Information processing device, information processing method, program, and system

Abstract

To provide useful review assistance without imposing a burden on reviewers of legal documents. [Solution] An information processing device comprising: a reception unit that receives at least input of legal text; a generation unit that generates prompts that instruct the extraction of corresponding description sections for each review item from the legal text based on review item information stored in a memory unit that describes review items to be reviewed for the legal text; an extraction unit that extracts corresponding description sections for each review item by inputting the generated prompts into a trained model; and an output unit that associates the extracted description sections with each review item and outputs them to a terminal device.

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, a program, and a system. [Background technology]

[0002] Conventionally, there are known techniques for assisting the review of legal documents. For example, Patent Document 1 describes a technique for evaluating legal documents based on predefined evaluation rules and outputting the evaluation results to a user terminal. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-184251 Summary of the Invention [Problem to be solved by the invention]

[0004] However, as described above, the conventional techniques evaluate legal documents based on rules, and in order to obtain useful review assistance, detailed rules must be set, which can be a burden for reviewers. Furthermore, when evaluating legal documents based on rules, it is difficult to evaluate items that are not specified in the rules, and this can be unhelpful for reviewers.

[0005] The present invention has been made in consideration of the above circumstances, and one of its objectives is to provide an information processing device, an information processing method, and a program that can provide useful review assistance without imposing a burden on reviewers of legal documents. [Means for solving the problem]

[0006] One aspect of the present invention is an information processing device comprising: a reception unit that receives at least input of legal text; a generation unit that generates a prompt that instructs the extraction of corresponding description passages for each review item from the legal text based on review item information stored in a memory unit that describes review items to be reviewed for the legal text; an extraction unit that extracts corresponding description passages for each review item by inputting the generated prompt into a trained model; and an output unit that associates the extracted description passages with each review item and outputs them to a terminal device. [Effects of the Invention]

[0007] According to one aspect of the present invention, it is possible to provide an information processing device, an information processing method, a program, and a system that can provide useful review assistance without imposing a burden on reviewers of legal documents. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a diagram illustrating an example of a usage environment and configuration of an information processing apparatus 100 according to an embodiment of the present invention. [Figure 2] FIG. 2 is a diagram showing an example of a legal document 22. [Figure 3] FIG. 2 is a diagram showing an example of review item information 24. [Figure 4] FIG. 10 is a diagram showing an example of a display mode of review item information 24 by spreadsheet software 12. [Figure 5] FIG. 10 is a diagram showing an example of a prompt template 152. [Figure 6] FIG. 10 is a diagram showing an example of a prompt 154. [Figure 7] FIG. 2 is a diagram showing an example of a review result displayed by the terminal device 10. [Figure 8] 1 is a sequence diagram showing an example of a flow of processing executed by the terminal device 10, the information processing device 100, and the generation AI server 200 in cooperation with each other. [Figure 9]FIG. 10 is a diagram showing an example of a prompt template 152 according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0009] [Overall configuration] Hereinafter, an information processing device, an information processing method, and a program according to an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a diagram showing an example of the usage environment and configuration of an information processing device 100 according to an embodiment of the present invention. The information processing device 100 operates in cooperation with, for example, a terminal device 10 and a generation AI server 200.

[0010] The terminal device 10 is a computer device such as a personal computer, a smartphone, or a tablet terminal. The terminal device 10 is equipped with, for example, spreadsheet software 12 for tabulating, analyzing, and automatically calculating data, and the spreadsheet software 12 provides functions for displaying and editing information such as legal documents 22 and review item information 24 stored in a storage unit 20. The legal documents 22 are legal documents to be reviewed. The review item information 24, as will be described later, is definition information that describes review items to be reviewed for the legal documents 22. The storage unit 20 is, for example, a hard disk drive (HDD), a flash memory, or a random access memory (RAM). The storage unit 20 may be a network-attached storage (NAS) device that the terminal device 10 can access via a network.

[0011] The terminal device 10 communicates with the information processing device 100 via the network NW, and for example, transmits the legal text 22 and review item information 24 stored in the storage unit 20 to the information processing device 100 via the network NW, and requests a review of the legal text 22 based on the review item information 24. The terminal device 10 then receives the review results of the legal text 22 from the information processing device 100, and displays them, for example, on a spreadsheet program 12.

