Amendment draft generation support device and amendment draft generation support program

Abstract

This invention provides an amendment draft generation support device and an amendment draft generation support program that assist in generating an amendment draft so as to resolve the grounds for rejection due to lack of inventive step. [Solution] The amendment draft generation support device 100 includes a difference data detection unit 101 that detects data in the specification data that differs from the cited reference data that constitutes the cited documents described in the notice of reasons for rejection as difference data, a provisional amendment data generation unit 102 that generates provisional amendment data using the claim data and difference data, a classification processing unit 103 that performs classification processing on whether the provisional amendment data meets the requirements for inventive step using the provisional amendment data and cited reference data, and a presentation processing unit 109 that presents an amendment draft 50 generated using conforming amendment data that shows that the result of the classification processing of the provisional amendment data meets the requirements for inventive step.

Description

Technical Field

[0001] The present invention relates to an amendment draft generation support device and an amendment draft generation support program for supporting the generation of an amendment draft as the text of an amendment document.

Background Art

[0002] Conventionally, technologies for supporting the creation of documents related to patent applications have been known. For example, Patent Document 1 discloses a creation support device that reduces the time required for creating application documents by suppressing variations in quality when creating claims (claims).

[0003] In recent years, technologies for supporting the creation of documents related to patent applications using large language models (also referred to as LLMs) have also been known. For example, Patent Document 2 discloses a patent document creation support device that supports the creation of a specification, claims, etc. by generating sentences such as questions and answers using an LLM. Patent Document 3 discloses a document creation support device that supports the creation of intermediate documents using an intermediate document database that stores intermediate document information such as amendment documents and opinion documents.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0005] As described above, according to the prior art, the burden of creating documents related to patent applications is reduced, and it is possible to efficiently create application documents.

[0006] However, in practice, a request for examination is made for approximately 80% of patent applications, and with a few exceptions, grounds for rejection are notified. In such cases, in order to proceed with obtaining patent rights, it is necessary to resolve the notified grounds for rejection by taking response procedures, such as submitting documents such as procedural amendments and statements of opinion. For this reason, documents such as procedural amendments and statements of opinion (also called response documents) hold an important position in the examination practice of patent applications, but the prior art described above did not adequately support the preparation of such response documents.

[0007] Furthermore, in order to reduce the burden more effectively, it is desirable to provide support in preparing response documents that can resolve the grounds for rejection found. In particular, since the majority of the grounds for rejection that are notified are due to failure to meet the requirement of inventive step (Article 29, Paragraph 2 of the Patent Act) (also known as deficiency in inventive step), it is desirable to provide support in preparing response documents, especially procedural amendments, that can resolve such grounds for rejection.

[0008] However, prior art did not provide support for creating such procedural amendments. Therefore, the present invention was created to support the creation of such procedural amendments.

[0009] The present invention has been made to solve the above problems and aims to provide an amendment draft generation support device and an amendment draft generation support program that support the generation of an amendment draft as a draft text for a procedural amendment, so as to eliminate the grounds for rejection due to lack of inventive step. [Means for solving the problem]

[0010] To solve the above problems, the present invention is an amendment draft generation support device that assists in generating an amendment draft for the claims of a patent application which describes an invention related to a patent application, and is characterized by having: a difference data detection unit that detects data that differs from the cited reference data that constitutes a cited reference document described in a notice of reasons for rejection which states that the claim data or specification data that constitutes a description of the invention related to a patent application does not meet the requirement of inventive step, as difference data; a provisional amendment data generation unit that generates provisional amendment data using the claim data and the difference data; a classification processing unit that performs classification processing on whether the provisional amendment data meets the requirement of inventive step using the provisional amendment data and the cited reference data; and a presentation processing unit that presents an amendment draft generated using conforming amendment data which indicates that the result of the classification processing of the provisional amendment data meets the requirement of inventive step.

[0011] In the above-described amendment draft generation support device, the classification processing unit has an effectiveness classification unit that classifies provisional amendment data on whether or not it is effective in resolving the grounds for rejection, and an inventive step suitability classification unit that performs classification processing using effective provisional amendment data classified as effective by the effectiveness classification unit and reference data from among the provisional amendment data, and a difference data detection unit has a difference detection criterion set for detecting difference data, and the difference data detection unit detects invention text data as difference data when the similarity of the claim data or specification data divided into sentence units and the reference text data divided into sentence units is smaller than the detection criterion value, and the provisional amendment data generation unit has a sorting unit that sorts the difference data in ascending order of similarity, and generates provisional amendment data using at least one of the sorted difference data sorted by the sorting unit, and the provisional amendment data generation unit can generate provisional amendment data using the minimum difference data with the smallest similarity among the sorted difference data, and then generate provisional amendment data by adding the sorted difference data one by one to the minimum difference data in order of similarity.

[0012] Furthermore, in the case of the above-mentioned amendment draft generation support device, it may also further include an instruction sentence generation unit that generates an instruction sentence instructing a large-scale language model to generate an amendment draft using the appropriate amendment data, and the presentation processing unit may present the amendment draft using the language model draft generated by the large-scale language model in accordance with the instruction sentence.

[0013] Furthermore, the above-mentioned amendment draft generation support device may also include an amendment draft generation adjustment unit that generates an amendment draft by performing text and formatting adjustments using conforming amendment data and amendment draft formatting adjustments to adjust the format to conform to the procedural amendment.

[0014] Furthermore, the present invention provides a draft amendment generation support program for causing a computer to function as a draft amendment generation support device for assisting in the generation of a draft amendment to the claims of a patent application that describes an invention related to a patent application, and the computer to function as a draft amendment generation support program that causes the computer to function as a difference data detection unit that detects data that differs from the claim data constituting the claims or specification data constituting the specification that describes the invention related to the patent application, and which constitutes cited reference data that constitutes a notice of rejection that states that the requirements for inventive step are not met; a provisional amendment data generation unit that generates provisional amendment data using the claim data and the difference data; a classification processing unit that performs classification processing regarding the suitability of the provisional amendment data for the requirements for inventive step using the provisional amendment data and the cited reference data; and a presentation processing unit that presents a draft amendment generated using suitability amendment data that indicates that the result of the classification processing of the provisional amendment data conforms to the requirements for inventive step. [Effects of the Invention]

[0015] As described in detail above, according to the present invention, an amendment draft generation support device and an amendment draft generation support program are obtained that support the generation of an amendment draft as a draft text for a procedural amendment, so as to eliminate the grounds for rejection due to lack of inventive step. [Brief explanation of the drawing]

