bank, N.A., 776 F.3d 1343, 1347 (Fed. Cir. 2014) (extracting data from documents and recognizing specific information in it). That a computer, or a generative model, performs the comparison faster or over more documents does not change the character of the comparison. Bancorp Servs., L.L.C. v. Sun Life Assurance Co. of Canada (U.S.), 687 F.3d 1266, 1278 (Fed. Cir. 2012) (performing calculations “more efficiently via a computer does not materially alter the patent eligibility of the claimed subject matter”). Claim 4 therefore recites an abstract idea. The generative-model steps are not treated as part of the mental process; they are evaluated as additional elements below.
Step 2A, Prong Two: Independent claim 4
The additional elements are: (a) a computer; (b) a generative model; (c) inputting text (the independent claim, the characteristic part, the technical information, and the instruction texts) to the generative model; (d) acquiring the model’s output; and (e) outputting the investigation result.
Elements (a) and (b) are recited at a high level of generality. The computer is a generic computer with a CPU, memory, and storage (¶¶[0043]-[0045]). The generative model is any known trained text-generating model (¶¶[0022], [0031]). Elements (c) and (d) do no more than use the generative model, as a tool, to perform the identification and comparison that make up the abstract idea: the claim states what the model is to produce (a characteristic part; an investigation result) but not how. This is a mere instruction to apply the exception using a generic computer and a generic machine-learning model. MPEP 2106.05(f); Recentive, 134 F.4th 1205; July 2024 Subject Matter Eligibility Update, Example 47, claim 2. Element (e), outputting the result to a user, is insignificant extra-solution activity. MPEP 2106.05(g).
Technological-improvement check. The specification states the following advantages: the investigation result is more accurate because noise from the non-characteristic part is reduced (¶¶[0006], [0026]-[0027], [0052]); considering dependent claims gives “more accurately” investigated relevance (¶[0056]); and limiting the input to portions above a similarity threshold is “potentially improving the accuracy of the investigation result” (¶[0039]). Claim 4 reflects the first advantage and claim 6 reflects the third. Each advantage is an improvement in the accuracy of the analysis of information, that is, in the abstract idea itself, and not an improvement in the functioning of the computer or the generative model or in another technology. MPEP 2106.04(d)(1), 2106.05(a)(II); Trading Technologies, 921 F.3d at 1093-94; SAP America, 898 F.3d at 1163. The specification does not describe a reduction in processing load, memory, or latency, or any change to the operation of the generative model.
Cases applicant may rely on. Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), involved a self-referential table that improved how a computer stores and retrieves data; claim 4 does not change how the computer stores or retrieves anything. McRO, Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299 (Fed. Cir. 2016), involved specific rules that replaced an animator’s subjective judgment and that the claim itself recited; claim 4 recites no rules for identifying the characteristic part and leaves that judgment to the model. BASCOM Global Internet Servs., Inc. v. AT&T Mobility LLC, 827 F.3d 1341 (Fed. Cir. 2016), involved a non-conventional placement of a filtering tool within a network; claim 4 uses a generic computer and a generic model in their ordinary arrangement. DDR Holdings, LLC v. Hotels.com, L.P., 773 F.3d 1245 (Fed. Cir. 2014), involved a solution to a problem arising specifically in computer networks; the problem here, an analysis distracted by preamble language, arises equally when a person performs the comparison by hand. Ex parte Desjardins is distinguished in the Response to Arguments above. None of these decisions is analogous.
Considered individually and as a whole, the additional elements do not integrate the abstract idea into a practical application. Claim 4 is directed to the abstract idea.
Step 2B: Independent claim 4
The additional elements do not amount to significantly more than the abstract idea. As explained at Prong Two, the computer and generative model are generic, and the inputting and acquiring steps are mere instructions to apply the exception with them. MPEP 2106.05(f). Mere instructions to apply an exception cannot provide an inventive concept. Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208, 223-26 (2014). The outputting step was found to be insignificant extra-solution activity at Prong Two, and it is well-understood, routine, and conventional activity: transmitting the result over a network to the user terminal for display (¶¶[0041]-[0042]). MPEP 2106.05(d)(II) (“[r]eceiving or transmitting data over a network”), citing Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1321 (Fed. Cir. 2016); see also Electric Power Group, 830 F.3d at 1354 (displaying the results of an analysis is ancillary to the abstract idea). The conventional nature of the generative model is supported by the specification itself, which describes it as “a known generative AI” and names commercially available models (¶¶[0022], [0031]). MPEP 2106.05(d)(I)(2) and 2106.07(a)(III)(A) (a statement in the specification may establish that an element is conventional). The specification describes the computer as general-purpose hardware (¶¶[0043]-[0045], [0059]-[0060]).
