Document revision support system and document revision support method

Abstract

It helps to make proper document modifications in a simple way. [Solution] A document modification support system 40 includes a processing unit 41 that receives input of an outline of modifications to be made to a document, obtains similar past cases which are identical or similar to the modifications indicated in the outline of modifications from modification history information 1200, searches the document for modification candidate chunks which are the parts of the document that are the target of the modifications indicated in the outline of modifications, obtains related chunks which are parts of the document other than the modification candidate chunks that have content related to the modification candidate chunks registered, and inputs a prompt including the outline of modifications, similar past cases, modification candidate chunks, and related chunks into a trained model 31, thereby outputting the part of the document that should be modified as indicated in the outline of modifications and the content of the modifications to be made to that part.

Description

Technical Field

[0001] The present invention relates to a document correction support system and a document correction support method.

Background Art

[0002] With the development of digitization technology, companies are increasingly managing a large amount of documents. And these documents often require corrections even after they are created. Therefore, a technology for efficiently correcting a large amount of documents is demanded.

[0003] For that purpose, it is necessary to appropriately search for the locations that need correction. Regarding this point, for example, Patent Document 1 discloses an information processing apparatus including a first acquisition unit that acquires a query, a first generation unit that generates input information including one or more target chunks and an instruction sentence according to a first condition included in the query, and a second acquisition unit that acquires output information output by a language model to which the input information is input.

[0004] Also, in order to identify the correction locations, it may be useful to identify the outline of each document. Regarding this point, Patent Document 2 discloses an information processing apparatus including a first acquisition unit that acquires a query, a first extraction unit that extracts one or more matching chunks that are chunks matching the query from one or more target chunks by referring to the query, a first generation unit that generates instruction information for generating a summary of the one or more target chunks and generates instruction information including the one or more matching chunks, and a second acquisition unit that acquires output information output by a language model to which the instruction information is input.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

[0006] However, since companies manage various types of documents, the specific revisions (words, phrases) will differ from document to document. Therefore, it is necessary to broadly identify documents that require revision and make appropriate revisions according to the content of each document, but such work is generally cumbersome. The above-mentioned patent document does not necessarily adequately address such problems.

[0007] This invention has been made in view of these circumstances, and its purpose is to provide a document modification support system and a document modification support method that can assist in appropriately modifying documents in a simple manner. [Means for solving the problem]

[0008] One of the present inventions for solving the aforementioned problems is a document modification support system comprising: a storage device that stores information about a document, modification history information which is information about each modification made to the document in the past, and a trained model that outputs text corresponding to an input text; a modification summary input process that accepts input of an outline of modifications to be made to the document; a similar past case acquisition process that acquires similar past cases which are identical or similar modification parts to the modification indicated by the modification summary from the modification history information using a predetermined algorithm; a modification candidate chunk search process that searches the document for modification candidate chunks which are parts of the document that are the target of modification indicated by the modification summary; a related chunk acquisition process that acquires related chunks which are parts of the document other than the modification candidate chunk and which have content related to the acquired modification candidate chunk registered using a predetermined algorithm; and a prompt input process that inputs a prompt including the modification outline, the acquired similar past cases, the acquired modification candidate chunk, and the acquired related chunk to the trained model, thereby outputting the part of the document that should be modified as indicated by the modification summary and the content of the modification to be made to that part. [Effects of the Invention]

[0009] According to the present invention, it is possible to support the proper modification of documents in a simple manner. [Brief explanation of the drawing]

[0010] [Figure 1] This is an example of the configuration of a document management system according to this embodiment. [Figure 2] This figure shows an example of knowledge priority information. [Figure 3] This figure shows an example of prompt control information. [Figure 4] This figure shows an example of the control information for selecting a generated AI model. [Figure 5] This is a flowchart illustrating the overview of the document modification support process. [Figure 6] It is a sequence diagram for explaining an example of knowledge information registration processing. [Figure 7] It is a sequence diagram for explaining an example of knowledge information update processing. [Figure 8] It is a diagram showing an example of the knowledge information to be created. [Figure 9] It is a diagram showing an example of the knowledge-related information to be created. [Figure 10] It is a sequence diagram for explaining an example of correction history information registration processing. [Figure 11] It is a diagram showing an example of the correction history information to be created. [Figure 12] It is a flowchart for explaining an example of correction classification information registration processing. [Figure 13] It is a sequence diagram for explaining an example of inquiry processing. [Figure 14] It is a flowchart for explaining an example of the processing performed by the document correction support system in knowledge information registration processing and knowledge information update processing. [Figure 15] It is a flowchart for explaining an example of the processing performed by the document correction support system in knowledge information registration processing and knowledge information update processing. [Figure 16] It is an example of a prompt for requesting calculation of relevance. [Figure 17] It is an example of another prompt for requesting calculation of relevance. [Figure 18] It is a flowchart for explaining an example of the processing performed by the document correction support system in correction history information registration processing. [Figure 19] It is a flowchart for explaining an example of the processing performed by the document correction support system in correction classification information registration processing. [Figure 20] It is a diagram showing an example of a prompt for instructing to specify the correction classification to which a correction belongs. [Figure 21] It is a diagram showing an example of the correction classification information to be created. [Figure 22] It is a flowchart for explaining an example of the processing performed by the document correction support system in correction classification information setting processing. [Figure 23] It is a diagram showing an example of a prompt instructing to identify the correction classification to which the correction outline belongs. [Figure 24] It is a flowchart explaining an example of the process performed by the document correction support system in the inquiry process. [Figure 25] It is a flowchart explaining an example of the process performed by the document correction support system in the inquiry process. [Figure 26] It is a diagram showing an example of the knowledge search history information to be created. [Figure 27] It is a diagram showing an example of the inquiry response history information to be created.

Mode for Carrying Out the Invention

[0011] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0012] The document management system 1 of the present embodiment is an information processing system that supports the correction of documents (documents) managed by businesses and the like. The document management system 1 supports appropriately correcting a large number of documents in a batch using a learned model (generative AI (Artificial Intelligence)) described later according to a general (abstract) correction instruction specified by the user.

[0013] In this embodiment, documents are managed in bundles of documents called "document groups," and document modifications are performed on a document group basis. Each document consists of text (sentences, words, etc.), and each document consists of multiple text groups separated by predetermined text. Specifically, a text group is a portion of text separated by predetermined text (title, number, etc.) according to its format or content, such as a part, chapter, section, subsection, item, or paragraph. In this embodiment, each of these separated parts is referred to below as a "section." Furthermore, below, the text groups of a document divided based on a section are referred to as chunks.

[0014] Furthermore, in this embodiment, the document is assumed to include revision history information. Revision history information is added by the document administrator or automatically each time the content of the document is modified, and it indicates the location of the text changes in the document.

[0015] The document management system 1 creates a prompt (hereinafter also called inquiry information) for all documents in a document group, which includes text (hereinafter referred to as "correction summary") outlining the corrections requested by the user, chunks of past corrections in the document group that are identical or similar to the corrections outlined in the correction summary (hereinafter referred to as "similar past correction examples"), chunks of the document group to be corrected (hereinafter referred to as "correction candidate chunks"), and chunks in the document group that have content related to the correction candidate chunks (other than the correction candidate chunks) (hereinafter referred to as "related chunks"). The document management system 1 inputs the created prompt into each trained model in the model storage device 30 to obtain information on the locations and content of the corrections to be made in the document group (hereinafter referred to as "answer information").