[0016] [Figure 1] It is a system configuration diagram of a patent document generation system including a correction draft generation support device according to the first embodiment of the present invention. [Figure 2] It is a functional block diagram showing the internal configuration of a patent document server that constitutes a patent document generation system. [Figure 3] It is a block diagram showing the internal configuration of a user terminal device. [Figure 4] It is a functional block diagram showing the main configuration of a correction draft generation support device according to the first embodiment of the present invention. [Figure 5] It is a functional block diagram showing the main configuration of the classification processing unit included in the correction draft generation support device of FIG. 4. [Figure 6] It is a diagram showing an example of claim data. [Figure 7] It is a diagram showing an example of specification data. [Figure 8] It is a diagram showing an example of cited example data. [Figure 9] It is a diagram schematically showing the learned model of the validity classification unit and the learning document vectors used for its training. [Figure 10] It is a functional block diagram showing the main configuration of the inventiveness determination classification unit included in the classification processing unit of FIG. 5. [Figure 11] It is a diagram showing an example of difference data. [Figure 12] It is a diagram showing an example of provisional correction data. [Figure 13] (A) is a diagram showing an example of a language model draft, and (B) is a diagram showing an example of a correction draft. [Figure 14] It is a system configuration diagram of a patent document generation system including a correction draft generation support device according to the second embodiment of the present invention. [Figure 15] It is a functional block diagram showing the main configuration of a correction draft generation support device according to the second embodiment of the present invention. [Figure 16] It is a diagram showing an example of a correction draft generated by the correction draft generation support device of FIG. 15. [Modes for carrying out the invention]

[0017] Embodiments of the present invention will be described below with reference to the drawings. The same reference numerals are used for the same elements, and redundant descriptions will be omitted.

[0018] First Embodiment (Overall configuration of the patent document generation system) First, the configuration of the patent document generation system 1, which includes the amendment draft generation support device 100 according to an embodiment of the present invention, will be described. In the patent document generation system 1, an amendment draft 50 is generated using a large-scale language model.

[0019] Figure 1 is a system configuration diagram of the patent document generation system 1. As shown in Figure 1, the patent document generation system 1 has a patent document server 10, a user terminal device 30 operated by the user, and a large-scale language model server (LLM server) 400, and these are connected to each other via the internet N1.

[0020] The patent document server 10 has a draft amendment generation support device 100. The draft amendment generation support device 100 is realized in the patent document server 10 by performing data processing according to a draft amendment generation support program. The draft amendment generation support device 100 generates an instruction statement (also called a prompt) 114 used to generate a draft language model 49. The LLM server 400 generates a draft language model 49 according to the generated instruction statement 114, and the draft language model 49 is transmitted from the LLM server 400 to the patent document server 10. In the patent document server 10, the draft amendment generation support device 100 generates a draft amendment 50 using the draft language model 49, and the draft amendment 50 is presented to the user terminal device 30. Draft Amendment 50 is a draft of a procedural amendment document, and is used to generate a procedural amendment document to be submitted in response to a notice of rejection issued when an invention related to a patent application is found to not meet the requirements of inventive step (Article 29, Paragraph 2 of the Patent Act) (deficiency of inventive step).

[0021] The user terminal device 30 receives or transmits data to and from the patent document server 10, and transmits the claims data 45d constituting the claims 45, the specification data 46d constituting the specification 46, and the reference data 47d constituting the cited documents 47 to the patent document server 10. The user terminal device 30 also receives the draft amendment 50 from the patent document server 10.

[0022] (Configuration of patent document server 10) Next, the configuration of the patent document server 10 will be described with reference to Figure 2. Figure 2 is a block diagram mainly showing the internal configuration of the patent document server 10. The patent document server 10 is a server operated by a specialized company that provides services related to the generation of draft amendments.

[0023] The patent document server 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, and a RAM (Random Access Memory) 13. The CPU 11 operates according to the program stored in the ROM 12, receiving input data obtained from keyboard 19 and mouse 20 operations via the KBC (Keyboard controller) 17 through the main bus 19A, while also performing signal input and output with other components to control the overall operation of the patent document server 10. In the patent document server 10, the amendment draft generation support device 100 is realized when the CPU 11 operates according to the amendment draft generation support program. The ROM 12 stores control programs executed by the CPU 11, such as the amendment draft generation support program, and permanent data. The RAM 13 stores data and programs used when the CPU 11 operates.

[0024] In addition, the patent document server 10 includes a hard disk drive (HDD) 14, a communication control unit 15, a communication processing unit 16, and a video controller 18. The patent document server 10 may have an SSD (Solid State Drive) instead of the HDD 14.

[0025] The hard disk drive 14 has various storage units or databases used for executing the amendment draft generation support program. The hard disk drive 14 has a difference data storage unit 104, a provisional correction data storage unit 105, and an effective provisional correction data storage unit 113, which are used for executing the amendment draft generation support program (see Figures 4 and 5 for details of each unit, which will be explained in more detail later). The difference data storage unit 104 and provisional correction data storage unit 105 shown in Figure 4, and the effective provisional correction data storage unit 113 shown in Figure 5 store difference data 95d, provisional correction data 96d, and effective provisional correction data 97d, respectively.

[0026] The communication control unit 15 operates according to the instructions of the CPU 11 and controls the connection and disconnection of lines for communication with the user terminal device 30 and the LLM server 400. The communication processing unit 16 operates according to the instructions of the communication control unit 15 and performs the transmission and reception of data via the Internet N1.

[0027] The video controller 18 controls the image display on a display device (not shown) to display screens and the like used for various settings.

[0028] Claims 45 is a document in which the invention related to the patent application is described claim by claim, and the text data described in each claim corresponds to claim data 45d which constitutes Claims 45. The specification 46 is a document in which matters relating to the invention related to the patent application are described, and it includes the title of the invention, a brief description of the drawings, and a detailed description of the invention. The text data described in the specification 46 corresponds to specification data 46d which constitutes the specification 46.

[0029] Reference document 47 is a prior art document listed in the notice of rejection due to lack of inventive step, and is cited as the basis for the claim that the invention in the patent application does not meet the requirement of inventive step, such as a published patent gazette. Most notices of rejection list multiple documents as "Reference Document 1," "Reference Document 2," etc. In such cases, the prior art document that describes the structure and framework of the invention in the patent application and has common elements (hereinafter also called the "main cited invention") is called the main cited document (also called the primary reference), and the text data that constitutes it corresponds to the reference data 47d.

[0030] (Configuration of the amendment draft generation support device 100) Next, the configuration of the amendment draft generation support device 100 will be described with reference to Figures 4-5 and 10. Figure 4 is a functional block diagram showing the main configuration of the amendment draft generation support device 100 together with the user terminal device 30 and LLM server 400, and Figure 5 is a functional block diagram showing the main configuration of the classification processing unit 103 of the amendment draft generation support device 100. Figure 10 is a functional block diagram showing the main configuration of the inventiveness suitability classification unit 121.

[0031] As shown in Figure 4, the correction draft generation support device 100 includes a difference data detection unit 101, a provisional correction data generation unit 102, a classification processing unit 103, an instruction data generation unit 106, an instruction text generation unit 107, an instruction text presentation unit 108, and a presentation processing unit 109.

[0032] The discrepancy data detection unit 101 detects discrepancy data 95d using the claims data 45d, the specification data 46d, and the reference data 47d, and stores the detected discrepancy data 95d in the discrepancy data storage unit 104. The discrepancy data detection unit 101 will be described in more detail later.