Viewed as an ordered combination, the elements are a generic computer prompting a generic model twice in sequence, with the first output becoming part of the second prompt and displaying the answer. Prompting a language model in stages, and using one output as the next input, is the ordinary way such models are used; see, for example, Plotkin, ¶¶[0318], [0600], and Lin, ¶¶[0030], [0051], [0065], applied below. The ordered combination adds nothing beyond the elements taken individually. Claim 4 is not patent eligible.
Independent claims 1 and 5
Claim 1 recites the same processes as claim 4, performed by “a processor configured to” perform them. Claim 5 recites the same processes stored as a program on “a non-transitory storage medium” executed by a computer. The processor and the storage medium are generic computer components (¶¶[0043]-[0044], [0059]-[0060]) and are mere instructions to apply the exception. MPEP 2106.05(f). Claims 1 and 5 are ineligible for the same reasons as claim 4.
Dependent claims
Claim 2. Claim 2 adds “inputting, to the generative model, at least a portion of descriptions of dependent claims in the patent document together with the characteristic part, the technical information, and the second instruction text.” Considering the dependent claims together with the characteristic part in the comparison is part of the mental comparison itself; a person comparing a claim with a document routinely reads the dependent claims as well. The only additional element is supplying that text to the generative model, which is again a mere instruction to apply the exception. MPEP 2106.05(f). The asserted benefit, a more accurate comparison (¶[0056]), is an improvement in the abstract idea. The limitation does not integrate the exception at Prong Two and does not add significantly more at Step 2B. Claim 2 is not patent eligible.
Claim 6. Claim 6 adds “vectorize the characteristic part …; vectorize information present in each of portions of the technical information; calculate a similarity between a vector of the characteristic part and a vector of the information present in each of the portions of the technical information; and input, to the generative model in the second generative-model process, as the technical information, only information present in a portion … for which the calculated similarity is equal to or greater than a predetermined threshold.” Vectorizing text and calculating a similarity between vectors are mathematical calculations and fall within the mathematical concepts grouping. MPEP 2106.04(a)(2)(I)(C). Selecting only the portions whose score meets a threshold is a judgment about which information to consider, which is part of the mental process. Claim 6 therefore adds further abstract ideas. The only additional element is supplying the selected portions to the generative model, which is a mere instruction to apply the exception (MPEP 2106.05(f)). The specification describes the vectorization as performed with “a known vectorization technique” (¶[0039]), which is conventional at Step 2B. MPEP 2106.05(d).
Closest to eligibility. Claim 6 comes closest, because it adds a concrete computational step that a person would not ordinarily perform by hand. However, it does not cross the line. The specification attributes the threshold filtering only to “potentially improving the accuracy of the investigation result” (¶[0039]), which is an improvement in the analysis. The specification does not describe, and the claim does not recite, any reduction in the computing resources the generative model consumes or any change to how the model processes its input. The calculation itself is recited at a high level of generality (“a similarity”; “a predetermined threshold”). Claim 6 is not patent eligible.
Suggested amendment (§ 101)
The examiner has not located in the specification a description of an improvement to the functioning of the computer or of the generative model; the stated benefits relate to the accuracy of the analysis (¶¶[0006], [0027], [0039], [0052]-[0053], [0056]). An amendment would likely need to recite, with specification support, a specific technical implementation that changes how the generative model or the computer operates, rather than what text is supplied to the model. If applicant contends that the threshold filtering of claim 6 provides a technical benefit, such as keeping the input within the model’s input limit or reducing processing, applicant is invited to identify the specification support, because the examiner has not found it. Applicant may also submit evidence, for example a declaration under 37 CFR 1.132, showing that the claimed arrangement improves the operation of the model or computer. Claim 6 is the claim closest to eligibility.
Claim Rejections - 35 U.S.C. 112
No rejection under 35 U.S.C. 112 is made. The examiner reviewed each of claims 1, 2, and 4-6 for written description, enablement, and definiteness. Two close questions are resolved on the record as follows.
Claim 2, written description. Claim 2 requires inputting the dependent-claim descriptions “together with the characteristic part, the technical information, and the second instruction text.” ¶[0056] describes inputting dependent-claim descriptions as a modification of the embodiment that inputs the characteristic part, and original claim 2 depended from original claim 1, which input the characteristic part. The original disclosure therefore supports the combination. MPEP 2163(I)(B) (“The claims as filed in the original specification are part of the disclosure”).
“Characteristic part,” definiteness. The term is defined functionally by the claim (it is what the generative model returns in response to an instruction to extract it), and the specification explains its meaning with an example (¶¶[0025]-[0026], [0034]). A person of ordinary skill would understand the scope with reasonable certainty. Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.S. 898, 910 (2014). The breadth of the term is addressed under Claim Interpretation and in the § 103 rejection, not as indefiniteness. MPEP 2173.04.