[0016] Figure 1 shows an example of the configuration of the document management system 1 according to this embodiment. The document management system 1 is composed of information processing devices including a user device 10 used by users who modify documents using the document management system 1, an administrator device 20 used by the administrator of the document management system 1, a model storage device 30 that stores one or more types of trained models 31, and a document modification support system 40 that assists in modifying documents.

[0017] The pre-trained model 31 is a mathematical model (e.g., a Large Language Model (LLM)) that takes text prompts as input and outputs text corresponding to the input prompts. The pre-trained model 31 is constructed using, for example, BERT (Bidirectional Encoder Representations from Transformers), XLNet, or GPT (Generative Pre-trained Transformer). In this embodiment, a predetermined fee will be charged for using each pre-trained model 31 in the model storage device 30.

[0018] The administrator device 20, user device 10, model storage device 30, and document modification support system 40 are connected by a wired or wireless communication network 5, such as the Internet, LAN (Local Area Network), WAN (Wide Area Network), or dedicated line.

[0019] As shown in Figure 1, the document modification support system 40 comprises a processing unit 41 such as a CPU (Central Processing Unit), a storage device 42 such as RAM (Random Access Memory), ROM (Read Only Memory), HDD (Hard Disk Drive), or SSD (Solid State Drive), an input device 43 consisting of a keyboard, mouse, touch panel, etc., an output device 44 for screen display consisting of a monitor (display), etc., and a communication device 45 consisting of a NIC (Network Interface Card), wireless communication module, USB (Universal Serial Interface) module, or serial communication module, etc. Other information processing devices in the document management system 1 also have similar hardware.

[0020] The document revision support system 40 stores the following information: knowledge information 900, knowledge relevance information 1000, knowledge selection criterion information 200, knowledge priority information 300, revision history information 1200, revision classification information 2200, knowledge retrieval method control information 56, prompt control information 400, generation AI model selection control information 500, knowledge retrieval history information 2700, and inquiry response history information 2800.

[0021] Knowledge Information 900 is aggregated information from chunks, which are units of document divided based on sections (for example, text for each chapter). Details of Knowledge Information 900 will be described later.

[0022] Knowledge relevance information 1000 is information that registers the relevance level, a parameter representing the strength of the relationship between chunks in a document group, for each chunk. In this embodiment, chunks whose relevance level exceeds a predetermined threshold (hereinafter referred to as the relevance chunk threshold) are considered relevant chunks. The relevance chunk threshold varies depending on the combination of chunks. Details of knowledge relevance information 1000 will be described later.

[0023] Knowledge selection criterion information 200 is the lower limit (threshold) of relevance required to consider something as a related chunk.

[0024] Knowledge priority information 300 is information about the priority criteria (hereinafter referred to as the priority rule) among multiple chunks that have the same degree of relevance to a given chunk.

[0025] Revision history information 1200 is a compilation of information summarizing the history of each revision made to each document in a document group. Details of revision history information 1200 will be described later.

[0026] Revision Classification Information 2200 is information that summarizes the categories of revisions (hereinafter referred to as revision classifications) that have been made to documents in a document group, for each document group. Details of Revision Classification Information 2200 will be described later.

[0027] Knowledge retrieval method control information 56 is information about the search conditions when searching for chunks (hereinafter also called candidate chunks for modification) that correspond to the summary of modifications specified by the user from each document in the document group. Knowledge retrieval method control information is information that specifies one or more types of search methods (full-text search, vector search, semantic search), and the union of the search results for each method is used as the search result. Knowledge retrieval method control information is, for example, in the format "key=value".

[0028] The prompt control information 400 contains information about the constraints on the prompt to be created.

[0029] The generated AI model selection control information 500 is information about the selection method or selection criteria for the trained model 31 stored in the model storage device 30. The generated AI model selection control information 500 is referenced when the document modification support system 40 calls up the trained model 31.

[0030] (Knowledge priority information) Figure 2 shows an example of knowledge priority information 300. Knowledge priority information 300 stores the following priority rules: History priority rule 301, which prioritizes chunks with a revision history within the same revision classification; Proximity priority rule 302, which prioritizes chunks that are close to related chunks (close in terms of document chapter structure); Document order rule 303, which prioritizes chunks according to the order of chapters in the document to which related chunks belong (1. ○○ → 1.1 △△ → 2. ×× → ...); Size ascending order rule 304, which prioritizes chunks with smaller sizes; Size descending order rule 305, which prioritizes chunks with larger sizes; and Custom rule 306, which applies any rule set in advance by the user or administrator.

[0031] (Prompt control information) Figure 3 shows an example of prompt control information 400. The prompt control information 400 has information for each setting item 401 and the content 402 of that setting item. The content 402 of the setting item includes, for example, the upper limit size of the prompt 403, the maximum number of similar past correction examples to be attached to the prompt 404, and the order in which the correction candidate chunks and related chunks attached to the prompt are sorted in the prompt (chunk sort 405). If these sorting orders in the prompt are different, the answer output by the trained model 31 will be different.

[0032] (Generated AI model selection control information) Figure 4 shows an example of the generated AI model selection control information 500. The generated AI model selection control information includes information for each criterion item (setting item 501) and its content 502. In the example shown in the figure, setting item 501 includes "selection mode" and "model". If "selection mode" is "fixed", the trained model 31 specified in "model" is used. If "selection mode" is "dynamic", the trained model 31 with the lowest usage fee among the trained models 31 specified in "model" is used. Note that if "selection mode" is "fixed", one trained model 31 is specified, and if "selection mode" is "dynamic", multiple trained models 31 are specified.

[0033] Next, as shown in Figure 1, the document modification support system 40 includes the following functional units: a knowledge information registration unit 51, a modification history information registration unit 52, a modification classification information registration unit 53, a modification classification setting unit 54, and a query processing unit 55.

[0034] The knowledge information registration unit 51 creates or updates knowledge information 900 and knowledge relevance information 1000.

[0035] The revision history information registration unit 52 creates or updates revision history information 1200.

[0036] The revised classification information registration unit 53 creates or updates the revised classification information 2200.

[0037] The modification classification setting unit 54 classifies the modification summary 1206 of the modification history information 1200 into one of the modification classifications based on the modification classification information 2200, and registers the classification result in the modification history information 1200.

[0038] The inquiry processing unit 55 (prompt input processing unit) inputs a prompt, which includes an outline of the modifications to be made to the document group to be modified, similar past modification examples, candidate chunks for modification, and related chunks, as specified by the user, into the trained model 31 in the model storage device 30. This causes the unit to output the parts (chunks) to be modified in the document group to be modified, as indicated by the outline of the modifications, and the content of the modifications to be made to those parts (chunks).

[0039] Specifically, the query processing unit 55 requests that the document to be modified be modified according to the modification summary, and also prompts the trained model 31 to modify one of the chunks, including the candidate modification chunk and the related chunk, according to the modification summary, using similar past modification examples as examples of modification content.

[0040] The functions of the document correction support system 40 (and the same applies to other information processing devices in the document management system 1) are realized by the hardware of the document correction support system 40, or by the arithmetic unit 41 of the document correction support system 40 reading and executing each program stored in the storage device 42. These programs are stored in storage devices such as secondary storage devices, non-volatile semiconductor memory, hard disk drives, SSDs, or recording media readable by each device, such as IC cards, SD cards, or DVDs. Each device may be realized in whole or in part using virtual information processing resources provided using virtualization technology or process space isolation technology, such as virtual servers provided by a cloud system. Furthermore, all or part of the functions provided by these devices may be realized by services provided by the cloud system via APIs (Application Programming Interfaces), for example. Next, we will explain the processes performed in document management system 1.