[0033] The provisional correction data generation unit 102 includes a sorting unit 110 and a generation unit 111. In the provisional correction data generation unit 102, provisional correction data 96d is generated by the generation unit 111 using the claim scope data 45d and the difference data 95d.

[0034] The sorting unit 110 performs a sorting process to rearrange the difference data 95d and outputs sorted difference data 110d. In the difference data detection unit 101, the text data of the claims data 45d, specification data 46d, and reference data 47d are separated into sentences to generate sentence data. The similarity of the sentence data of the specification data 46d and the sentence data of the reference data 47d is calculated, and the sentence data of the specification data 46d that meets the difference detection criteria described later is detected as difference data 95d. In many cases, the specification 46 and reference 47 contain multiple sentences, so the difference data 95d also contains multiple sentences. In the sorting unit 110, the difference data 95d is sorted in ascending order using the similarity calculated by the difference data detection unit 101. The sorted difference data 95d corresponds to sorted difference data 110d.

[0035] The generation unit 111 generates provisional correction data 96d using the claim range data 45d and at least one sorted difference data 110d, as will be described in more detail later.

[0036] The difference data 95d contains multiple sentence data. Each of these sentence data is used in order of similarity to generate the provisional correction data 96d. For each generated provisional correction data 96d, the classification processing unit 103 classifies whether or not it meets the requirement of inventive step. If the provisional correction data 96d is classified as not meeting the requirement of inventive step, it is highly likely that the draft amendment using that provisional correction data 96d will not be able to resolve the grounds for rejection (grounds for rejection due to inventive step). Therefore, the draft amendment generation support device 100 prioritizes using the smallest difference data (described later), which has the smallest similarity, in the generation of the draft amendment, as it is considered to be the data with the largest difference and is highly likely to be useful in resolving the grounds for rejection (details will be described later).

[0037] As shown in Figure 5, the classification processing unit 103 includes an effectiveness classification unit 120 and an inventiveness suitability classification unit 121. In the classification processing unit 103, after the effectiveness classification processing described later by the effectiveness classification unit 120 for the provisional correction data 96d, the inventiveness suitability classification unit 121 performs the inventiveness suitability classification processing described later, and generates output data 130d including the effective provisional correction data 97d and classification data 98d described later (see Figure 10).

[0038] The effectiveness classification unit 120 has a trained model 120M, which will be described later. The effectiveness classification unit 120 uses the trained model 120M to perform effectiveness classification processing. Through the effectiveness classification processing, the provisional correction data 96d is classified into an effective group (effective group) and an ineffective group (ineffective group). The provisional correction data 96d classified into the effective group is output as effective provisional correction data 97d and stored in the effective provisional correction data storage unit 113.

[0039] In this embodiment, "effective" means that the grounds for rejection, namely failure to meet the requirement of inventive step, can be resolved. When a procedural amendment is submitted as a response to a notice of rejection stating failure to meet the requirement of inventive step, there are cases where the grounds for rejection are resolved and cases where they are not. It is desirable that the draft amendment 50 presented to the user terminal device 30 corresponds to the former case. In order to extract the data corresponding to the former case from the provisional amendment data 96d by the effectiveness classification process, the effectiveness classification unit 120 has a trained model 120M in this embodiment. Although not shown in the figures, the effectiveness classification unit 120 may have a trained model other than the trained model 120M.

[0040] (Pre-trained model 120M) The trained model 120M can be explained as follows with reference to Figure 9. As shown in Figure 9, for the provisional correction data 96d and the reference data 47d, the provisional correction vector 131a and the reference vector 132a are assumed to be the document vectors corresponding to them. The difference between the provisional correction data 96d and the reference data 47d corresponds to the difference data 133, and the document vector corresponding to that difference data 133 is also assumed to be the difference vector 133a.

[0041] For a rejection due to lack of inventive step to be resolved, there must be a difference between the invention in the patent application and the prior art invention described in the references, and this difference must exceed a certain degree (to the extent that it could have been easily invented). Therefore, it is considered that the difference data 133 must also exceed a certain degree. Using the examination documents of an actual patent application (amendment and cited documents), multiple difference vectors v1 and v2 shown in Figure 9 were prepared as training data, and a training label (for example, "0" if the grounds for rejection were resolved, and "1" if the grounds for rejection were not resolved) was assigned to each difference vector. Machine learning was performed in this manner to construct the trained model 120M. Difference vector v1 corresponds to the difference vector between the invention in the application (amended invention) and the references when a patent grant is notified (grounds for rejection are resolved), and difference vector v2 corresponds to the difference vector between the invention in the application (amended invention) and the references when a patent grant is notified (grounds for rejection were not resolved).

[0042] The effectiveness classification unit 120 calculates the difference vector 133a between each provisional correction data 96d and its corresponding reference data 47d, and then classifies whether the provisional correction data 96d is effective or ineffective based on the output data (in the above case, "0" or "1") when this difference vector is input to the trained model 120M.

[0043] As shown in Figure 10, the inventiveness classification unit 121 includes an input vector generation unit 128, a machine learning unit 129, and a data output unit 130. The input vector generation unit 128 includes a summary data generation unit 140 and a vector generation unit 141. The summary data generation unit 140 receives the valid provisional amendment data 97d and the specification data 46d to generate summary data 46da, and the vector generation unit 141 receives the summary data 46da and the reference data 47d to generate corresponding input vectors, and generates a vector (provisional amendment vector) Vmtx corresponding to the difference between the two input vectors. The summary data is text data that can identify the gist of the invention related to the application, and includes the data relating to each claim in the claims data 45d and the text data described in the problem to be solved column of the specification data 46d. In the case of summary data 46da, it includes the valid provisional amendment data 97d and the text data described in the problem to be solved column of the specification data 46d. The data output unit 130 outputs output data 130d, which includes the classification data 98d output from the machine learning unit 129 along with the effective provisional correction data 97d, to the instruction data generation unit 106.

[0044] (Machine Learning Section 129) The machine learning unit 129 classifies the provisional correction vector Vmtx into one of the following three inventiveness classification classes and outputs classification data 98d according to the classification result. The inventiveness classification classes correspond to a class with a "high" probability of meeting the requirements for inventiveness (no OA class), a class with a "medium" probability (medium OA class), and a class with a "low" probability (with OA class). For example, "A" (no OA class), "B" (medium OA class), and "C" (with OA class) can be used as classification labels to indicate each class. The inventiveness classification unit 121, as will be described in more detail later, classifies the provisional correction vector Vmtx into one of the above three classes and outputs classification data 98d according to the classification result. The classification data 98d is data that shows the classification result regarding the compliance with the requirements for inventiveness.