Claim Rejections - 35 U.S.C. 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: (1) determining the scope and contents of the prior art; (2) ascertaining the differences between the prior art and the claims at issue; (3) resolving the level of ordinary skill in the pertinent art; and (4) considering objective evidence present in the application indicating obviousness or nonobviousness. Graham v. John Deere Co., 383 U.S. 1, 17-18 (1966).
Prior Art of Record
Lin. US 2025/0245762 A1, “Method and System to Generate Patent Claim Charts,” first-named inventor Daniel Lin, Appl. No. 18/424,829, filed January 28, 2024, published July 31, 2025 (no assignee listed). Lin was effectively filed before December 26, 2024, and names another inventor, so it is prior art under 35 U.S.C. 102(a)(2). Lin discloses a service platform that generates patent claim charts by prompting an LLM, for each claim element of a selected independent claim, with the prior-art document, the claim, and the claim element, and instructing the LLM to assess whether the element is disclosed and to find the most similar passage. Lin is in applicant’s field: using a generative model to compare patent-claim content with technical information.
Plotkin. US 2024/0211685 A1, “Language Model-Based Multi-Objective Optimization,” first-named inventor Robert Plotkin, applicant Quabbin Patent Holdings, Inc., Appl. No. 18/391,114, filed December 20, 2023, published June 27, 2024. Plotkin was published before December 26, 2024, and is prior art under 35 U.S.C. 102(a)(1). Plotkin discloses using a language model to analyze patent claims against products and prior art, including a workflow in which the language model generates output describing the features of a patent claim, and that output is used as the input for a further language-model pass. Plotkin is in the same field of endeavor as the claims.
Chang. WO 2024/173577 A1, “Information Retrieval of Patent-Related Corpora with Semantic and Large Language Models,” first-named inventor Wayne Chang, applicant Licensing, Inc., International Appl. No. PCT/US2024/015833, filed February 14, 2024, published August 22, 2024. Chang was published before December 26, 2024, and is prior art under 35 U.S.C. 102(a)(1). Chang discloses generating embeddings of patent claims and of document chunks, comparing them for similarity, and supplying to an LLM only the content whose embeddings meet a threshold proximity, for patent infringement and claim-chart analysis. Chang is in the same field of endeavor as the claims.
Designation of Primary and Secondary References; Level of Ordinary Skill
Lin is the Primary reference. Lin, on its own, teaches the device, the second generative-model process (a prompt containing claim-element text, technical information, and an instruction to determine whether and where the element is disclosed), acquiring the LLM’s result, and outputting it. Lin also teaches incorporating dependent-claim limitations into the prompt (claim 2). Lin does not teach the first generative-model process: Lin divides the claim into elements by its punctuation and lets the user identify the language of particular importance (Lin, ¶[0024]), rather than prompting the LLM with the independent claim to extract that part. Plotkin is the Secondary reference, combined to supply that first process and the use of its output in the second process. Chang is a further Secondary reference, combined with Lin and Plotkin for the vector-similarity threshold filtering of claim 6.
A person of ordinary skill in the art would have had a bachelor’s degree in computer science or a related field, or equivalent experience, and working familiarity with prompting large language models, text embeddings, and patent claim analysis. The applied references themselves reflect that level of skill. Okajima v. Bourdeau, 261 F.3d 1350, 1355 (Fed. Cir. 2001).
Rejection statements
Claims 1, 2, 4, and 5 are rejected under 35 U.S.C. 103 as being unpatentable over Lin (US 2025/0245762 A1) in view of Plotkin (US 2024/0211685 A1).
19/408,620 – Claim 1
Claim 1 recites “An information processing device comprising a processor configured to perform a first generative-model process and a second generative-model process using a generative model”. Lin teaches this feature. Lin discloses a service platform whose backend runs on cloud infrastructure and may interact with “one or more large language models (LLMs), collectively referred to as LLM 120”; LLM 120 may include “OpenAI's GPT-4 service, Anthropic's Claude 2 service, Google's Gemini service” (Lin, ¶[0014]), and states that the system may be “one or more general-purpose computers selectively activated or configured by a computer program stored therein” (Lin, ¶[0068]). A general-purpose computer executing a program includes a processor. Lin’s LLMs are generative models; they have “the capability to comprehend and generate text that closely resembles human communication” (Lin, ¶[0003]). Lin performs more than one LLM process with the same LLM: an element-analysis prompt (Lin, ¶[0025]), a feedback prompt (Lin, ¶[0033]), and a motivation-to-combine prompt that includes an earlier LLM analysis as context (Lin, ¶[0051]).