[0041] Figure 5 is a flowchart illustrating the overview of the document modification support process performed in the document management system 1.

[0042] The document management system 1 executes a knowledge information registration process s1 that creates knowledge information 900 and knowledge relationship information 1000.

[0043] Furthermore, after executing the knowledge information registration process s1, the document management system 1 executes a knowledge information update process s3 to update the knowledge information 900 and the knowledge relationship information 1000. The knowledge information update process s3 is started, for example, when the document modification support system 40 receives an update request from the administrator device 20. The knowledge information update process s3 may be executed multiple times.

[0044] Meanwhile, the document management system 1 executes a revision history information registration process s5 to create or update revision history information 1200. Note that the revision history information registration process s5 may be executed multiple times.

[0045] Furthermore, the document management system 1 executes a revision classification information registration process s7 to create or update revision classification information 2200 based on the revision history information 1200 created or updated in the revision history information registration process s5. Note that the revision classification information registration process s7 may be executed multiple times.

[0046] Furthermore, the document management system 1 executes a revision classification information setting process s8, which classifies each revision history in the revision history information 1200 based on the revision classification information 2200. Note that the revision classification information setting process s8 may be executed multiple times.

[0047] Then, the document management system 1 executes a query process s9 based on a prompt (query information) specified by the user that includes an outline of modifications to a certain document group, and presents each document in that document group that will be modified as indicated in the outline of modifications, as well as the details of those modifications. Note that the query process s9 may be executed multiple times.

[0048] The following explains these processes in detail.

[0049] <Knowledge Information Registration Process> Figure 6 is a sequence diagram illustrating an example of the knowledge information registration process s1.

[0050] First, the document modification support system 40 displays a screen for setting knowledge information 900 (knowledge information registration screen) on the administrator device 20 based on a request from the administrator device 20 (s11).

[0051] The administrator device 20 displays a knowledge information registration screen and accepts the designation of a document group from the administrator. The administrator device 20 then sends each document in the designated document group (each document has revision history information attached) to the document revision support system 40 (s12).

[0052] When the document modification support system 40 receives each document from the administrator device 20, it divides each received document into multiple chunks (s13). Then, the document modification support system 40 registers each divided chunk in the knowledge information 900 (s14).

[0053] The document modification support system 40 sends information to the administrator device 20 indicating that the registration of knowledge information 900 has been completed, and the administrator device 20 displays this information on its screen (s15).

[0054] The document modification support system 40 generates a prompt requesting the calculation of the degree of relevance between the chunks generated in s14, and sends the generated prompt to the model storage device 30 (s16). Based on the prompt received from the document modification support system 40, the model storage device 30 outputs the degree of relevance between the chunks generated in s14 (s17), and sends the outputted degree of relevance information to the document modification support system 40. Subsequently, the document modification support system 40 registers the degree of relevance information received from the model storage device 30 into the knowledge relevance information 1000 (s18). This completes the knowledge information registration process s1.

[0055] <Knowledge information update process> Figure 7 is a sequence diagram illustrating an example of the knowledge information update process s3.

[0056] First, the document revision support system 40 generates a screen (revision history information update recording screen) for updating the knowledge information 900, and displays the generated screen on the administrator device 20 (s31).

[0057] The administrator device 20 displays the revision history information update recording screen and accepts the administrator's designation of each document in the document group (hereinafter referred to as the document to be updated). Then, similar to s12, the administrator device 20 transmits each document to be updated specified by the administrator to the document revision support system 40 (s32).

[0058] The document modification support system 40 divides each received document to be updated into multiple chunks, similar to s13 (s33).

[0059] The document modification support system 40 compares each chunk divided in s33 with each chunk in the document registered in the knowledge information registration process s1 (hereinafter referred to as the pre-update document) (s34).

[0060] Based on the comparison results in s34, the document modification support system 40 identifies the pre-update document and the presence or absence of each chunk corresponding to the document to be updated (s35).

[0061] The document revision support system 40 updates the knowledge information 900 based on the document to be updated, according to the result of s35 (s36).

[0062] For example, the document modification support system 40 adds information about chunks that do not exist in the pre-update document but do exist in the document being updated to the knowledge information 900. The document modification support system 40 also removes information about chunks that exist in the pre-update document but do not exist in the document being updated from the knowledge information 900, focusing on the differences. Furthermore, the document modification support system 40 adds information about chunks that exist in both the pre-update document and the document being updated but have differences in content to the knowledge information 900 (updating the differences). The document modification support system 40 does not update the knowledge information 900 for chunks that exist in both the pre-update document and the document being updated but have no differences in content.

[0063] The document revision support system 40 retrieves all currently defined revision classifications from the revision classification information 2200 (s37).

[0064] The document modification support system 40 generates a prompt in the same manner as in s16, requesting the calculation of the degree of relevance between chunks of the knowledge information 900 whose information was updated in s36, and sends the generated prompt to the model storage device 30 (s38).

[0065] Based on prompts received from the document modification support system 40, the model memory device 30 outputs the degree of relevance between the chunks generated in s38 (s39) and sends the outputted relevance information to the document modification support system 40. Subsequently, the document modification support system 40 updates the knowledge relevance information 1000 using the relevance information received from the model memory device 30 (s40). The document modification support system 40 updates only the information of the chunks related to the updated document.

[0066] The document revision support system 40 sends information to the administrator device 20 indicating that the update of the knowledge information 900 is complete, and the administrator device 20 displays this information on its screen (s41). This completes the knowledge information update process s3.

[0067] (Knowledge information) Figure 8 shows an example of the knowledge information 900 that is created. For each chunk, the knowledge information 900 includes the chunk ID 901, the document group 902 to which the chunk belongs, the original file name of the chunk 903, the section name 904 of the section to which the chunk belongs, the sequence ID 905 of the chunk sequence (the unit of division when the same section is divided into multiple chunks), the chunk content 906 (text), the vector value 907 of the chunk text, the chunk size 908 (number of tokens), and the relevance update 909 indicating whether the chunk is in a state of being newly registered ("I") or has been registered at least once ("U").

[0068] (Knowledge-related information) Figure 9 shows an example of the knowledge relevance information 1000 that is created. In this example, the knowledge relevance information 1000 is information in which the degree of relevance between chunks is registered in a matrix format. For each modification classification 1001 to which the modification made to a chunk belongs, the degree of relevance 1002 between chunks to which the modification belonging to that modification classification 1001 has been made is registered in the knowledge relevance information 1000.

[0069] <Revision History Information Registration Process> Figure 10 is a sequence diagram illustrating an example of the modification history information registration process s5.

[0070] First, when the document modification support system 40 receives a predetermined request from the administrator device 20, it generates a screen for registering the modification history information 1200 (modification history information registration screen) and displays the generated screen on the administrator device 20 (s51).

[0071] The administrator device 20 displays the revision history information registration screen and accepts the designation of a document group from the administrator. The administrator device 20 then sends each document in the designated document group to the document revision support system 40 (s52).

[0072] The document revision support system 40 extracts revision history information attached to the received document (including information on the revision summary and information on the sections that were revised) (s53), and registers the extracted revision history information in the revision history information 1200 (s54).