[0045] When classifying into one of the three inventiveness suitability classification classes, the machine learning unit 129 can be constructed by performing machine learning using training data that includes, for example, three types of learning document vectors (first, second, and third) and a target vector, as shown below. Alternatively, the machine learning unit 129 can be constructed to classify into either "A" (no OA class) or "C" (with OA class), in which case it can be constructed by performing machine learning using training data that includes two types of learning document vectors (first and third) and a target vector.

[0046] The first learning document vector is a first moving document vector corresponding to the difference between the document vector corresponding to claim 1 (at the time the notice of rejection was issued) of an application that, as a result of examination, was granted a patent without a notice of rejection (an application without rejection) or an application that was issued a notice of rejection but did not include a reason for rejection due to lack of inventive step (an application without rejection due to lack of inventive step), and the document vector corresponding to the document with the highest similarity (most similar document for learning) as a result of a conceptual search targeting those applications without rejection or applications without rejection due to lack of inventive step (a non-cited document vector).

[0047] The second learning document vector is a second moving document vector corresponding to the difference between the document vector corresponding to the claim (at the time the rejection notice was issued) for an application that, among the published applications, received its first rejection notice (1st action) as a result of the examination by the Japan Patent Office, and in that 1st action, while no rejection notice for novelty (a rejection reason stating that the requirements of Article 29, Paragraph 1, Item 3 of the Patent Act were met) was pointed out as a rejection reason for infringement of inventive step (a rejection reason stating that the requirements of Paragraph 2 of the same Article were not met) and the document vector corresponding to the cited document 1 (the document cited as the principal publication) at that time.

[0048] The third learning document vector is a third moving document vector corresponding to the difference between the document vector corresponding to the claim (at the time the notice of rejection was issued) for an application that has been published and for which the Japan Patent Office has issued a 1st action as a result of its examination, and for which the 1st action pointed out grounds for rejection for lack of novelty and inventive step (grounds for rejection that the requirements of Article 29, Paragraph 1, Item 3 and Paragraph 2 of the Patent Act were not met) for which the grounds for rejection were pointed out (at the time the notice of rejection was issued), and the document vector corresponding to cited document 1 (the document cited as the main publication) at that time.

[0049] Approximately 300,000 patent applications are filed annually, and first actions have already been issued for a large number of published patent applications. Among these, there are many applications that have been rejected due to a lack of inventive step (applications rejected for lack of inventive step).

[0050] In an application rejected for lack of inventive step, the examination result indicates that although there were differences between the invention claimed at the time of examination and the primary cited invention, the examiner did not determine that these differences alone were sufficient to establish inventive step. In contrast, some patent applications have been granted a patent without the issuance of a first action, and others have been issued a notice of rejection, but the reason for rejection did not include a lack of inventive step (applications without rejection for lack of inventive step).

[0051] If it were possible to determine the extent of the differences that would lead to the affirmation of an invention's inventive step and the extent of the differences that would lead to its denial, this could serve as objective criteria for determining whether the requirements for inventive step are met. To do this, it is considered effective to determine the distance corresponding to the differences between two inventions (the invention in which the lack of inventive step has been pointed out and the main cited invention). The machine learning unit 129 determines this using machine learning based on training data from past examination records and outputs classification data 98d regarding the compliance of the requirements for inventive step.

[0052] As described above, in the patent document server 10, the amendment draft generation support device 100 has a classification processing unit 103, and the classification processing unit 103 is equipped with a machine learning unit 129, so that the amendment draft is generated in a way that reflects the examination record of the Japan Patent Office. This allows the judgment results of artificial intelligence to be utilized where previously it was necessary to rely solely on the experience and intuition of experts such as examiners and patent attorneys, making it possible to bring objectivity to the amendment draft and improving the efficiency of patenting work by reducing the burden of document creation in the intermediate processing.

[0053] Next, as shown in Figure 4, the instruction data generation unit 106 receives the effective provisional correction data 97d and the output data 130d, which includes the classification data 98d, and generates the generation instruction data 114. The instruction data generation unit 106 generates the generation instruction data 114 when the classification data 98d indicates that it is a class that is "highly" likely to meet the requirements for inventive step (classification data 98d: "A"). The generation instruction data 114 includes the effective provisional correction data 97d when the classification data 98d is "A", and the pattern data and form data corresponding to the effective provisional correction data 97d, which will be described later.

[0054] The instruction generation unit 107 generates an instruction 115 using the generation instruction data 114. The instruction 115 includes instruction data indicating that a language model proposal 49 should be generated using the valid provisional correction data 97d. The instruction 115 is sent to the LLM server 400 by the instruction presentation unit 108 so that the language model proposal 49, described later, can be generated. The instruction presentation unit 108 performs the presentation process of the instruction 115 by sending the instruction 115 to the LLM server 400.

[0055] The presentation processing unit 109 generates a revised draft 50 using the language model draft 49 generated by the LLM server 400, and performs the presentation processing of the revised draft 50 by sending the revised draft 50 to the user terminal device 30.

[0056] (Configuration of user terminal device 30) As shown in Figure 1, the user terminal device 30 is equipped with an internet connection environment N1 and can communicate with the patent document server 10. While the user terminal device 30 is assumed to be a stationary (or portable notebook) personal computer, a tablet-type terminal device may also be used.

[0057] As shown in Figure 3, the user terminal device 30 includes a CPU 31, ROM 32, RAM 33, data storage unit 34, and liquid crystal display unit 35. The user terminal device 30 also includes a voice conversion processing unit 36, a communication control unit 37, a communication processing unit 38a, a wireless communication unit 38b, a speaker 39, and a microphone 40.

[0058] The CPU 31 operates according to the program stored in the ROM 32 and controls the operation of the entire user terminal device 30. The ROM 32 stores the program executed by the CPU 31, for example, a communication control program for data communication. The RAM 33 stores data necessary for the CPU 31 to execute the program.

[0059] Various types of data are stored in the data storage unit 34. The liquid crystal display unit 35 has an LCD (Liquid Crystal Display) and its drive unit, and is an image display means that displays images such as characters, figures, and symbols. The voice conversion processing unit 36 ​​decompresses the voice data and outputs it to the speaker 39, while converting and compressing the analog voice signal input from the microphone 40 into digital voice data and inputting it to the communication processing unit 38a. The communication control unit 37 operates under the instructions of the CPU 31 and controls the connection and disconnection of the line for data communication. The communication processing unit 38a operates according to the instructions of the communication control unit 37 and performs data transmission and reception via the Internet N1. The wireless communication unit 38b is a wireless communication means that performs wireless data transmission and reception according to the control of the communication control unit 37. The speaker 39 is an audio output means that outputs sound, and the microphone 40 inputs voice such as the content of the user's conversation and converts it into an electrical signal.

[0060] (LLM Server 400 Configuration) LLM400M is built on LLM Server 400. For example, LLM400M can be built using a large-scale underlying language model (also called the "underlying model," for example, Chat GPT developed by OpenAI). LLM400M is then built by fine-tuning the underlying model using the training dataset shown below, so that language model draft 49 can be generated.