[0012] The information processing device 100 is a server device such as a web server. The information processing device 100 receives input of a legal document 22 and review item information 24 from the terminal device 10, and generates a prompt that instructs the device to extract, based on the review item information 24, a passage corresponding to each review item described in the review item information 24 from the legal document 22. After generating the prompt, the information processing device 100 transmits the generated prompt to the generation AI server 200 and obtains an answer.

[0013] The generation AI server 200 is a server device such as a web server. The generation AI server 200 is equipped with a large language model (LLM) that has been trained to receive prompts as text from the information processing device 100 and generate answers as text in response to the received prompts. In this embodiment, the generation AI server 200 is installed outside the terminal device 10 and the information processing device 100, but the present invention is not limited to such a configuration, and the generation AI server 200 may be part of the functions of the information processing device 100. The large language model equipped in the generation AI server 200 is an example of a "trained model" in the claims.

[0014] [Function details] The information processing device 100 includes, for example, a receiving unit 110, a generating unit 120, an extracting unit 130, and an output unit 140. The components such as the receiving unit 110, the generating unit 120, the extracting unit 130, and the output unit 140 are realized by, for example, a hardware processor such as a CPU executing a program (software). Some or all of these components may be realized by hardware (including circuitry) such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a GPU (Graphics Processing Unit), or may be realized by a combination of software and hardware. The program may be stored in advance in a storage device (a storage device having a non-transitory storage medium) such as an HDD (Hard Disk Drive) or flash memory, or may be stored in a removable storage medium (a non-transitory storage medium) such as a DVD or CD-ROM, and installed in the storage device by inserting the storage medium into a drive device. The storage unit 170 stores information such as the prompt template 152 and the prompt 154. The storage unit 170 is a hard disk drive (HDD), a flash memory, a random access memory (RAM), etc. The storage unit 170 may be a network attached storage (NAS) device that can be accessed by the terminal device 10 via a network.

[0015] The reception unit 110 receives input of the legal text 22 and the review item information 24 from the terminal device 10 and temporarily stores them in the storage unit 150. At this time, the reception unit 110 only needs to receive input of at least the legal text 22, and the review item information 24 may be stored in the storage unit 150 in advance.

[0016] 2 is a diagram showing an example of legal document 22. As shown in FIG. 2, in this embodiment, legal document 22 is, as an example, a data processing agreement (DPA) concluded between a software vendor that provides software for use as a cloud-based application in response to external access and its user. However, legal document 22 in the present invention is not limited to a data processing agreement, and may be legal document 22 in any format, such as a sales contract, lease agreement, contract for work, power of attorney agreement, non-disclosure agreement, employment contract, or license agreement.

[0017] FIG. 3 is a diagram showing an example of review item information 24. As shown in FIG. 3, in this embodiment, the review item information 24 is information representing review items that are predefined for the legal document 22, which is a data processing agreement. The review item information 24 includes, for example, review items, review subitems, and alternative expressions (prompts) for the review subitems. The review items are information that prescribe the aspects to be reviewed for the legal document 22. For example, if the legal document 22 is a data processing agreement, the aspects to be reviewed include prohibition of use for purposes other than those intended, safety control measures, employee supervision, and leak reporting.

[0018] Review subitems are information that specifies the perspectives that should be further subdivided when a review item is reviewed. For example, in FIG. 3, for the review item "Leak Report," "Leak Report" can be further subdivided into perspectives such as "Reporting Obligation in the Event of a Leak" and "Obligation and Responsibility for Accident Response." By predefining one or more review subitems for a review item in this way, the legal document 22 can be reviewed more appropriately without missing any perspectives. Alternative expressions (prompts) for review subitems are alternative expressions for the review subitems prepared by the reviewer to prompt the generation AI server 200 to generate more appropriate answers. Alternative expressions (prompts) for review subitems are not limited to paraphrases of the review subitems and may include, for example, elaboration or specificity of the content of the review subitems.