[0073] Then, the document revision support system 40 displays a screen (registration completion screen) on the administrator device 20 indicating that the registration of the revision history has been completed (s55). This completes the revision history information registration process s5.

[0074] (Revision history information) Figure 11 shows an example of the revision history information 1200 that is created. The revision history information 1200 contains information on each document group 1201, each document name 1202 within that document group, a revision number 1203 assigned to each revision made to that document, a branch number 1204 assigned to each section in which the revision was made (set when the same revision is made to multiple sections), a revision classification 1205 (revision classification ID) to which the revision belongs, a summary of the revision 1206, and the section in which the revision was made (revision section name 1207, corresponding to branch number 1204). Note that the revision classification 1205 is set by the revision classification information setting process s9 described later.

[0075] <Processing for registering corrected classification information> Next, Figure 12 is a flowchart illustrating an example of the modified classification information registration process s7.

[0076] The document modification support system 40 generates a screen (modification classification information registration screen) for specifying the document group to be identified for modification classification, and displays the generated screen on the administrator device 20 (s71).

[0077] Subsequently, the administrator device 20 receives a document group designation from the administrator and transmits the information of the designated document group to the document modification support system 40 (s72).

[0078] The document modification support system 40 obtains information on the modification history for each modification of each document in the received document group from the corresponding record in the modification history information 1200 (s73).

[0079] The document modification support system 40 then generates a prompt asking which modification category each modification belongs to, and sends the generated prompt to the model storage device 30 (s74).

[0080] The model memory device 30 inputs prompts received from the document modification support system 40 into the trained model 31, and outputs information about the modification classification to which each modification belongs (s75). The model memory device 30 sends the outputted modification classification information to the document modification support system 40. If a modification classification does not exist to which a particular modification belongs, the model memory device 30 creates a new modification classification for that attribute.

[0081] The document modification support system 40 registers the modification classification information for each modification received from the model storage device 30 into the modification classification information 2200 (s76). The document modification support system 40 displays a screen (registration completion screen) on the administrator device 20 indicating that the registration of the modification classification is complete (s77).

[0082] Subsequently, the document revision support system 40 updates the knowledge relevance information 1000 based on the revision classification updated by the processing up to s77.

[0083] In other words, first the document modification support system 40 obtains the chunks of each document in the document group specified in s72 from the knowledge information 900 (s78).

[0084] Then, the document revision support system 40 performs the following processing s79-s81 for each of the revision classifications that have been updated by the processing up to s77.

[0085] In other words, the document modification support system 40 sends a prompt to the model storage device 30 instructing it to output the degree of relevance between chunks in its modification classification (s79).

[0086] The model memory device 30 inputs prompts received from the document correction support system 40 into the trained model 31 and outputs the relevance between each chunk (s80). The model memory device 30 sends the outputted relevance information to the document correction support system 40.

[0087] The document modification support system 40 registers each relevance level received from the model storage device 30 into the knowledge relevance information 1000 (s81). This completes the modification category information registration process s7. The document modification support system 40 may choose not to register information on relevance levels for chunks whose modification classification has not been identified into the knowledge relevance information 1000.

[0088] <Inquiry Processing> Figure 13 is a sequence diagram illustrating an example of query processing s9.

[0089] The document modification support system 40 generates a screen (inquiry screen) for users to input inquiry information, and displays the generated screen on the administrator device 20 (s91).

[0090] Subsequently, the administrator device 20 receives from the administrator the designation of the document group to be modified (hereinafter referred to as the document group to be modified) and the input of text (document) indicating the outline of the modifications to be made, and transmits the entered information of the document group and the modification outline to the document modification support system 40 (s92). This modification outline information may be text such as "I want to change XX to YY," and may specify only abstract modifications rather than individual modification locations or specific modification details.

[0091] The document revision support system 40 retrieves all revision classifications from the revision classification information 2200 (s93).

[0092] The document modification support system 40 generates a prompt (s94) requesting that the modification classification to which the modification summary received by s92 belongs be output, based on all modification classifications obtained in s93. The document modification support system 40 then sends the generated prompt to the model storage device 30.

[0093] The model storage device 30 inputs a prompt received from the document correction support system 40 into the trained model 31, outputs information about the correction classification to which the correction belongs (s95), and sends the outputted correction classification information to the document correction support system 40.

[0094] The document modification support system 40 obtains all information on modification history (similar past modification examples) of the same type of modification classification (i.e., identical or similar) as the modification classification received from the model storage device 30 in the document group received by s92 from the modification history information 1200 (s96).

[0095] Meanwhile, the document revision support system 40 searches the knowledge information 900 for all chunks (candidate revision chunks) of each document containing the revision summary text received by s92 (s97).

[0096] The document revision support system 40 identifies all related chunks, i.e., associated chunks, for each revision candidate chunk found in s97, based on the knowledge relevance information 1000 (s98).

[0097] Then, the document revision support system 40 obtains information about each related chunk acquired in s98 from the knowledge information 900 (s99).

[0098] Based on the information obtained from the above process regarding similar past revision examples, revision summaries, candidate revision chunks, and related chunks, the document revision support system 40 generates a prompt (inquiry information) instructing that the revisions indicated in the revision summary be applied to all documents in the document group to be revised (s100). Then, the document revision support system 40 sends the prompt generated in s100 to the model storage device 30 (s101).

[0099] The model storage device 30 inputs the prompt received from the document revision support system 40 into the trained model 31, and outputs information on suggested revisions for the revisions indicated by the prompt (s102). The model storage device 30 then transmits the outputted information on suggested revisions to the document revision support system 40.

[0100] The document revision support system 40 displays the information of the proposed revisions received from the model storage device 30 on the screen of the user device 10 (s103). This completes the inquiry process s9.

[0101] Next, we will explain in detail the processes performed by the document modification support system 40 in each of the above processes.

[0102] <Details of Knowledge Information Registration and Update Processes> Figures 14 and 15 are flowcharts illustrating an example of the processes performed by the document modification support system 40 in the knowledge information registration process s1 and the knowledge information update process s3 (divided into two figures due to space limitations).

[0103] First, as shown in Figure 14, when the document modification support system 40 receives each document in the document group from the administrator device 20 (s101), it repeats the following processes s103-s113 for each received document (s102, s114).

[0104] In other words, first the document modification support system 40 divides the document into sections and stores each divided section as a chunk (s103).

[0105] For example, the document editing support system 40 analyzes the text in the document, divides the document into sections (units such as parts, chapters, and subsections), and treats the text of each divided section as a chunk.

[0106] Furthermore, if the document modification support system 40 determines that the size of the identified chunk exceeds a predetermined size, it may further divide the chunk so that the size does not exceed that limit.

[0107] The document modification support system 40 repeats the following processing s105-s109 for each chunk divided in s103 (s104, s110).

[0108] In other words, first the document modification support system 40 checks whether the chunk is already registered in the knowledge information 900 (s105). If the chunk is already registered in the knowledge information 900 (s105: Yes), the document modification support system 40 executes the process in s106; if the chunk is not registered in the knowledge information 900 (s105: No), the document modification support system 40 executes the process in s109.

[0109] In s106, the document modification support system 40 determines whether there is a difference between the chunk registered in the knowledge information 900 and the chunk divided in s103. If there is a difference (s106: Yes), the document modification support system 40 executes the process in s107; if there is no difference (s106: No), the document modification support system 40 executes the process in s110.