[0061] In this embodiment, a claim dataset is used as the training dataset. The claim dataset contains multiple text data of the claims described in the patent publication corresponding to the registered patent. The claims contain one or more claims. Each claim has a description pattern, and there are multiple patterns (claim patterns) such as "newline enumeration type," "non-newline (plain writing) type," and "plain writing type." The claim dataset contains a large amount of text data of the claims for each claim pattern. The claim dataset includes pattern data corresponding to each claim pattern (for example, "CP1" indicating the newline enumeration type, "CP2" indicating the "non-newline type," etc.). In addition, the text data of the claims before amendment and the claims after amendment are also included in the claim dataset.

[0062] Furthermore, there are two forms for claim description: an independent form that does not refer to other claims, and a dependent form that refers to other claims. A claim dataset is prepared to clearly distinguish between these two forms. The claim dataset includes form data corresponding to the description format (for example, "f0" corresponding to the independent form, "f1" corresponding to the dependent form).

[0063] For example, when using Python as the programming language, the Transformers library (provided by Hugging Face) can be used to achieve fine-tuning. The Transformers library provides an API (Application Programming Interface) and tools for downloading and training a base model from a server (not shown in the diagram). Furthermore, the Dataset library of the Transformers library can be used to prepare the training dataset. By using the Dataset library, natural language processing datasets can be used. The Trainer class can be used to perform tuning. On the LLM server 400, fine-tuning is performed by saving the model obtained using the Trainer class, etc.

[0064] (Operation details of the patent document generation system) Next, referring to Figures 1 to 5, 9 and 10, as well as Figures 6 to 8, 11 and 12, the operation of the patent document server 10 will be explained, mainly focusing on the operation of the amendment draft generation support device 100. Here, Figure 6 shows an example of the claims 45 and claim data 45d, Figure 7 shows an example of the specification 46 and specification data 46d, and Figure 8 shows an example of the cited references 47 and reference data 47d. Figures 11 and 12 show an example of difference data and an example of provisional amendment data, respectively.

[0065] As mentioned above, the claims (assuming the claims of the initial application before any amendments have been filed) 45 contains claims data 45d, and the specification 46 (assuming the specification of the initial application before any amendments have been filed) contains specification data 46d. In addition, the cited document 47 contains reference data 47d. The amendment draft generation support device 100 detects the difference data 95d using the claims data 45d, specification data 46d, and reference data 47d as follows.

[0066] The amendment draft generation support device 100 can use text data entered by the user from the user terminal device 30 as claim data 45d, specification data 46d, and reference data 47d. Alternatively, corresponding document data (for example, PDF (Portable Document Format) files of published patent gazettes) may be transmitted from the user terminal device 30 to the amendment draft generation support device 100, and text data extracted from each of these document data may be used. Furthermore, the amendment draft generation support device 100 may be provided with a search processing unit (not shown), which may search for and obtain the relevant document data using the document number (for example, patent application publication number) transmitted from the user terminal device 30, and text data extracted from the obtained document data may be used.

[0067] As shown in Figures 6 to 8, the claims data 45d, the specification data 46d, and the reference data 47d are text data, but each contains multiple sentence data. Sentence data is data obtained by dividing text data into sentence units. For example, the claims data 45d contains sentence data c1 containing "The base portion and...stand device," and sentence data c2 containing "The cylindrical body is...stand device." The specification data 46d contains sentence data d1 to d11 as shown in Figure 7, and the reference data 47d contains sentence data e1 to e10 as shown in Figure 8. The difference data detection unit 101 takes the sentence data of the claims data 45d (c1, c2 in the case of Figure 6) and the sentence data of the specification data 46d (d1 to d11 in the case of Figure 7) as invention sentence data, and the sentence data e1 to e10 of the reference data 47d as reference sentence data, extracts one of the invention sentence data and one of the reference sentence data, calculates their similarity, and detects the invention sentence data as difference data 95d when the similarity is smaller than the detection criterion value described later.

[0068] In this case, the invention description data includes sentence data c1, c2 and sentence data d1 to d11. In many cases, sentence data corresponding to sentence data c1 and c2 is included in one of sentence data d1 to d11 (in the case of Figure 7, sentence data d4). There are parts, such as sentence data d1, d2, and d3, where it is difficult to find differences from the reference data 47d, or where it is difficult to find appropriate constituent elements to describe in the claims. Therefore, in this embodiment, the sentence data in the "Modes for Carrying Out the Invention" column of the specification data 46d (in the case of Figure 7, d5 to d11) is used to calculate the similarity.

[0069] The difference data detection unit 101 then extracts each of the sentence data d5 to d11 one by one and calculates the similarity (for example, cosine similarity) between each of them and each of the sentence data e1 to e10. For example, it calculates the cosine similarity between sentence data d5 and sentence data e1, the cosine similarity between sentence data d5 and sentence data e2, and so on, for all combinations. In this case, when each sentence data is converted into data of a predetermined number of dimensions by vector transformation (for example, when using Python as the programming language, tfidfVectorizer can be used), each sentence data is placed in a vector space of that number of dimensions. When cosine similarity is calculated, if the relationship between two sentence data is high (high similarity), the vectors corresponding to the two sentence data are placed at positions where the distance between the two points in that vector space is close, and in that case, the cosine similarity is large. Conversely, when the cosine similarity is small, the vectors corresponding to the two sentence data are placed at positions far apart in the vector space. In this case, the relationship between the two sentence data is low (the similarity is low), and the difference between the two sentence data is considered large. Therefore, by prioritizing the use of sentence data with small cosine similarity, sentence data with large differences can be used to generate provisional correction data. For this reason, the difference data detection unit 101 has a preset upper limit for cosine similarity when detecting difference data, and the invention sentence data is detected as difference data when it is smaller than that detection criterion. The "difference detection criterion" according to the present invention can be set as the cosine similarity being smaller than the detection criterion (less than the detection criterion), and sentence data with cosine similarity less than that detection criterion is detected by the difference data detection unit 101 as conforming to the "difference detection criterion". The detected difference data 95d is stored in the difference data storage unit 104. Alternatively, the number of sentence data to be detected may be set as the detection criterion, and conforming to the "difference detection criterion" may be considered as not exceeding that number.

[0070] For example, if the difference data detection unit 101 detects sentence data d9 to d11 as difference data 95d, then the difference data storage unit 104 stores difference data 95d1, d2, and d3, as shown in Figure 11 (in Figure 11, paragraph numbers are shown along with the difference data 95d1, d2, and d3 for illustrative purposes).