Claim 1 recites the first generative-model process: “inputting, to the generative model, an independent claim in a patent document and a first instruction text instructing extraction of a characteristic part of a disclosure described in the independent claim” and “acquiring, from the generative model, the characteristic part extracted from the independent claim”. Lin teaches part of this feature. Lin selects an independent claim: the platform presents patent data “so that the user may select one of the independent claim numbers of the entered patent to evaluate” (Lin, ¶[0022]). Lin then “divides the patent claim into a plurality of default claim elements or phrases,” and alternative embodiments allow “the user to highlight or identify more specific language in the claim element that is of particular importance” (Lin, ¶[0024]). Lin uses that focus language in later prompts, which ask the LLM to proceed “with a particular emphasis on the phrase “${elementFocusText}” in the claim element” (Lin, ¶[0045]). The language “of particular importance” in an independent claim is a characteristic part of the claim under the broadest reasonable interpretation (see Claim Interpretation). Primary prior-art reference Lin does not explicitly teach “inputting, to the generative model, an independent claim in a patent document and a first instruction text instructing extraction of a characteristic part of a disclosure described in the independent claim” or “acquiring, from the generative model, the characteristic part extracted from the independent claim”; in Lin, the claim is divided by punctuation and the important language is identified by the user, not by the LLM.
However, secondary prior-art reference Plotkin teaches this feature as follows. Plotkin extracts claim text from a patent document and supplies it to a language model with a query: the system may “automatically extract … text of some or all of the claims of the patent document” and “may include some or all of that extracted data within the initial text 118, such as within any of the query text” (Plotkin, ¶[0297]). In one workflow, the system generates, “based on the user input 104 and the language model 112, language model output 114 which describes features encompassed by the patent document,” which may include “one or more of the claims of the patent document (such as claim 1 of the patent document)” and “a part and/or summary of any of the preceding data” (Plotkin, ¶[0317]). Claim 1 of a patent document is ordinarily an independent claim. The query text that accompanies the claim text is an instruction text, and the model’s output describing the claim’s features, including a part or summary of claim 1, is text extracted from the independent claim that identifies what the claim is about. Plotkin also explains why a user would want such text: claims “may describe the particular feature in complex language,” and the user may generate “data describing the particular feature in simpler or otherwise different language” and use it in the query instead of “the original complex language” (Plotkin, ¶[0298]), and any such input “may be generated using a language model” (Plotkin, ¶[0600]). Plotkin does not use the words “characteristic part,” and Plotkin’s query asks for the features of the claim rather than its “point.” In the combination, Lin supplies the target of the extraction, namely the language of the claim “of particular importance” (Lin, ¶[0024]), and Plotkin supplies the technique of having the language model generate that text from the claim.
Claim 1 further recites, in the second generative-model process, “inputting, to the generative model, the characteristic part acquired in the first generative-model process, technical information, and a second instruction text instructing at least one of a first investigation of whether a technical feature related to the characteristic part is present in the technical information and a second investigation of a portion of the technical information where the technical feature related to the characteristic part is present”. Lin teaches this feature, except that Lin’s claim-element text is not acquired in a first generative-model process. Lin’s prompt includes the claim element, the technical information, and the instruction: “For each claim element, the method generates a prompt for an LLM to assess whether the claim element is disclosed by the prior art document. To provide proper context for the LLM, the service platform includes the prior art document, the patent claim, and the claim element in the prompt” (Lin, ¶[0005]). The prompt template instructs: “Find the passage of the prior art document that is most similar to the claim element of “${claimElement}” in the claim and provide a summary on why the passage in the prior art document is most similar to the claim element” (Lin, ¶¶[0025]-[0029]). The prior-art document is technical information (see Claim Interpretation; applicant’s ¶[0023]). Instructing the LLM to assess “whether the claim element is disclosed” is an instruction for the first investigation (whether a related technical feature is present). Instructing the LLM to “[f]ind the passage … most similar to the claim element” is an instruction for the second investigation (the portion where the related feature is present). Lin’s claim 4 confirms the second alternative: “the at least one prompt requests the LLM to include in the initial analysis a passage from the prior art document that discloses the claim element if one exists” (Lin, claim 4). Lin also describes using “multiple prompts to first engage LLM 120 to ascertain whether a passage with strong enough similarity exists in the prior art document” (Lin, ¶[0030]). The ${claimElement} text, and the focus phrase of ¶[0045], is a feature related to the part of the independent claim under investigation.