[0110] In s107, the document modification support system 40 deletes the chunk registered in the knowledge information 900. Then, the document modification support system 40 registers the chunk that was divided in s103 into the knowledge information 900 (s108). The document modification support system 40 also sets the relevance update 909 of the record related to the chunk in the knowledge information 900 to "U" to indicate an update. After that, the process in s110 is performed.

[0111] In s109, the document modification support system 40 registers the chunks divided in s103 into a new record in knowledge information 900. The document modification support system 40 also sets the relevance update 909 of the relevant record in knowledge information 900 to "I" to indicate a new registration. After that, the process in s110 is performed.

[0112] In s110, the document modification support system 40 checks whether the processes s105-s109 have been repeated for all chunks divided in s103. If the processes s105-s109 have been repeated for all chunks divided in s103, it performs the processes from s111 onwards.

[0113] In s111, the document correction support system 40 determines whether information for chunks that do not belong to any of the chunks divided in s103 is registered in the knowledge information 900. If information for chunks that do not belong to any of the chunks is registered in the knowledge information 900 (s111: Yes), the document correction support system 40 executes the process in s112. If information for chunks that do not belong to any of the chunks is not registered in the knowledge information 900, it executes the process in s114.

[0114] In s112, the document revision support system 40 deletes the information of that chunk from the knowledge relationship information 1000. The document revision support system 40 also deletes the information of that chunk from the knowledge information 900 (s113). After that, the process in s114 is performed.

[0115] In s114, the document correction support system 40 checks whether the processes in s103-s113 have been repeated for all documents received in s101. If the processes in s103-s113 have been repeated for each document received in s101, it performs the processes from s115 onwards as shown in Figure 16.

[0116] In s115 of Figure 15, the document correction support system 40 obtains data for all chunks of the document group received in s101 from the knowledge information 900.

[0117] The document modification support system 40 then determines whether or not there are any chunks in which the data has been updated (chunks in which the relevance update 909 is set to "U") (s116). If there are chunks in which the data has been updated (s116: Yes), the document modification support system 40 identifies all of those chunks (hereinafter referred to as updated chunks) and executes the process in s117. If there are no chunks in which the data has been updated (s116: Yes), the document modification support system 40 executes the process in s121.

[0118] In s117, the document modification support system 40 repeats the following processes s118-s119 for all update chunks (s120).

[0119] In other words, the document modification support system 40 uses a prompt to instruct the trained model 31 in the model memory 30 to output the degree of relevance between the updated chunk and other chunks in the document group (s118).

[0120] For example, the document modification support system 40 generates a prompt that includes the update chunk and each other chunk, requests the calculation of the degree of relevance between the chunks, and inputs the generated prompt into the trained model 31 in the model storage device 30.

[0121] Furthermore, for example, the document revision support system 40 generates a prompt that includes information on each revision classification, update chunks, and other chunks, and requests the calculation of the degree of relevance between chunks for each revision classification, and inputs the generated prompt into the trained model 31 in the model storage device 30. If, at this point, the revision classifications necessary for calculating the degree of relevance for each revision classification have not yet been registered in the revision history information 1200, the document revision support system 40 may execute the revision classification setting process s8.

[0122] Subsequently, the document revision support system 40 updates the knowledge relevance information 1000 using the relevance information output in s118 (s119).

[0123] The document modification support system 40 then checks whether the processes s118-s119 have been repeated for all update chunks (s120), and if the processes s118-s119 have been repeated for all update chunks, it performs the processes from s121 onwards.

[0124] In s121, the document modification support system 40 determines whether there are any newly registered chunks (chunks in which the relevance update 909 is set to "I"). If there are any newly registered chunks of data (s121:Yes), the document modification support system 40 identifies all of those chunks (hereinafter referred to as newly registered chunks) and executes the process in s122. If there are no newly registered chunks of data (s121:No), the process terminates.

[0125] In s122, the document modification support system 40 repeats the following processes s123-s124 for all newly registered chunks (s125).

[0126] In other words, the document modification support system 40 uses a prompt to instruct the calculation of the degree of relevance between the newly registered chunk and other chunks, and causes the trained model 31 in the model memory device 30 to output the degree of relevance (s123).

[0127] For example, the document modification support system 40 generates a prompt that includes the newly registered chunk and other chunks, and requests the calculation of the degree of relevance between chunks, and inputs the generated prompt into the trained model 31 in the model storage device 30.

[0128] Furthermore, for example, the document modification support system 40 generates a prompt that includes information on each modification classification, newly registered chunks, and other chunks, and requests the calculation of the degree of relevance between chunks for each modification classification, and inputs the generated prompt into the trained model 31 in the model storage device 30.

[0129] Subsequently, the document revision support system 40 registers the relevance information output in s123 into the knowledge relevance information 1000 (s124).

[0130] Then, the document modification support system 40 checks whether the process s123-s124 has been repeated for all newly registered chunks (s125), and if the process s123-s124 has been repeated for all newly registered chunks, the process terminates.

[0131] Here, Figure 16 is an example of a prompt requesting the calculation of relevance in s118. This prompt 1700 includes instruction text 1701 for calculating relevance, a document summary 1702, an update chunk 1703 (or a new entry chunk), other chunks 1704, an explanatory text 1705 that describes the criteria for calculating relevance, citing the document's contents as specific examples, and the relevance output format 1706.

[0132] Figure 17 also shows an example of another prompt that requests the calculation of relevance in s118. This prompt 1800 is similar to the prompt 1700 mentioned above, but the instruction wording 1801 for calculating relevance includes wording 1802 that identifies the modified classification of the chunks for which relevance is to be calculated.

[0133] <Details of the revision history information registration process> Next, Figure 18 is a flowchart illustrating an example of the process performed by the document modification support system 40 in the modification history information registration process s5.

[0134] The document revision support system 40 retrieves revision history information (files) associated with each document registered in the knowledge information 900 (s501).

[0135] The document revision support system 40 extracts information from the revision history information, including the name of the document that was revised (hereinafter referred to as the target document), a document describing the revision summary (hereinafter referred to as the revision summary), and the name of the section that was revised (hereinafter referred to as the revised section) (s502).

[0136] The document modification support system 40 repeats the processes s504-s508 for each target document (s503, s509). Furthermore, the document modification support system 40 repeats the processes s505-s509 for each modification made to that target document (s504, s510).

[0137] In other words, in s505, the document modification support system 40 determines whether a summary of the modification related to that modification has already been registered in the modification history information 1200. If the summary of the modification has already been registered in the modification history information 1200 (s505:Yes), the document modification support system 40 executes the process in s506. If the summary of the modification has not been registered in the modification history information 1200 (s505:No), the document modification support system 40 executes the process in s508.

[0138] In s506, the document modification support system 40 determines whether the modified section in the target document differs from (has a difference from) the modified section name 1207 in the modification history information 1200. If the modified section has a difference from the modified section name 1207 in the modification history information 1200 (s506: Yes), the document modification support system 40 executes the process in s507. If the modified section does not have a difference from the modified section name 1207 in the modification history information 1200 (s506: No), the document modification support system 40 executes the process in s510.

[0139] In s507, the document revision support system 40 updates the revision history information 1200 with information about revisions to the target document.

[0140] For example, if the document revision support system 40 has deleted a revision section, it deletes the record in the revision history information 1200. Also, if the document revision support system 40 has added a revision section, it adds the revision information related to that revision section to a new record in the revision history information 1200. After that, the process in s509 is performed.