[0071] Next, in the provisional correction data generation unit 102, the sorting unit 110 performs a sorting process in which the difference data 95d are sorted in ascending order of similarity (cosine similarity in this embodiment), and outputs sorted difference data 110d. In this way, the difference data 95d are sorted in ascending order of cosine similarity, so sentence data with large differences are given priority for use in generating provisional correction data 96d. The generation unit 111 generates provisional correction data 96d using at least one of the sorted difference data 110d and each sentence data of the claim scope data 45d. For example, for the difference data 95d1, 95d2, and 95d3 in the difference data storage unit 104, let's assume that their respective cosine similarities increase in the order of difference data 95d1, 95d2, and 95d3 (difference data 95d1 < 95d2 < 95d3). In this case, the sorting unit 110 outputs sorted difference data 95d1, 95d2, and 95d3 in order as sorted difference data 110d. These, along with the statement data c1 of the claim scope data 45d, are used by the generation unit 111 to generate the provisional correction data 96d1, 96d2, and 96d3 shown in Figure 12. The generated provisional correction data 96d1, 96d2, and 96d3 are stored in the provisional correction data storage unit 105.

[0072] As shown in Figure 12, the provisional correction data 96d1, 96d2, and 96d3 are generated using sentence data c1 and difference data 95d1, sentence data c1 and difference data 95d1 and 95d2, and sentence data c1 and difference data 95d1, 95d2, and 95d3, respectively. Of the difference data 95d1, 95d2, and 95d3, difference data 95d1 has the smallest similarity, so difference data 95d1 corresponds to the minimum difference data according to the present invention. The generation unit 111 generates provisional correction data 96d1 using the difference data 95d1, which is the minimum difference data, and then generates provisional correction data 96d2 and 96d3 by adding difference data 95d2 and 95d3 one by one to difference data 95d1 (by adding sorted difference data 110d one by one in order of similarity).

[0073] Furthermore, the generation unit 111 refers to the statement data (c1, c2) of the claim data 45d to determine the description pattern and description format of each claim, and generates the provisional correction data 96d by setting (including) pattern data ("CP1", "CP2", ...) and form data ("f0", "f1") according to the result of the determination. In the case of statement data c1, since there is a line break for each requirement, the pattern data is set to "CP1", which indicates a line break enumeration type, and since the text "as described in claim*" is not included, the form data is set to "f0", which indicates an independent format. In the case of statement data c2, since there is no line break for each requirement, the pattern data is set to "CP2", which indicates a non-line break type, and since the text "as described in claim*" is included, the form data is set to "f1", which indicates a dependent format.

[0074] Next, in the classification processing unit 103, the effectiveness classification unit 120 performs effectiveness classification processing on the provisional correction data 96d (96d1 to 96d3) using the trained model 120M, outputs the provisional correction data 96d classified into the effective group as effective provisional correction data 97d, and stores it in the effective provisional correction data storage unit 113. For example, if the effectiveness classification processing classifies provisional correction data 96d1 into the ineffective group and provisional correction data 96d2 and 96d3 into the effective group, then provisional correction data 96d2 and 96d3 are stored in the effective provisional correction data storage unit 113 as effective provisional correction data 97d.

[0075] Then, as shown in Figure 10, in the inventiveness classification unit 121, the abstract data generation unit 140 of the input vector generation unit 128 uses each of the provisional correction data 96d2 and 96d3, and generates abstract data 46da using them and the specification data 46d. Subsequently, the vector generation unit 141 generates a difference vector between the abstract data 46da and the reference data 47d, and outputs the provisional correction vector Vmtx.

[0076] Next, the machine learning unit 129 classifies the provisional correction vector Vmtx corresponding to the provisional correction data 96d2 and 96d3 into one of the inventiveness suitability classification classes, and outputs classification data 98d corresponding to each class. For example, if provisional correction data 96d2 is classified into the moderate OA class (classification label: "B") and provisional correction data 96d3 is classified into the no OA class (classification label: "A"), then classification data 98d containing classification labels "B" and "A" will be output to correspond to provisional correction data 96d2 and 96d3, respectively. Then, the data output unit 130 generates output data 130d by pairing provisional correction data 96d2 with classification label "B" and provisional correction data 96d3 with classification label "A", and outputs the data.

[0077] Furthermore, the instruction data generation unit 106 generates the generation instruction data 114 when the classification label is set to "A" in the output data 130d. In this case, since the classification label "A" is set in the classification data 98d of the provisional correction data 96d3, the provisional correction data 96d3 corresponds to the conformance correction data according to the present invention, and the generation instruction data 114 is generated including this, as well as form data (f0 in the above case) and pattern data (CP1 in the above case) corresponding to the provisional correction data 96d3.

[0078] Then, as shown in Figure 4, the instruction text generation unit 107 generates an instruction text 115 using the generation instruction data 114, and the instruction text presentation unit 108 executes the instruction text presentation process by sending the instruction text 115 to the LLM server 400. Subsequently, when the presentation processing unit 109 receives the language model draft 49 from the LLM server 400, the presentation processing unit 109 generates a revised draft 50 using the language model draft 49 and sends the revised draft 50 to the user terminal device 30 to perform the revised draft presentation process. The presentation processing unit 109 can generate the revised draft 50 by changing the title using the language model draft 49 (for example, changing "language model draft" to "revised draft" as shown in Figures 13(A) and (B)). When the user terminal device 30 receives the revised draft 50, the user can easily create a procedural amendment by transcribing the contents of the revised draft 50 into a procedural amendment document (not shown). As shown in Figures 13(A) and (B), the proposed language model 49 and the proposed amendment 50 include provisional amendment data 96d3 as valid provisional amendment data 97d. This provisional amendment data 96d3 includes sentence data c1 obtained from the claims data 45d and sentence data d9, d10, and d11 obtained from the specification data 46d. Even if a rejection ground for lack of inventive step is pointed out for sentence data c1, sentence data d9, d10, and d11 were detected as data with a large difference from the reference data 47d. Therefore, if provisional amendment data 96d3 includes sentence data d9, d10, and d11 along with sentence data c1, there is a high possibility that the rejection ground can be resolved. Furthermore, provisional amendment data 96d3 is classified by the classification processing unit 103 as valid and conforming to the inventive step requirement. For this reason, by using the proposed amendment 50, a procedural amendment that can resolve the rejection ground for lack of inventive step can be efficiently created.

[0079] (Details of LLM Server 400 operation) Meanwhile, when the LLM server 400 receives the generation instruction data 114, it determines the format of the document to be generated using the claim dataset corresponding to the pattern data and form data contained therein, and then uses the provisional correction data 96d3 as the valid provisional correction data 97d to perform the following text and formatting adjustments to generate the language model proposal 49.

[0080] In this embodiment, text and formatting adjustment means adjusting the text and formatting of the valid provisional correction data 96d7 (provisional correction data 96d3) received (acquired) from the correction document generation support device 100 so that it is in a format corresponding to the form data and pattern data, conforms to the format of the procedural correction document, and conforms to the description requirements of the claim, which state that it should be written in a single sentence. Specifically, the following adjustments p1, p2, p3 and adjustments t1 to t6 are performed.

[0081] Adjustment p1: Write in the format that follows the form data. If the valid provisional correction data 96d7 (provisional correction data 96d3) is determined to be in an independent format, write it in an independent format. Adjustment p2: Write in the format that follows the pattern data. If the valid provisional correction data 96d7 (provisional correction data 96d3) is determined to be in a newline enumeration type, write the valid provisional correction data 96d7 (provisional correction data 96d3) with a newline for each sentence data c1, d9, d10, d11. Adjustment p3: Write it in one sentence. If the valid provisional correction data 96d7 (provisional correction data 96d3) is written in one sentence from "The base part and," in sentence data c1 to "The stand device" in sentence data d11.