[0019] 3 is merely an example, and the review item information 24 may include only review items and alternative expressions (for prompts) of the review items without including review subitems, or may include only review items and review subitems without including alternative expressions. In the latter case, the review subitems are used as they are to generate prompts. Alternatively, for example, the review item information 24 may include only review items, and the review items may be used as they are to generate prompts.

[0020] FIG. 4 is a diagram illustrating an example of a display mode of the review item information 24 by the spreadsheet software 12 of the terminal device 10. FIG. 4 illustrates a scene in which the spreadsheet software 12 is displaying the review item information 24 shown in FIG. 3 on a spreadsheet. The reviewer can add and edit each review item in the review item information 24 on the spreadsheet as needed. Furthermore, in FIG. 4, the symbol B1 represents a button provided for requesting the execution of a review of the legal document 22. That is, when the reviewer presses the review execution button B1, the spreadsheet software 12 transmits the review item information 24 displayed on the spreadsheet together with the legal document 22 to the information processing device 100, requesting the execution of a review. Note that, while FIG. 4 exemplarily illustrates an interface provided as the review execution button B1 for requesting the execution of a review, the present invention is not limited to such a configuration, and a request may also be made from the settings screen of the spreadsheet software 12.

[0021] The generation unit 120 generates a prompt that instructs the user to extract a corresponding passage for each review item from the legal text 22 based on the review item information 24. More specifically, the generation unit 120 generates a prompt 154 by combining the review item information 24 with a prompt template 152 stored in the storage unit 150.

[0022] Fig. 5 is a diagram showing an example of the prompt template 152. As shown in Fig. 5, the prompt template 152 includes, for example, an area A1 showing a review request outline, an area A2 showing review request details, an area A3 showing supplemental information for the review, an area A4 showing an example of feedback for the review, an area A5 showing instructions for outputting the review, and an area A6 showing the context (perspective) of the review.

[0023] Area A1 shows the review request summary. For example, area A1 includes information such as the role specification for the generation AI server 200 to act as a legal department member checking legal documents, to perform checks and provide feedback according to internal company standards (corresponding to the review subitems (prompts) in Figure 3) linked to area A6 (review summary specification), and to satisfy the constraints described below (constraint specification). This allows the generation AI server 200 to improve its accuracy in functioning appropriately as a reviewer of legal documents 22.

[0024] Area A2 shows the review request details. For example, area A2 may respond "no description" if the legal document 22 does not contain any statements that meet the company's standards, but may also include information such as a specification to extract the relevant passages (review details specification) if the legal document 22 contains any statements that meet the company's standards. This allows the generation AI server 200 to output a response that precisely matches the reviewer's purpose.

[0025] Area A3 indicates supplementary information for the review. For example, area A3 includes information such as outputting the item number and item title without processing if the legal document 22 contains a statement that meets company standards (specification of constraint details), and translating the extracted relevant passage into Japanese and outputting the translation result together (specification of supplementary information) if the extracted relevant passage is in a language other than Japanese. This allows the generation AI server 200 to output an answer that is highly convenient for the reviewer.

[0026] Area A4 shows example feedback sentences for reviews. The example feedback sentences in area A4 shown in Figure 5 represent example feedback sentences that satisfy the conditions specified in areas A2 and A3. In this way, by not only specifying the conditions in areas A2 and A3 but also providing example feedback sentences in area A4, it is possible to reduce the possibility that the generation AI server 200 will misinterpret the conditions and output an answer that does not meet the reviewer's intentions.

[0027] Area A5 indicates a review output instruction. For example, area A5 includes information such as a specification to output a review (output specification) based on the feedback sentence example in area A4. By specifying conditions in areas A2 to A4 and providing additional output instructions, the generation AI server 200 can increase the likelihood of outputting a response that is in line with the reviewer's intentions.

[0028] Area A6 indicates the context (perspective) of the review. As described above, the context of the review is specified in advance as an alternative expression (prompt) for the review subitem in the review item information 24. Therefore, the generation unit 120 combines the content of the review item information 24 with area A6 of the prompt template 152 to obtain the prompt 154 to be provided to the generation AI server 200.