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify Lin’s claim-element preparation step (Lin, ¶[0024], step 310), in which the claim is divided by punctuation and the user highlights the language of particular importance, to instead input the selected independent claim to the LLM together with a query instructing the LLM to identify that language and to acquire it from the LLM, as Plotkin teaches for generating text describing a claim’s features with a language model and using that output as the input to the next language-model pass (Plotkin, ¶¶[0297], [0317]-[0319], [0600]), and then to insert the acquired text into Lin’s element-analysis prompt in place of the user-highlighted text (Lin, ¶¶[0025]-[0029], [0045]). The resulting combination inputs the independent claim and a first instruction text to the generative model, acquires the extracted characteristic part, and inputs that acquired part with the prior-art document and the second instruction text, as claimed. The reason to make the modification comes from the references. Plotkin teaches that “[t]he system 100 may use the language model output 114 as new user input 104 and/or generate new user input 104 based on the language model output 114” (Plotkin, ¶[0318]) and that inputs generated in this way yield language-model outputs whose “quality (e.g., accuracy and/or relevance) … may be higher than if the original input … had been provided as input to the language model” (Plotkin, ¶[0597]; see also ¶[0599]). Lin itself seeks to focus the analysis on the language “of particular importance” (Lin, ¶[0024]) and already contemplates breaking the assessment “into a series of more targeted prompts” and using “chain-of-thought” prompting (Lin, ¶¶[0030], [0065]); Lin also feeds one LLM analysis into a later prompt “for additional context” (Lin, ¶[0051]). Automating Lin’s manual highlighting step with the LLM Lin already uses also removes a manual step from a workflow whose stated purpose is to reduce the professional’s review time (Lin, ¶[0004]). One of ordinary skill would have had a reasonable expectation of success because both references prompt the same kind of commercially available LLM with text and read back text, and the modification only changes where Lin’s ${claimElement} or ${elementFocusText} text comes from. The rationale is MPEP 2143(I)(G), with (C): Plotkin explicitly suggests generating language-model inputs with a language model and chaining one model output into the next pass to improve output quality, and applying that known technique to Lin’s claim-element prompt improves Lin’s prompt in the same way. The modification does not change Lin’s principle of operation (prompting an LLM with a claim element and a prior-art document) or make Lin unsatisfactory for its intended purpose of generating claim charts, and the predictable result is a claim chart whose element analyses are focused on LLM-identified important claim language. KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 416 (2007). This reason to combine comes from Plotkin and Lin themselves, not from applicant’s disclosure.
Claim 1 further recites “acquiring, from the generative model, an investigation result output in response to the second instruction text”. Lin teaches this feature: “Service platform 100 transmits each of these prompts to LLM 120 to obtain an initial analysis for each claim element,” and “LLM 120 transmits the initial analyses for each claim element back to service platform 100 in step 314” (Lin, ¶[0031]). Each initial analysis “has a summary 416 x (including a rating 418 x) as well as quoted passages 420 x from the prior art document” (Lin, ¶[0031]), which is an investigation result output in response to the prompt. No secondary reference is needed to meet this limitation.
Claim 1 further recites “outputting the investigation result”. Lin teaches this feature: “service platform 100 then presents the results in a claim chart to the user in step 316” (Lin, ¶[0031]; FIG. 4E). No secondary reference is needed to meet this limitation.
19/408,620 – Claim 2
Claim 2 recites “wherein the second generative-model process further includes inputting, to the generative model, at least a portion of descriptions of dependent claims in the patent document together with the characteristic part, the technical information, and the second instruction text”. Lin teaches this feature. Lin states that “service platform 100 may incorporate the substantive limitations of a dependent claim into its corresponding independent claim, and utilize the combined claim in its generated prompts for LLM 120” (Lin, ¶[0067]). Lin’s element-analysis prompt contains the claim text (${claimText}) together with the claim element and the prior-art document and the instruction (Lin, ¶¶[0025]-[0029]). Using the combined claim as the ${claimText} therefore inputs at least a portion of the descriptions of the dependent claims together with the claim-element text, the technical information, and the second instruction text. Lin also creates a claim chart object for each “patent claim set (e.g., independent claim and claims dependent thereupon)” (Lin, ¶[0017]). In the combination of Lin and Plotkin set out for claim 1, the claim-element text is the characteristic part acquired in the first process, so the prompt includes the dependent-claim descriptions together with the characteristic part, the technical information, and the second instruction text. Plotkin likewise uses “a combination (e.g., concatenation) of the text of the dependent claim's ancestors with the text of the dependent claim” in its language-model queries (Plotkin, ¶[0321]). No further modification beyond that given for claim 1 is needed; the combination is the same, for the reasons given for claim 1.
19/408,620 – Claim 4
Claim 4 is the method counterpart of claim 1. Claim 4 recites “An information processing method comprising causing a computer to execute a first generative-model process and a second generative-model process using a generative model.” Lin teaches a method (“One method that generates a claim chart for a patent claim against a prior art document can be performed … by a web application that provides a service platform” (Lin, ¶[0005])) executed on “one or more general-purpose computers” (Lin, ¶[0068]), using an LLM as discussed for claim 1. The remaining limitations of claim 4, “inputting, to the generative model, an independent claim in a patent document and a first instruction text instructing extraction of a characteristic part of a disclosure described in the independent claim”; “acquiring, from the generative model, the characteristic part extracted from the independent claim”; “inputting, to the generative model, the characteristic part acquired in the first generative-model process, technical information, and a second instruction text instructing at least one of a first investigation of whether a technical feature related to the characteristic part is present in the technical information and a second investigation of a portion of the technical information where the technical feature related to the characteristic part is present”; “acquiring, from the generative model, an investigation result output in response to the second instruction text”; and “outputting the investigation result”, are identical to those of claim 1 and are met by Lin in view of Plotkin for the reasons given for claim 1, including the combination paragraph given there.