[0141] In s508, the document modification support system 40 adds information about the modification in the target document to the modification history information 1200. Specifically, the document modification support system 40 adds the modification information to a new record in the modification history information 1200. After that, the process in s509 is performed.

[0142] In s509, the document modification support system 40 checks whether the processes in s505-s509 have been repeated for all modifications in the target document. If the processes in s505-s509 have been repeated for all modifications in the target document, it performs the process in s510.

[0143] In s510, the document modification support system 40 checks whether the processes s504-s509 have been repeated for all target documents. If the processes s504-s509 have been repeated for all target documents, the process terminates.

[0144] <Details of the modified classification information registration process> Figure 19 is a flowchart illustrating an example of the process performed by the document modification support system 40 in the modification classification information registration process s5.

[0145] The document modification support system 40 obtains all modification history information corresponding to each document in the document group (target document group) specified by the administrator device 20 from the modification history information 1200 (s701).

[0146] Then, the document revision support system 40 obtains all the revision summary information for each revision from the revision history information acquired in s701, from the revision summary 1206 of the revision history information 1200 (s702).

[0147] Based on the information on the correction summary of each correction obtained in s702, the document correction support system 40 uses a prompt to instruct the system to identify the correction classification to which each correction belongs, and the trained model 31 in the model memory device 30 outputs the correction classification to which each correction belongs (s703).

[0148] For example, the document revision support system 40 generates a prompt that includes information about the revision summary and instructs the system to identify the revision category to which the revision belongs, and inputs the generated prompt into the trained model 31 in the model storage device 30.

[0149] The document revision support system 40 determines whether each revision classification output in s703 is a revision classification already registered in the revision classification information 2200 (s704). If the revision classification output in s703 is a registered revision classification (s704: Yes), the document revision support system 40 executes the process in s705. If the revision classification output in s703 is not a registered revision classification (s704: No), the document revision support system 40 executes the process in s706.

[0150] In s705, the document modification support system 40 deletes the registered modification classification information from the modification classification information 2200 (it deletes the records related to the registered modification classification of the target document group in the modification classification information 2200). After that, the process in s706 is performed.

[0151] In s706, the document modification support system 40 adds the information of each modification classification output in s703 to the modification classification information 2200. For example, the document modification support system 40 registers the information of each modification classification output in s703 (such as the modification classification ID) in a new record related to the target document group in the modification classification information 2200.

[0152] Subsequently, the document correction support system 40 repeats the following processes s708-s710 for each correction classification registered in the correction classification information 2200 (s707, s711).

[0153] In other words, the document revision support system 40 determines whether or not knowledge relationship information 1000 corresponding to the revision classification exists (s708). Specifically, the document revision support system 40 determines whether or not the degree of relationship between chunks (hereinafter referred to as corresponding chunks) in the revision classification is registered in the knowledge relationship information 1000.

[0154] If knowledge-related information 1000 corresponding to the revision classification exists (s708:Yes), the document revision support system 40 executes the process in s711. If knowledge-related information 1000 corresponding to the revision classification does not exist (s708:No), the document revision support system 40 executes the process in s509.

[0155] In s709, the document modification support system 40 uses a prompt to instruct the calculation of the degree of relevance between each corresponding chunk, and the trained model 31 in the model storage device 30 outputs the degree of relevance.

[0156] For example, the document modification support system 40, similar to the knowledge information registration process s1 and the knowledge information update process s3, sends a prompt to the model storage device 30 instructing it to output the degree of relevance between each corresponding chunk, including each corresponding chunk.

[0157] Then, the document revision support system 40 registers the information on each degree of relevance output in s709 into the knowledge relevance information 1000 (s710). After that, the processing in s711 is performed.

[0158] In s711, the document modification support system 40 checks whether the processing in s708-s710 has been performed for all modification classifications. If the processing in s708-s710 has been performed for all modification classifications, the process is terminated.

[0159] Figure 20 shows an example of a prompt that instructs the user to identify the revision category to which the revision belongs. This prompt 2100 includes wording 2101 that instructs the user to identify the revision category, instructions 2102 on the output format of the revision category, and the contents 2103 of each revision (revision history) (revision number, revision date, revision summary, etc.).

[0160] (revised classification information) Figure 21 shows an example of the created revision classification information 2200. The revision classification information 2200 includes information for each document group 2201, the ID of each revision classification set for that document group (revision classification ID 2202), and the revision classification 2203. Revision classifications include, for example, basic revisions during version upgrades, functional changes, functional additions, and corrections of typographical errors or omissions.

[0161] <Details of the correction classification setting process> Next, Figure 22 is a flowchart illustrating an example of the process performed by the document modification support system 40 in the modification classification information setting process s8.

[0162] The document modification support system 40 obtains all modification history information for each document in the document group (target document group) specified by the administrator device 20 from the modification history information 1200 (s801).

[0163] Then, the document revision support system 40 obtains all the revision summary information for each revision from the revision history information acquired in s801, from the revision summary 1206 of the revision history information 1200 (s802).

[0164] Furthermore, the document modification support system 40 acquires all records of modification classification information 2200 related to the target document group (s803).

[0165] Furthermore, the document revision support system 40 retrieves all of the revision classification information (revision classification ID 2202) from the records of revision classification information 2200 acquired in s803 (s804).

[0166] The document correction support system 40 repeats the following processes s806-s807 for each correction summary obtained in s802 (s805, s808).

[0167] In other words, the document modification support system 40 uses a prompt that instructs the system to identify the modification classification to which the modification summary belongs, and causes the trained model 31 in the model storage device 30 to output the modification classification (s806).

[0168] For example, the document modification support system 40 sends a prompt to the model storage device 30 that includes information on the modification summary and each modification classification, and instructs the system to output the modification classification to which the modification summary belongs (one of the modification classifications obtained in s804).

[0169] Then, the document correction support system 40 sets the correction classification output in s806 as correction classification 1205 in the correction history information 1200 (s807). After that, the process in s808 is performed.

[0170] In step s808, the document revision support system 40 checks whether the processes s806-s807 have been performed for all revision summaries. If the processes s806-s807 have been performed for all revision summaries, the process terminates. On the other hand, if there are revision summaries for which the processes s806-s807 have not been performed, the document revision support system 40 repeats the processes s806-s807 for those revision summaries.

[0171] Figure 23 shows an example of a prompt in s806 that instructs the user to identify the correction category to which the correction summary belongs. This prompt 2400 includes wording 2401 that instructs the user to identify the correction category, the content of the correction summary 2402, an instruction 2403 on the output format of the correction category, and the content 2404 for each correction category.

[0172] <Details of inquiry processing> Figures 24 and 25 are flowcharts illustrating an example of the processing performed by the document modification support system 40 in the inquiry processing s9 (divided into two figures due to space limitations).

[0173] As shown in Figure 24, the document modification support system 40 receives inquiry information from the user device 10 via the inquiry screen (s901). The inquiry information includes the document group to be modified (the document group to be modified) and an overview of the modification.

[0174] The document modification support system 40 retrieves all records related to the document group to be modified from the modification classification information 2200 (s902). Then, the document modification support system 40 extracts all modification classifications 2203 from the retrieved records (s903).

[0175] The document modification support system 40, similar to the modification classification setting process s8, uses a prompt that instructs the system to identify the modification classification to which the modification summary in the query information belongs, and causes the trained model 31 in the model storage device 30 to output the modification classification (s904).