[0029] 6 is a diagram showing an example of the prompt 154. As shown in area A6 in FIG. 6, the generation unit 120 obtains the prompt 154 by combining the content of an alternative expression (for the prompt) of the review subitem of the review item information 24 into area A6, which was blank in the prompt template 152. At this time, the generation unit 120 may further include the legal text 22 itself to be reviewed in the prompt 154.

[0030] In this way, when the generation unit 120 generates the prompt 154, the extraction unit 130 inputs the generated prompt 154 to the generation AI server 200 via the network NW and obtains the answer output from the generation AI server 200. If the prompt 154 includes the legal text 22 itself, the extraction unit 130 inputs only the prompt 154 to the generation AI server 200, whereas if the prompt 154 does not include the legal text 22, the extraction unit 130 inputs the legal text 22 to the generation AI server 200 as additional data for the prompt 154.

[0031] The answers acquired in this manner are extracted by associating the corresponding description (or "no description") for each review subcategory. The output unit 140 outputs the acquired answers to the terminal device 10 via the network NW, and the terminal device 10, for example, reads the output answers using the spreadsheet software 12, develops them on a spreadsheet, and displays them as review results. Therefore, the reviewer does not need to use a different app to review the legal document 22, and can instruct the execution of the review and view the review results on the existing spreadsheet software 12.

[0032] FIG. 7 is a diagram showing an example of a review result displayed by the terminal device 10. The review result shown in FIG. 7 is displayed when, for example, the reviewer presses the review execution button B1 shown in FIG. 4, the terminal device 10 sends a review request to the information processing device 100, and then receives the review result from the information processing device 100. As described above, the terminal device 10 acquires the description portion (or "no description") corresponding to each review subcategory from the information processing device 100 as the review result. Therefore, as shown in FIG. 7, the terminal device 10 displays the description portion (or "no description") for each review subcategory on the spreadsheet. Furthermore, as described above, the prompt 154 in FIG. 6 requests the addition of a Japanese translation in the area A3, and therefore the terminal device 10 also displays the Japanese translation of the description portion.

[0033] This allows a reviewer of the legal document 22 to determine whether or not a statement corresponding to the specified viewpoint exists simply by specifying the viewpoint (subitem) of the review, without having to visually check the legal document 22. In particular, if the legal document 22 is a data processing agreement concluded with a software vendor that provides software use as a cloud-based application in response to external access, there is little room for the user to make changes. Therefore, it may be sufficient for the reviewer to simply check whether or not a statement corresponding to the specified viewpoint exists in the data processing agreement. Furthermore, the chapter, article, or paragraph in which each item in the legal document 22 is described typically varies depending on the counterparty (software vendor). Therefore, the reviewer faces a heavy burden of checking, especially when checking legal document 22 for the first time, and there is also a risk of overlooking an item to be checked. According to the review support method of this embodiment, regardless of the order in which the items to be checked are listed in the legal document, a review result such as that shown in FIG. 7 can be obtained in the order of review items specified by the user (reviewer). In other words, the review support method for legal text 22 according to this embodiment has the effect of being able to sort the items to be checked in legal text 22 in a desired order based on review item information 24. As described above, this embodiment can significantly reduce the burden on reviewers who do not need to review legal text 22 in detail.

[0034] Furthermore, as shown in Figure 7, the review item information 24 and the review results are displayed side by side by the spreadsheet software 12. Therefore, when viewing the review results, the reviewer can update the review results by editing the alternative wording (prompt) for any review sub-item that the reviewer feels did not convey their intention to the generation AI server 200 and pressing the Review Execution button B1 again. For example, by adding additional information to the alternative wording (prompt) for the review sub-item, the reviewer can improve the accuracy of the answer output by the generation AI server 200 (prompt extension).

[0035] [Processing flow] Next, a description will be given of the flow of processing executed by the terminal device 10, the information processing device 100, and the generation AI server 200 in cooperation with each other with reference to Fig. 8. Fig. 8 is a sequence diagram showing an example of the flow of processing executed by the terminal device 10, the information processing device 100, and the generation AI server 200 in cooperation with each other.