19/408,620 – Claim 5
Claim 5 is the storage-medium counterpart of claim 1. Claim 5 recites “A non-transitory storage medium storing an information processing program causing a computer to execute a first generative-model process and a second generative-model process using a generative model.” Lin teaches this feature: “One or more embodiments of the invention may be implemented as one or more computer programs or as one or more computer program modules embodied in computer readable media,” and “[e]xamples of computer readable media are HDDs, SSDs, network-attached storage (NAS) systems, read-only memory (ROM), RAM, compact disks (CDs), digital versatile disks (DVDs), magnetic tapes, and other optical and non-optical data storage devices” (Lin, ¶[0069]). These are non-transitory media. The remaining limitations of claim 5, “inputting, to the generative model, an independent claim in a patent document and a first instruction text instructing extraction of a characteristic part of a disclosure described in the independent claim”; “acquiring, from the generative model, the characteristic part extracted from the independent claim”; “inputting, to the generative model, the characteristic part acquired in the first generative-model process, technical information, and a second instruction text instructing at least one of a first investigation of whether a technical feature related to the characteristic part is present in the technical information and a second investigation of a portion of the technical information where the technical feature related to the characteristic part is present”; “acquiring, from the generative model, an investigation result output in response to the second instruction text”; and “outputting the investigation result”, are identical to those of claim 1 and are met by Lin in view of Plotkin for the reasons given for claim 1.
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Lin (US 2025/0245762 A1) in view of Plotkin (US 2024/0211685 A1), and further in view of Chang (WO 2024/173577 A1).
19/408,620 – Claim 6
Claim 6 depends from claim 1, whose limitations are met by Lin in view of Plotkin for the reasons given for claim 1. Claim 6 recites that the processor is further configured to “vectorize the characteristic part acquired in the first generative-model process”; “vectorize information present in each of portions of the technical information”; “calculate a similarity between a vector of the characteristic part and a vector of the information present in each of the portions of the technical information”; and “input, to the generative model in the second generative-model process, as the technical information, only information present in a portion, among the portions, for which the calculated similarity is equal to or greater than a predetermined threshold”.
Lin teaches limiting the technical information placed in the prompt: “Alternative embodiments may use only portions of the prior art document (e.g., only the “Detailed Description” of a prior art document, etc.) in the prompt” (Lin, ¶[0030]). The combination of Lin and Plotkin does not explicitly teach vectorizing the characteristic part and the portions of the technical information, calculating a similarity between the vectors, or inputting only the portions whose similarity meets a predetermined threshold.
However, secondary prior-art reference Chang teaches these features as follows. Chang vectorizes a patent document by portions: “the patent infringement module then splits the patent into chunks in step 2420,” and “[f]or each chunk the model then obtains embeddings in step 2425,” and it also “obtains embeddings for the product description in step 2430” (Chang, p. 28, l. 29 to p. 29, l. 8). Chang calculates a similarity between the vectors and prompts the LLM with the most similar portions: “The system then searches for patent chunks that are most similar to the product description in step 2435. This is illustratively performed by comparing the embeddings from the various chunks of the patent to the embeddings from the product description. … Then a prompt is created to the large language model in step 2440” (Chang, p. 29, ll. 10-15). Chang applies a threshold to the similarity and passes on only the content that meets it: the system compares document embeddings “to embeddings of the patents of interest, its claims,” and “can optionally focus on documents whose embeddings have a threshold proximity to the embeddings of patents and claims of interest, which saves significant computing resources (i.e., no need to analyze the entirety of documents, only the similar ones)” (Chang, p. 3, ll. 22-29); “[r]esulting product descriptions with relatedness above a threshold are injected into a large language model (LLM)” (Chang, Abstract). Embedding text is vectorizing it, and a threshold proximity of embeddings is a similarity equal to or greater than a predetermined threshold. Chang differs from the claim in two respects: in the passage at pp. 28-29 Chang divides the patent, rather than the technical document, into chunks, and in the passage at p. 3 Chang applies the threshold to whole documents rather than to chunks. Both differences are addressed in the combination below.