[0176] The document revision support system 40 retrieves similar revision histories (similar past revision examples) made in the same document group in the past from the revision history information 1200 (s905). Specifically, the document revision support system 40 retrieves records from the revision history information 1200 that are associated with the revision classification and the document group to be revised output in s904.

[0177] Then, the document modification support system 40 obtains the modification summary 1206, document name 1202, and modified section name 1207 of the record of modification history information 1200 obtained in s905 (s906).

[0178] Furthermore, the document revision support system 40 uses the text of the revision summary as a key and searches all chunks related to the document group to be revised from the knowledge information 900 using the search method indicated by the knowledge search method control information 56, and sets each of them as a candidate chunk for revision (s908).

[0179] Then, the document correction support system 40 repeats the following processes s910-s911 for each correction candidate chunk obtained in s908 (s901, s912).

[0180] In other words, the document revision support system 40 identifies all related chunks of the candidate revision chunk from the knowledge relevance information 1000 based on the knowledge selection criterion information 200 (s910, s911).

[0181] Specifically, the document revision support system 40 obtains the related chunk threshold between each chunk from the knowledge selection criterion information 200 related to the revision classification identified in s904, and determines that each chunk is a related chunk if the degree of relevance between each chunk in the knowledge relevance information 1000 is greater than the corresponding related chunk threshold.

[0182] Next, as shown in Figure 25, the document modification support system 40 obtains a template prompt (s913).

[0183] The document revision support system 40 sets the number of similar past revision cases obtained in s905 to the prompt obtained in s913 (s915). However, if the number of similar past revision cases exceeds the maximum number of similar past revision cases 404 in the prompt control information 400, the document revision support system 40 sets the maximum number of similar past revision cases 404 to the prompt (s914, s915).

[0184] The document revision support system 40 sets the inquiry information received in s901 as the prompt (s916).

[0185] The document correction support system 40 repeats the following processing s918-s930 for each candidate chunk for correction (s917, s931).

[0186] Specifically, the document revision support system 40 first obtains information about each related chunk of the candidate revision chunk from the knowledge information 900 (s918-s920).

[0187] The document modification support system 40 then refers to the chunk sort 405 of the prompt control information 400 and determines the processing order (s921, s922). If the processing order is by relevance, the document modification support system 40 executes the process in s924; if the processing order is by document order, the document modification support system 40 executes the process in s923.

[0188] In s923, the document modification support system 40 sorts the candidate chunks and related chunks in order of document ID, and adds them to the prompt in the sorted order. After that, the process in s926 is performed.

[0189] Meanwhile, in s924, the document revision support system 40 sorts related chunks in order of relevance. If there are multiple related chunks with the same degree of relevance, the document revision support system 40 sorts these multiple related chunks based on the priority rules of the knowledge priority information 300, thereby assigning a priority order to each of them.

[0190] The document modification support system 40 then adds the candidate chunks for modification and each sorted related chunk to the prompt (s925). After that, the process in s926 is performed.

[0191] In s926 and s927, the document correction support system 40 determines whether the size of the current prompt exceeds the upper limit size 403 of the prompt in the prompt control information 400. If the size of the current prompt exceeds the upper limit size 403 of the prompt in the prompt control information 400 (s927: Yes), the document correction support system 40 executes the process in s928. If the current prompt does not exceed the upper limit size 403 of the prompt in the prompt control information 400 (s927: No), the document correction support system 40 executes the process in s929.

[0192] In s928, the document modification support system 40 deletes the chunk information set in the prompt until the size of the prompt becomes less than or equal to the upper limit size 403 of the prompt in the prompt control information 400.

[0193] For example, the document modification support system 40 deletes related chunks in order from those listed after the prompt. In particular, the document modification support system 40 deletes related chunks in order from those with lower priority among the multiple related chunks that were prioritized in s924, so that only a portion of the multiple related chunks are included in the prompt. After that, the process in s929 is performed.

[0194] In s929, the document modification support system 40 refers to the generated AI model selection control information 500 and identifies the trained model 31 to be used.

[0195] The document modification support system 40 then inputs the current prompt into the trained model 31 identified in s929 of the model memory device 30, thereby obtaining the response information to the prompt output from the trained model 31 (s930). The document modification support system 40 also attaches to the prompt wording that instructs the use of similar past cases as reference examples for modification, and wording that requests modification to at least one of the candidate chunks for modification and each related chunk.

[0196] Furthermore, the document revision support system 40 registers the contents of the search process in the query process s9 into the knowledge search history information 2700. In addition, the document revision support system 40 registers the contents of the answer information in the query process s9 into the query answer history information 2800.

[0197] (Knowledge search history information) Figure 26 shows an example of the knowledge retrieval history information 2700 that is created. The knowledge retrieval history information 2700 includes the date and time when the user made a query (search) (search execution date and time 2701), the query ID 2702 which is the ID assigned to that search, the modification summary set in that search (query content 2703), the search format 2704, and the chunk ID 2705 of the retrieved chunk. The search format 2704 can be, for example, full-text search, vector search, or semantic search.

[0198] (Inquiry response history information) Figure 27 shows an example of the query response history information 2800 that is created. The query response history information 2800 stores the content of the user's search and the answer output by the trained model 31. Specifically, the query response history information 2800 contains the execution date and time 2801 of each search, the query ID 2802 assigned to that search, the query content 2803 which is the prompt in that search, the answer ID 2804 of the answer corresponding to that search, the document group to be modified 2805 indicated in the answer, the document name 2806 of the document to be modified, the section name 2807 of the section to be modified in that document, and the text of the answer (answer content 2808).

[0199] As described above, the document modification support system 40 of this embodiment receives input of an outline of modifications to be made to a document group, obtains similar past examples of modifications indicated by the modification outline from the modification history information 1200, searches for candidate modification chunks in the document group, obtains related chunks for the candidate modification chunks, and inputs a prompt including the modification outline, similar past examples, candidate modification chunks, and related chunks to the trained model 31, thereby outputting the portion of the document group that should be modified as indicated by the modification outline and the content of the modifications to be made to that portion.

[0200] In other words, the document revision support system 40 inputs a prompt into the trained model 31 that adds a summary of the revision to be made, corresponding revision candidates (revision candidate chunks), chunks related to the revision candidate chunks, and similar past revision examples corresponding to the revision summary. This allows the system to broadly extract the areas that need revision and propose appropriate revisions for them, simply by specifying a summary of the revision to be made to the document group.

[0201] Thus, the document revision support system 40 can assist in making appropriate revisions to documents in a simple manner.

[0202] Furthermore, the above prompt requests that the document be modified as outlined in the modification summary, and that the modification summary be applied to either the candidate modification chunk or any of the related chunks, while referring to similar past cases as modification examples.

[0203] This helps to correct the parts of the document that need correction with the appropriate content.

[0204] Furthermore, the document modification support system 40 of this embodiment identifies the category (modification classification) to which each modification made in the past belongs in each document of the document group, and based on the identified category and the modification classification to which the modification indicated in the modification summary belongs, it retrieves from the modification history information 1200, as similar past examples, the parts (chunks) of modifications made in the past in each document of the document group that belong to the same modification classification as the modification indicated in the modification summary.

[0205] In this way, by identifying similar past cases to the revision summary based on the revision classification, it is possible to collect similar past cases within an appropriate scope.

[0206] Furthermore, the document modification support system 40 of this embodiment calculates the degree of relevance between the candidate modification chunks and the parts of the document other than the candidate modification chunks, and acquires related chunks based on the calculated degrees of relevance.