[0036] First, the reviewer presses a review execution button on, for example, a spreadsheet program on the terminal device 10 (S10). When the terminal device 10 accepts the pressing of the review execution button, it transmits the legal text 22 and the review item information 24 to the information processing device 100, requesting a review of the legal text 22 (S12). When the information processing device 100 receives the review request, it generates a prompt 154 based on the review item information 24 and the prompt template 152 (S14).

[0037] Next, the information processing device 100 inputs the generated prompt 154 to the generation AI server 200 (S16). The generation AI server 200 generates an answer based on the input prompt 154 (S18) and transmits the generated answer to the information processing device 100 (S20). The information processing device 100 transmits the received answer to the terminal device 10 (S22), and the terminal device 10 displays the received answer on, for example, a spreadsheet program. This completes the processing of this sequence diagram.

[0038] [Variations] In the above embodiment, prompt template 152 was configured to respond "no description" if legal document 22 does not contain any description that meets the company standards, but to extract the relevant passage if legal document 22 contains any description that meets the company standards. However, the present invention is not limited to this configuration, and prompt template 152 may be configured to not only extract the relevant passage if legal document 22 contains any description that meets the company standards, but also to output the completeness and comprehensiveness of the extracted relevant passage. This allows the reviewer to prioritize checking the relevant passages that are less complete or comprehensive, thereby improving the accuracy of the review.

[0039] FIG. 9 is a diagram showing an example of a prompt template 152 according to a modified example. The contents of areas A1 and A3 of prompt template 152 shown in FIG. 9 are the same as the contents of areas A1 and A3 of prompt template 152 shown in FIG. 5, and therefore description thereof will be omitted. As shown in FIG. 9, area A2 of prompt template 152 includes a specification to respond "no description" if legal document 22 does not contain a description that satisfies the company's internal standards, and to extract the relevant passage if legal document 22 contains a description that satisfies the company's internal standards and output a score value (in FIG. 9, as an example, a five-level rating from 1 to 5) of the completeness of the extracted relevant passage. Area A4 of prompt template 152 includes a description regarding the completeness of the description (in FIG. 9, completeness of description: 2) as an example of a feedback message.

[0040] Furthermore, area A2 includes a specification for attaching a sample contract corresponding to each completeness level. In this case, the prompt template 152 may include the original description of the sample contract corresponding to each completeness level, or the generation unit 120 may include the description of the sample contract corresponding to each completeness level in the prompt 154 when generating the prompt 154 from the prompt template 152, or the extraction unit 130 may input the sample contract corresponding to each completeness level to the generation AI server 200 as additional data when inputting the prompt 154 to the generation AI server 200. In this way, not only is it specified that the completeness of the description of the corresponding section be displayed using a score value, but sample contract data is also provided to the generation AI server 200, thereby improving the accuracy of the answer output by the generation AI server 200.

[0041] According to the above-described embodiment, an information processing device, an information processing method, a program, and a system are provided that can provide useful review assistance to reviewers of legal documents without imposing a burden on them. The device receives input of at least legal documents, generates prompts that instruct the device to extract corresponding passages from the legal documents for each review item based on review item information stored in a storage unit and describing review items to be reviewed for the legal documents, and inputs the generated prompts into a trained model to extract corresponding passages for each review item. The device then outputs the extracted passages in association with each review item. This provides an information processing device, an information processing method, a program, and a system that can provide useful review assistance to reviewers of legal documents without imposing a burden on them.

[0042] The above describes the form for carrying out the present invention using an embodiment, but the present invention is not limited to such an embodiment, and various modifications and substitutions can be made within the scope that does not deviate from the gist of the present invention. [Explanation of symbols]

[0043] 10 Terminal Equipment 12 Spreadsheet software 20 Memory section 22 Legal texts 24 Review Item Information 100 Information processing device 110 Reception 120 Generation part 130 Extraction part 140 Output section 150 Storage section 152 prompt templates 154 prompts 200 Generation AI Server

Claims

1. a reception unit that receives input of at least legal documents; a generating unit that generates a prompt that instructs the extraction of a description portion corresponding to each review item from the legal text based on review item information stored in a storage unit and describing review items whose existence should be confirmed in the legal text; an extraction unit that extracts a corresponding description for each review item by inputting the generated prompt into a trained model; an output unit that outputs the extracted description portions to a terminal device in association with each review item; The extraction unit extracts a translation result of the description together with the description from the trained model, the output unit outputs the description and the translation result of the description to the terminal device in association with each review item. Information processing device.