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the combination of Lin and Plotkin so that, before Lin’s element-analysis prompt is built (Lin, ¶¶[0025]-[0030]), the system embeds the characteristic part acquired in the first process, splits the prior-art document into chunks and embeds each chunk, computes the similarity between the characteristic-part embedding and each chunk embedding, and places in the ${priorArtDocument} portion of the prompt only the chunks whose similarity meets a threshold, as Chang teaches for comparing claim embeddings with document embeddings, comparing chunk embeddings, and passing to the LLM only content within a threshold proximity (Chang, p. 3, ll. 22-29; p. 28, l. 29 to p. 29, l. 15; Abstract). The resulting combination vectorizes the characteristic part and each portion of the technical information, calculates their similarity, and inputs to the generative model only the portions at or above the threshold, as claim 6 requires. The motivation comes from the references: Chang teaches that restricting the LLM’s input to content within a threshold proximity “saves significant computing resources (i.e., no need to analyze the entirety of documents, only the similar ones)” (Chang, p. 3, ll. 22-29) and that data preparation “can avoid exceeding an LLM’s context window” (Chang, p. 19, ll. 17-19), and Lin already contemplates using “only portions of the prior art document” in the prompt (Lin, ¶[0030]). Applying Chang’s chunk-level embedding comparison to the technical document instead of the patent and applying Chang’s threshold to those chunks instead of to whole documents, uses the same technique for the same purpose of selecting the most relevant text for the LLM. One of ordinary skill would have had a reasonable expectation of success because embedding-based similarity search and thresholding were established techniques for selecting LLM input text, and Chang applies them to the same kind of claim-versus-document analysis Lin performs. The rationale is MPEP 2143(I)(C) and (D): the combination of Lin and Plotkin is a known claim-analysis system ready for improvement, and Chang’s embedding-and-threshold selection was known to improve such systems in the same way, by reducing the content the LLM must analyze to the relevant portions. The modification does not change the principle of operation of the combination (prompting an LLM with claim text, a document, and an instruction), and the predictable result is the same element analysis performed on a shorter, more relevant input. KSR, 550 U.S. at 416. This reason to combine comes from Chang and Lin, not from applicant’s disclosure.
Suggestions for overcoming the art
The following amendments are suggested for applicant’s consideration. Each would require further search and consideration, and each is offered for compact prosecution, not as an indication of allowability.
Claim 6 limited to video technical information, with per-frame vectorization: recite that the technical information is video information, that each frame image constituting the video is vectorized, and that only frames whose similarity to the vectorized characteristic part meets the threshold are input to the generative model (¶¶[0023], [0039]). Chang vectorizes text chunks (Chang, p. 28, l. 29 to p. 29, l. 15), and Lin analyzes text documents. The closest art for image input is CN 118569653 A (made of record below), which compares image vectors and inputs related images to a multimodal language model.
Claims 1, 4, and 5 limited to the excluded premise portion: recite that the first instruction text instructs the generative model to exclude the preamble and the premise portion of the independent claim and to output only the component or components that constitute the point of the disclosure, and that the second instruction text instructs the model to output, for each extracted point, a determination, a reason, and the original text of the described portion (¶¶[0025]-[0026], [0034], [0038]). Lin’s chart outputs a summary, a rating, and quoted passages (Lin, ¶[0031]), and Lin’s focus language is chosen by the user (Lin, ¶[0024]). Lin and Plotkin would remain the closest art.
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Bray, GB 2622298 A (Added Matter Ltd; first-named inventor Edward William Arthur Bray), “Patent claim feature mapping,” filed June 26, 2023, published March 13, 2024 (§ 102(a)(1)). This reference is pertinent to claims 1, 4, 5, and 6. It inserts “the entire claim” into a large-language-model prompt “together with instructions to generate one or more semantic search terms associated with each feature of the input claim” (p. 22, ll. 17-20), then embeds chunks of the text document and the search terms, selects the closest chunk by embedding distance, and inserts the selected text into a second prompt “further comprising the text of the one or more patent claim features, and text comprising instructions to generate text identifying where in the extended first chunk the one or more claim features are found” (claim 1, p. 44, ll. 26-29). The reference correlates with the claimed first process (an LLM prompted with the claim), the second investigation (where the feature is found), and the vector selection of claim 6. It selects the closest chunk and extends it to fill the context window, rather than applying a similarity threshold, and its first-stage output is used as a search term rather than inserted into the second prompt.
Lundberg, US 2025/0315906 A1 (Black Hills IP Holdings, LLC), “Generative Artificial Intelligence Patent Claim Mapping,” filed April 8, 2025, claiming the benefit of provisional application 63/631,154 filed April 8, 2024, published October 9, 2025. This reference is pertinent to the first generative-model process of claims 1, 4, and 5 (“a first instruction text instructing extraction of a characteristic part”). Its generative AI tool analyzes claim text and generates “scope concepts,” where “[a] scope concept may define the scope to which a patent claim is limited” (¶¶[0023]-[0025]), and presents the presence or absence of each concept across claims. It maps concepts among claims of a portfolio rather than investigating a technical document. Its status as prior art depends on support in the provisional application.
Sullivan, US 2023/0325422 A1 (Daystrom Information Systems LLC), “Hybrid Artificial Intelligence System for Semi-Automatic Patent Claims Analysis,” published October 12, 2023 (§ 102(a)(1)). This reference is pertinent to claim 6. It transforms claim elements and a reference document into embeddings and performs “an element-by-element search” to match claim elements to relevant portions of the reference document, with confidence scores based on similarity (¶¶[0026], [0045]-[0047]). It does not use a generative model to extract the claim’s characteristic part or to perform the investigation.