[0207] In this way, by calculating the degree of relevance between chunks and using it to identify related chunks, it is possible to appropriately extract related chunks that are relevant to the candidate chunks for modification.

[0208] Furthermore, the document modification support system 40 of this embodiment calculates the degree of relevance between each chunk for each category (modification classification) to which past modifications made to the document belong, and acquires related chunks based on the degree of relevance related to the modification classification to which the modification summary belongs.

[0209] The content of modifications to each part (chunk) of a document often differs depending on the content of that part. Therefore, as described above, by identifying related chunks based on their relevance to the modification classification to which the modification summary belongs, it is possible to identify the appropriate related chunks according to the content of the modifications.

[0210] Furthermore, when the document modification support system 40 of this embodiment acquires multiple related chunks with the same degree of relevance, it determines the priority of those multiple related chunks based on the knowledge priority information 300 (priority rule), and based on the determined priority, includes some of those multiple related chunks in the prompt.

[0211] This allows us to determine the priority of related chunks of similar importance, thereby identifying the appropriate modifications.

[0212] Furthermore, the document modification support system 40 of this embodiment divides a document in a document group into multiple sections separated by predetermined text (such as part or chapter titles), and searches for candidate chunks for modification from the divided sections. Similarly, the document modification support system 40 of this embodiment divides a document into multiple sections separated by predetermined text, and retrieves related chunks from the document from the divided sections.

[0213] In this way, by defining sections as the constituent units of candidate chunks and related chunks, it becomes possible to appropriately identify candidate chunks and related chunks according to their content and propose appropriate document revisions on a topic-by-topic basis.

[0214] The above description of embodiments is for the purpose of facilitating understanding of the present invention and does not limit it. The present invention can be modified and improved without departing from its spirit, and equivalents thereof are included.

[0215] For example, some of the hardware components of each device in this embodiment may be provided in other devices.

[0216] Furthermore, each program of each device may be provided in other devices, a program may consist of multiple programs, or multiple programs may be integrated into a single program.

[0217] Furthermore, in this embodiment, the degree of relevance between chunks for identifying related chunks was calculated based on the trained model 31. However, the degree of relevance between chunks may also be calculated using algorithms other than the trained model, such as calculating the degree of match or similarity of each word between chunks, or the degree of relevance may be set based on input from the administrator.

[0218] Furthermore, in this embodiment, the registration of the revised classification is performed by the trained model 31 in the revised classification information registration process s7, but the revised classification may also be registered based on input from the administrator.

[0219] Furthermore, in this embodiment, the determination of the corrected classification is performed by the trained model 31 in the corrected classification information setting process s8. However, the determination of the corrected classification may also be performed by an algorithm other than the trained model, for example, by calculating the degree of word match or similarity.

[0220] Furthermore, while this embodiment identifies similar past cases based on a revised classification, similar past cases may also be identified using other algorithms, such as calculating word matching or similarity scores. [Explanation of Symbols]

[0221] 40 Document revision support systems, 31 pre-trained models, 900 knowledge pieces, 1200 revision history pieces.

Claims

1. A storage device that stores document information, revision history information which is information on each revision made to the said document in the past, and a trained model that outputs text corresponding to the input text, and A revision summary input process that accepts input of a summary of the revisions to be made to the aforementioned document, A process for acquiring similar past cases, which involves obtaining similar past cases from the revision history information using a predetermined algorithm, where the parts of the revisions that are identical or similar to the revisions indicated in the summary of the revisions, A correction candidate chunk search process searches the document for correction candidate chunks, which are the parts of the document that are subject to correction as indicated in the summary of the corrections, Related chunk acquisition process: Acquires related chunks, which are parts of the document other than the acquired modification candidate chunk, using a predetermined algorithm, where related content is registered with the acquired modification candidate chunk. A computing device that inputs a prompt including the summary of the correction, the acquired similar past examples, the acquired candidate correction chunks, and the acquired related chunks into the trained model, thereby executing a prompt input process that outputs the portion of the document that should be corrected as indicated by the summary of the correction, and the content of the correction to be made to that portion. A document revision support system equipped with the following features.

2. The aforementioned computing device is In the prompt input process described above, a prompt is input to the trained model requesting that the document be modified as indicated in the modification summary, and that, while referring to the acquired similar past examples as modification examples, the model is requested to modify any of the chunks, including the acquired modification candidate chunk and the acquired related chunk, as indicated in the modification summary. The document modification support system according to claim 1.

3. The aforementioned computing device is Further, a modification classification information registration process is performed to identify the category to which each modification made in the past belongs in the aforementioned document. In the process of acquiring similar past cases, based on each of the identified categories and the category to which the modification indicated in the summary of the modification belongs, the portion of the document that has been modified in the past and belongs to the same category as the modification indicated in the summary of the modification, is acquired from the modification history information as a similar past case. The document modification support system according to claim 1.

4. The aforementioned computing device is A knowledge information registration process is performed in which the degree of association between the acquired candidate chunks for correction and the parts of the document other than the candidate chunks for correction is calculated using a predetermined algorithm. In the aforementioned related chunk acquisition process, the related chunks are acquired based on the calculated degree of relevance. The document modification support system according to claim 1.

5. The aforementioned computing device is In the knowledge information registration process described above, for each category to which past modifications to the document belong, the degree of relevance between each part of the document to which the modifications belonging to that category were made is calculated using a predetermined algorithm. In the aforementioned related chunk acquisition process, the related chunks are acquired based on the calculated relevance levels for each modification category to which the summary of the modification belongs. The document modification support system according to claim 4.

6. The memory device stores knowledge priority information, which is information about the priority criteria between each related chunk. The aforementioned computing device is In the related chunk acquisition process described above, if multiple related chunks with the same degree of relevance are acquired, the priority of the multiple related chunks is determined based on the knowledge priority information. In the prompt input processing, based on the determined priority, some of the multiple related chunks are included in the prompt. The document modification support system according to claim 4.

7. The aforementioned computing device is In the aforementioned correction candidate chunk search process, the document is divided into multiple sections separated by predetermined text, and the correction candidate chunks are searched from the divided multiple sections. In the aforementioned related chunk acquisition process, the document is divided into multiple sections separated by predetermined text, and the related chunks are acquired from the document from the divided multiple sections. The document modification support system according to claim 1.

8. A document modification support method using an information processing device comprising a storage device that stores document information, modification history information which is information on each modification made to the document in the past, and a trained model that outputs text corresponding to the input text, and a processing unit, The aforementioned computing device A revision summary input process that accepts input of a summary of the revisions to be made to the aforementioned document, A process for acquiring similar past cases, which involves obtaining similar past cases from the revision history information using a predetermined algorithm, where the parts of the revisions that are identical or similar to the revisions indicated in the summary of the revisions, A correction candidate chunk search process searches the document for correction candidate chunks, which are the parts of the document that are subject to correction as indicated in the summary of the corrections, Related chunk acquisition process: Acquires related chunks, which are parts of the document other than the acquired modification candidate chunk, using a predetermined algorithm, where related content is registered with the acquired modification candidate chunk. By inputting the aforementioned correction summary, the acquired similar past examples, the acquired correction candidate chunks, and the acquired related chunks into the trained model, a prompt input process is executed that outputs the portion of the document to be corrected as indicated by the correction summary, and the content of the correction to be made to that portion. Document revision support methods.

Images