2. The review item information includes review subitems, which are detailed contents whose existence should be confirmed by subdividing the review items into viewpoints for reviewing the legal document, and the generating unit generates a prompt instructing extraction of a corresponding description portion for each of the review subitems. The information processing device according to claim 1 .

3. The review item information includes alternative expressions for the review subitems, the generating unit generates a prompt instructing the user to extract a corresponding description portion for each of the alternative expressions. The information processing device according to claim 2 .

4. a reception unit that receives input of at least legal documents; a generating unit that generates a prompt that instructs the extraction of a description portion corresponding to each review item from the legal text based on review item information stored in a storage unit and describing review items whose existence should be confirmed in the legal text; an extraction unit that extracts a corresponding description for each review item by inputting the generated prompt into a trained model; an output unit that outputs the extracted description portions to a terminal device in association with each review item; The extraction unit extracts, from the trained model, a score value that indicates the completeness of the description together with the description, the output unit associates the description with the score value of the description for each review item and outputs the score value to the terminal device. Information processing device.

5. When there is no description in the legal text corresponding to the review item, the output unit outputs information indicating that there is no description in the legal text corresponding to the review item to the terminal device. The information processing device according to claim 1 .

6. a reception unit that receives input of at least legal documents; a generating unit that generates a prompt that instructs the extraction of a description portion corresponding to each review item from the legal text based on review item information stored in a storage unit and describing review items whose existence should be confirmed in the legal text; an extraction unit that extracts a corresponding description for each review item by inputting the generated prompt into a trained model; an output unit that outputs the extracted description portions to a terminal device in association with each review item; the output unit associates the extracted description portions with each of the review items and outputs the extracted description portions to the terminal device in a spreadsheet; the output unit outputs the content of the review items to the terminal device in an extensible manner on the spreadsheet; the generating unit regenerates a prompt instructing the user to extract a corresponding description portion for each of the expanded review items; The extraction unit re-extracts corresponding description portions for each review item by inputting the re-generated prompt into the trained model. Information processing device.

7. The computer At a minimum, it accepts legal text input. generating a prompt that instructs the system to extract a description portion corresponding to each review item from the legal text based on review item information stored in a storage unit and describing review items whose existence should be confirmed for the legal text; The generated prompt is input into a trained model to extract corresponding descriptions for each review item; The extracted description portions are output to a terminal device in association with each of the review items; Extracting the translation result of the description from the trained model together with the description, outputting the description and the translation result of the description to the terminal device in association with each review item; Information processing methods.

8. On the computer, At the very least, it accepts legal text input, generating a prompt that instructs the system to extract a description portion corresponding to each review item from the legal text based on review item information stored in a storage unit and describing review items whose existence should be confirmed in the legal text; The generated prompt is input into a trained model to extract corresponding descriptions for each review item; The extracted description portions are output to a terminal device in association with each of the review items; Extracting the translation result of the description from the trained model together with the description; outputting the description and the translation result of the description to the terminal device in association with each review item; program.

9. A system comprising a terminal device, an information processing device, and a generation AI server having a trained model, The terminal device inputs at least legal documents into the information processing device; The information processing device accepts input of the legal text, generates a prompt that instructs the device to extract corresponding description passages for each review item from the legal text based on review item information stored in a storage unit that describes review items whose existence should be confirmed for the legal text, extracts corresponding description passages for each review item by inputting the generated prompt into the trained model, associates the extracted description passages with each review item and outputs them to the terminal device, extracts a translation of the description passage from the trained model together with the description passage, and associates the description passage with the translation of the description passage with each review item and outputs them to the terminal device, the terminal device displays the outputted description and the translation result in association with each review item. system.

Images