[0082] Adjustment t1: Change periods "." in sentences to commas ",". Adjustment t2: Delete the name of the invention from the end of the first sentence data (sentence data c1 in the case of provisional correction data 96d3), and change the preceding verb from the terminal form to the conjunctive form (g1 in Figures 13(A), (B)). Adjustment t3: Delete Arabic numerals (for example, "12" in "LED chip 12"). Adjustment t4: Underline parts other than the first sentence data (sentence data d9, d10, d11 in the case of provisional correction data 96d3). Adjustment t5: If the end of sentence data other than the first sentence data (sentence data d9, d10, d11 in the case of provisional correction data 96d3) is a verb, change that verb from the terminal form to the conjunctive form, and add a comma "," after it (g2, g3 in Figures 13(A), (B)). Adjustment t6: Add the name of the invention to the end (g4 in Figures 13(A) and (B)). When performing t2 and t5, use the part-of-speech data obtained by performing morphological analysis.

[0083] The language model draft 49 is then transmitted from the LLM server 400 to the revised draft generation support device 100 (see Figures 1 and 4). After receiving the language model draft 49, the presentation processing unit 109 generates the revised draft 50 by making changes to the title and other functions, and presents it to the user terminal device 30.

[0084] As described above, in the patent document generation system 1, the language model draft 49 and the amendment draft 50 are generated using the effective provisional amendment data 97d (provisional amendment data 96d3) obtained by the amendment draft generation support device 100 of the patent document server 10. The text data described in the language model draft 49 and the amendment draft 50 uses the effective provisional amendment data 97d (provisional amendment data 96d3). This effective provisional amendment data 97d (provisional amendment data 96d3) is conforming amendment data, which is effective data from the provisional amendment data 96d including the difference data 95d, and moreover, it is data that has been classified by the inventiveness suitability classification unit 121 as conforming to the requirements of inventiveness. Therefore, the language model draft 49 and the amendment draft 50 contain text data that is effective in resolving the grounds for rejection due to lack of inventiveness and that has been deemed to conform to the requirements of inventiveness. Therefore, by using the proposed amendment 50, users can create a procedural amendment that can resolve the grounds for rejection due to lack of inventive step (by making adjustments to terminology as needed, such as changing "ultraviolet LED module" to "ultraviolet generating unit"), and the burden of preparing the documents required for the response procedure (procedural amendment) can be greatly reduced.

[0085] In practice, when a rejection notice is issued due to lack of inventive step, the response procedure places great emphasis on examining the contents of the application documents and the contents of the cited documents, and identifying the differences between the two from the information contained in the application documents. This work requires practical experience and considerable effort. In this respect, the amendment draft generation support device 100 of the present invention detects difference data by the difference data detection unit 101, so the work of identifying the differences among the above-mentioned tasks is greatly reduced in effort.

[0086] Furthermore, since the proposed language model 49 and the proposed amendment 50 are generated with text and formatting adjustments, they follow the format of the original claims before amendment, conform to the claim description guidelines such as writing in a single sentence, and are also generated in accordance with the format required for procedural amendments (underlining the corrected parts). Therefore, by using the proposed amendment 50, users can reduce the effort required to create procedural amendments and significantly reduce the burden on the response procedure.

[0087] Furthermore, in the proposed language model 49 and the proposed amendment 50, sentence data d9, d10, and d11 are added to the claim data at the time of filing (sentence data c1 in the case of the valid provisional amendment data 97d (provisional amendment data 96d3)), but these sentence data d9, d10, and d11 are generated from the specification data 46d that constitutes the original specification. Since the proposed language model 49 and the proposed amendment 50 are generated using the matters described in the original specification, no new matters are added, and therefore, the user can make a lawful amendment that conforms to the requirements for procedural amendment (Article 17-2, Paragraph 3 of the Patent Act) by using the proposed amendment 50.

[0088] Second Embodiment (Overall configuration of the patent document generation system) A patent document generation system 201 including a draft amendment generation support device 200 according to a second embodiment of the present invention will be described with reference to Figures 14 and 15. Figure 14 is a system configuration diagram of the patent document generation system 201 including the draft amendment generation support device 200, and Figure 15 is a functional block diagram showing the main configuration of the draft amendment generation support device 200.

[0089] Unlike the aforementioned patent document generation system 1, the patent document generation system 201 provides support for generating draft amendments without using the LLM server 400. The patent document generation system 201 differs from the aforementioned patent document generation system 1 in that it does not use the LLM server 400 and has a patent document server 210 instead of the patent document server 10. The patent document server 210 differs from the patent document server 10 in that it has a draft amendment generation support device 200 instead of the draft amendment generation support device 100.

[0090] In the aforementioned patent document generation system 1, the LLM server 400 is used to assist in the generation of the draft amendment 50. Therefore, an operational burden is incurred in operating the LLM server 400. In this regard, the patent document generation system 201 is configured to reduce the operational burden of the LLM server 400 by enabling the generation of the draft amendment 51 without using the LLM server 400.

[0091] As shown in Figure 15, the revised draft generation support device 200 differs from the revised draft generation support device 100 in that it has a revised draft generation adjustment unit 220 instead of an instruction data generation unit 106, an instruction text generation unit 107, and an instruction text presentation unit 108, and it has a presentation processing unit 209 instead of a presentation processing unit 109.

[0092] The draft amendment generation and adjustment unit 220 generates the draft amendment 51 using the valid provisional amendment data 97d of the output data 130d. In the aforementioned patent document generation system 1, text and formatting adjustments were performed by the LLM server 400, but in this case, the draft amendment generation and adjustment unit 220 performs the text and formatting adjustments instead of the LLM server 400. Also, similar to the instruction data generation unit 106 of the aforementioned patent document generation system 1, the draft amendment generation and adjustment unit 220 performs the draft amendment generation and adjustment process when the classification label in the output data 130d is set to "A".

[0093] The presentation processing unit 209 performs the presentation processing of the revised draft 51 generated by the revised draft generation and adjustment unit 220 without making any changes to the title as described in the presentation processing unit 109 above.

[0094] The amendment draft generation and adjustment unit 220 may use a claim dataset such as that of the LLM server 400, or it may use a dataset similar to the claim dataset. The amendment draft generation and adjustment unit 220 performs the aforementioned adjustments p1, p2, and p3 in accordance with the pattern data and form data of the valid provisional amendment data 97d, similar to the LLM server 400, and further performs adjustments t1 to t6.