Jin, CN 117151052 A (Zhiguagua Tianjin Big Data Technology Co., Ltd., et al.), published December 1, 2023 (§ 102(a)(1)) (machine translation). This reference is pertinent to the first generative-model process of claims 1, 4, and 5. It uses a large language model to extract technical feature points from an invention disclosure document, searches a patent knowledge base for related documents, and extracts technical features of each retrieved document with the large language model for comparison (claim 1). The input is an invention disclosure, not an independent claim, and the comparison uses graph matching rather than a second instruction to the model.
CN 118569653 A (Shenzhen Aimo Technology Co., Ltd.), “Patent risk assessment method based on large language model,” published August 30, 2024 (§ 102(a)(1)) (machine translation). This reference is pertinent to claim 6 and to the image and video technical information of ¶[0023]. It encodes patent text and images with an embedding model and inputs retrieved related information, or related pictures, together with the file to be analyzed into a large language model or multimodal language model to obtain a similarity result (claims 6, 7, and 9). It does not extract a characteristic part of an independent claim.
Haense et al., US 2022/0365956 A1 (previously cited). This reference extracts technical problem information from a target patent document using a pre-trained extraction model (claim 2). It is pertinent to the extraction concept of claims 1, 4, and 5, but it summarizes a patent and does not investigate technical information.
Wang et al., “EvoPat: A Multi-LLM-based Patents Summarization and Analysis Agent,” arXiv:2412.18100v1 [cs.CL] (Suyuan Wang, Xueqian Yin, Menghao Wang, Ruofeng Guo, and Kai Nan; v1 submitted December 24, 2024) (non-patent literature; § 102(a)(1)). This reference is pertinent to the first and second generative-model processes of claims 1, 4, and 5. It analyzes a patent with multiple LLM agents in sequence, including an agent that is “responsible for identifying the most valuable innovative methods within the patent” and a comparison agent that retrieves similar patents for comparative analysis, and the agents “share all content generated during prior interactions” (sections 2.1 and 2.3.2). It identifies the innovations of the patent as a whole, not the characteristic part of an independent claim. Its comparison is a patent-to-patent summary rather than an investigation of whether, or where, a feature is present in a given technical document.
Walters et al., US 2025/0086737 A1; Lillemo et al., US 2022/0180051 A1; and Perkowski et al., US 2016/0048936 A1 were applied in the prior action and are no longer relied upon. They remain of record. Walters compares inventive ideas against prior art with a language model; Lillemo identifies independent and dependent claims in a document model; Perkowski analyzes claim limitation language strings against products. Each is less pertinent to the amended two-stage generative-model processes than Lin and Plotkin.
Suggested Interview
The examiner invites applicant to schedule a telephone or video interview before filing a reply, to discuss the following topics. Interview requests may be made with the Automated Interview Request (AIR) form or by contacting the examiner. Because this action is final, an interview after final rejection is at the examiner’s discretion and is granted where it is likely to advance prosecution. MPEP 713.09.
Claim 6 and video technical information (claims 1, 4-6): whether limiting the technical information to video with per-frame vectorization and threshold selection (¶¶[0023], [0039]) distinguishes Lin, Plotkin, and Chang; CN 118569653 A is the closest art for image input.
The characteristic part (claims 1, 4, 5): whether reciting exclusion of the preamble and premise portion and the determination-reason-original-text output format (¶¶[0025]-[0026], [0034], [0038]) distinguishes Lin’s user-selected focus language and Plotkin’s feature summaries.
Subject matter eligibility (all claims): whether the specification supports any improvement in the functioning of the generative model or the computer, as opposed to the accuracy of the analysis (¶¶[0006], [0027], [0039]); claim 6 is closest to eligibility.
Applicant is asked to send any proposed claim amendments to the examiner before the interview. The examiner will then be able to state, for each proposed amendment, whether it would overcome each outstanding rejection and why. MPEP 713.01, 713.04. Applicant is reminded that an amendment after final rejection is governed by 37 CFR 1.116.
Conclusion
THIS ACTION IS MADE FINAL
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.
Contact Information
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MATTHEW T. SITTNER, whose telephone number is (571) 270-7137. The examiner can normally be reached Monday through Friday, 8:00 a.m. to 5:00 p.m. MT. Interview requests may be scheduled by email to the examiner at matthew.sittner@uspto.gov.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO-supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) form.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Sarah M. Monfeldt, can be reached at (571) 270-1833.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. For questions about access to Patent Center, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). For assistance from a USPTO Customer Service Representative, call 800-786-9199 (in USA or Canada) or 571-272-1000.
/MATTHEW T SITTNER/
Primary Examiner, Art Unit 3629