[0095] Furthermore, the Amendment Draft Generation and Adjustment Unit 220 performs an Amendment Draft Format Adjustment to make the format more suitable for a procedural amendment than the Amendment Draft 50, generating an Amendment Draft 51 as shown in Figure 16. In this Amendment Draft Format Adjustment process, a submission form 51s is added to the top of the Amendment Draft Display Unit 51A. The Amendment Draft Display Unit 51A is generated using the valid provisional amendment data 96d7 (for example, the aforementioned provisional amendment data 96d3), similar to the Amendment Draft 50, but the submission form 51s is added to the top of it. The submission form 51s has several items such as "Document Name," "Submission Date," "Application Number," and "Person Making the Amendment." These can be fixed displays ("Document Name: Procedural Amendment," etc.), system dates, or generated using data (such as examiner name and application number) that the user has previously entered into the patent document server 210 from the user terminal device 30. This submission form 51s contains bibliographic information, but it includes items that are essential for the format of a procedural amendment. Since the draft amendment 51 is generated in a format that includes this, its conformity to the procedural amendment is higher than that of the draft amendment 50. Therefore, the burden of preparing the procedural amendment can be further reduced by the support for generating the draft amendment 51. The draft amendment 51 is also generated by the amendment generation support device 200, so, as with the amendment generation support device 100, the burden of preparing the documents required for the response procedure (procedural amendment) can be greatly reduced, and furthermore, lawful amendments can be made that conform to the requirements for procedural amendment (Article 17-2, Paragraph 3 of the Patent Act).

[0096] (modified version) In the embodiments described above, the example is given in which the patent document servers 10 and 210 function as amendment draft generation support devices 100 and 200 by installing the amendment draft generation support program on the patent document servers 10 and 210. In addition, the present invention also applies to cases in which the user terminal device 30 functions as an amendment draft generation support device. In this case, the aforementioned amendment draft generation support program should be modified in accordance with at least the following modification 1), the modified amendment draft generation support program should be downloaded from the patent document servers 10 and 210 to the user terminal device 30, and installed on the user terminal device 30.

[0097] Change 1) The proposed language model 49 and the proposed amendments 50 and 51 are displayed and output on the user terminal device 30 without being received from the patent document server 10.

[0098] The above description concerns embodiments of the present invention and does not limit the apparatus and method of this invention, and various modifications can be easily implemented. Furthermore, apparatuses or methods configured by appropriately combining the components, functions, features, or method steps of each embodiment are also included in the present invention.

[0099] For example, the user terminal device does not have to be a high-function mobile phone or a tablet device; it could be a laptop or a PDA. The correction draft generation support program executed by CPU 11 can be recorded on various recording media such as magnetic recording media, CD-ROMs, and DVDs, or it can be downloaded from a server (not shown) via a network. [Industrial applicability]

[0100] By applying the present invention, it is possible to support the generation of a draft amendment document as a draft of a procedural amendment document, so as to resolve the grounds for rejection due to lack of inventive step. The present invention can be used in the field of an amendment document draft generation support device and an amendment document draft generation support program. [Explanation of Symbols]

[0101] 1,201…Patent document generation system, 10,210…Patent document server, 11,31…CPU, 30…User terminal device, 46…Specification, 46d…Specification data, 47…References, 47d…Reference data, 46da…Abstract data, 49…Language model proposal, 50,51…Amendment proposal, 95d…Difference data, 96d…Provisional amendment data, 97d…Valid provisional amendment data, 98d…Classification data, 100,200…Supplement Orthographic draft generation support device, 101... Difference data detection unit, 102... Provisional correction data generation unit, 103... Classification processing unit, 107... Instruction text generation unit, 108... Instruction text presentation unit, 109, 209... Presentation processing unit, 110... Sorting unit, 110d... Sorted difference data, 114... Instruction text, 120... Effectiveness classification unit, 121... Inventiveness suitability classification unit, 129... Machine learning unit, 220... Correction draft generation adjustment unit, 400... LLM server.

Claims

1. A draft amendment generation support device that assists in generating a draft amendment to the claims of a patent application that describes an invention related to the patent application, A difference data detection unit detects data that differs from the cited reference data in a notice of rejection that states that the claim data constituting the claims or the specification data constituting the description of the invention relating to the patent application differs from the cited reference data that constitutes the cited references in the notice of rejection which states that the requirements for inventive step are not met. A provisional correction data generation unit that generates provisional correction data using the claim scope data and the difference data, A classification processing unit performs classification processing on whether the provisional correction data meets the requirements for inventive step with respect to the provisional correction data, using the provisional correction data and the reference data. A draft amendment generation support device having a presentation processing unit that presents a draft amendment generated using conforming correction data that shows that the result of the classification processing among the provisional correction data conforms to the requirement of inventive step.

2. The classification processing unit includes an effectiveness classification unit that classifies the provisional correction data on whether or not it is effective in resolving the grounds for rejection, and an inventive step suitability classification unit that performs the classification processing using the effective provisional correction data that has been classified as effective by the effectiveness classification unit and the cited data. The difference data detection unit has a difference detection criterion set for detecting the difference data. The difference data detection unit detects the invention text data as difference data when the similarity between the claim data or specification data divided into sentence units and the reference text data divided into sentence units is smaller than the detection criterion value, and the invention text data conforms to the difference detection criterion. The provisional correction data generation unit includes a sorting unit that sorts the difference data in ascending order of similarity, and generates the provisional correction data using at least one of the sorted difference data sorted by the sorting unit. The draft correction document generation support device according to claim 1, wherein the provisional correction data generation unit generates the provisional correction data using the smallest difference data with the smallest similarity among the sorted difference data, and then generates the provisional correction data by adding the sorted difference data one by one to the smallest difference data in order of similarity.

3. The system further includes an instruction generation unit that generates an instruction sentence that instructs a large-scale language model to generate the draft correction document using the aforementioned conformance correction data, The amendment draft generation support device according to claim 1 or 2, wherein the presentation processing unit presents the amendment draft using the language model draft generated by the large-scale language model in accordance with the instruction statement.

4. The amendment draft generation support device according to claim 1 or 2, further comprising an amendment draft generation adjustment unit that generates the amendment draft by performing text and formatting adjustment using the conformity correction data and amendment draft formatting adjustment to adjust the format to conform to the procedural amendment document.

5. A draft amendment generation support program for causing a computer to function as a draft amendment generation support device that assists in generating a draft amendment to the claims of a patent application which describes an invention related to a patent application, wherein the computer A difference data detection unit detects data that differs from the cited reference data in a notice of rejection that states that the claim data constituting the claims or the specification data constituting the description of the invention relating to the patent application differs from the cited reference data that constitutes the cited references in the notice of rejection which states that the requirements for inventive step are not met. A provisional correction data generation unit that generates provisional correction data using the claim scope data and the difference data, A classification processing unit performs classification processing on whether the provisional correction data meets the requirements for inventive step with respect to the provisional correction data, using the provisional correction data and the reference data. A draft amendment generation support program that functions as a presentation processing unit that presents a draft amendment generated using conforming amendment data that shows that the result of the classification processing among the provisional amendment data conforms to the requirement of inventive step.

Images