Document generation system and program

Abstract

The objective is to provide a document generation system and program that can generate documents with high accuracy by setting appropriate input information conditions. [Solution] The document generation system according to the present invention is characterized by comprising: acquisition means for acquiring business information relating to business content; setting means for setting conditions for input information necessary for document creation based on the business information acquired by the acquisition means; input means for acquiring input information based on the conditions set by the setting means and input by a user; and generation means for referencing a generation model learned using training data that uses the input information as input data and document information relating to the document as output data, and generating document information relating to the document based on the input information acquired by the input means.

Description

Technical Field

[0001] The present invention relates to a document generation system and a program.

Background Art

[0002] Conventionally, a system that generates a document using AI based on input information has been utilized (see Patent Document 1).

[0003] In Patent Document 1, there is an answer system that generates an answer to an inquiry, which has a learning unit that is machine-learned by the inquiry, general information, the answer, and a satisfaction index regarding the answer, and based on the learning result of the learning unit, an AI answer system is disclosed that includes an answer generation unit that generates an answer to be transmitted to each mobile terminal for an inquiry from a plurality of mobile terminals.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the AI answer system disclosed in Patent Document 1 does not assume setting appropriate input information conditions according to the generated document. As a result, in the AI answer system disclosed in Patent Document 1, since appropriate input information conditions are not set, appropriate input information for the generated document cannot be obtained, and there is a problem that the document cannot be generated with high accuracy.

[0006] Therefore, the present invention has been devised in view of the above problems, and an object thereof is to provide a document generation system and a program that can generate a document with high accuracy by setting appropriate input information conditions. [Means for solving the problem]

[0007] The document generation system according to the first invention comprises: acquisition means for acquiring business information relating to business content; setting means for setting condition information indicating the conditions for input information necessary for generating documents relating to the business based on the business information acquired by the acquisition means; input means for acquiring input information based on the condition information set by the setting means and input by a user; and generation means for generating document information based on the input information acquired by the input means, by referring to a generation model learned using training data which uses the input information as input data and document information relating to the document as output data. The setting means sets table of contents information indicating the table of contents of a document based on the business information acquired by the acquisition means, and the generation means refers to a generation model trained using training data that takes input information and table of contents information as input data and document information as output data, and generates the document information based on the input information and table of contents information acquired by the input means. It is characterized by the following.

[0009] The 2 The document generation system according to the invention is further characterized in that, in the first invention, it comprises a storage means for storing document information and a search means for searching the document information stored by the storage means based on the business information and the input information.

[0010] The 3 The document generation system according to the invention is characterized in that, in the first invention, the input means corrects the input information entered by the user.

[0011] The 4 The business plan setting system relating to the invention is the 1 In the invention, the generation means is characterized by referring to a generation model that has learned the degree of association based on the scoring information which includes the classification information determined based on the words contained in the learning data.

[0012] The 5 The document generation system according to the invention is characterized in that, in the first invention, it refers to a condition model learned using learning data in which business information is input data and conditions are output data, and sets the condition information based on the business information acquired by the acquisition means.

[0013] The6 The document generation program according to the invention causes a computer to execute the following steps: an acquisition step to acquire business information relating to business content; a setting step to set condition information indicating the conditions for input information necessary for generating a document relating to the business based on the business information acquired in the acquisition step; an input step to acquire input information based on the condition information set in the setting step, which is input by the user; and a generation step to generate document information based on the input information acquired in the input step, by referring to a generation model that has been trained using training data which uses the input information as input data and document information relating to the document as output data. The setting step sets table of contents information indicating the table of contents of a document based on the business information acquired in the acquisition step, and the generation step refers to a generation model trained using training data that takes input information and table of contents information as input data and document information as output data, and generates the document information based on the input information and table of contents information acquired in the input step. It is characterized by the following. [Effects of the Invention]

[0014] First Invention ~ 6 According to the invention, the document generation system and program set conditions for input information necessary for document creation based on business information, and generate document information based on input information based on the conditions entered by the user. This allows for highly accurate document generation by setting appropriate input information conditions.

[0015] In particular, 1 According to the invention, the document generation system generates document information based on input information and table of contents information. This allows for the generation of a document for each table of contents item, thus enabling the generation of appropriate documents according to the purpose.

[0016] In particular, 2 According to the invention, the document generation system searches for document information based on information input by the user. This allows for searching through stored documents, and furthermore, the retrieved documents can be used for learning and other purposes.

[0017] In particular, 3 According to the invention, the document generation system corrects the input information entered by the user. This allows for accurate document generation, for example, even when the user enters irregular sentences, by correcting the input information.

[0018] In particular, according to the 4 invention, the document generation system refers to the relationship table and sets the condition information. Thereby, according to the business information, it is possible to set the conditions of the appropriate input information.

[0019] In particular, according to the 5 invention, the document generation system refers to the condition model and sets the condition information based on the business information. Thereby, according to the business information, it is possible to set the conditions of the appropriate input information more flexibly.

Brief Description of Drawings

[0020] [Figure 1] FIG. 1 is a schematic diagram showing an example of the document generation system in the present embodiment. [Figure 2] FIG. 2(a) is a schematic diagram showing an example of the configuration of the user terminal of the document generation system in the present embodiment. FIG. 2(b) is a schematic diagram showing an example of the configuration of the server of the document generation system in the present embodiment. [Figure 3] FIG. 3 is a schematic diagram showing an example of the functions of the document generation system in the present embodiment. [Figure 4] FIG. 4 is a schematic diagram showing an example of the relevance of the generation model of the document generation system in the present embodiment. [Figure 5] FIG. 5 is a flowchart showing an example of the operation of generating the generation model of the document generation system in the present embodiment. [Figure 6] FIG. 6 is a flowchart showing an example of the operation of generating the plan information of the document generation system in the present embodiment.

Embodiments for Carrying out the Invention

[0021] Hereinafter, with reference to the drawings, an example of a document generation system 100, a document generation method, and a document generation program as embodiments of the present invention will be described in detail. Note that the configurations in each figure are schematically represented for illustrative purposes, and the size of each component, the size comparison between components, etc., may differ from those in the figures.

[0022] An example of the configuration of the document generation system 100 in this embodiment will be described with reference to Figures 1 to 3.

[0023] The document generation system 100 includes, for example, a user terminal 1, a server 2, and a communication network 9, as shown in Figure 1. The document generation system 100 receives input information from user U via the user terminal 1 and outputs document information.

[0024] User U is a user of the document generation system 100. User U refers, for example, to a user who generates documents using the document generation system 100. User U may also be a company.

[0025] In this embodiment, a database storing models such as large-scale language models is stored on server 2, and the document generation system 100 generates document information on server 2, but this is not the only example. The document generation system 100 may also have a database storing large-scale language models on user terminal 1, and generate document information on user terminal 1, in which case server 2 is not required.

[0026] <User Terminal 1> User terminal 1 is a terminal that controls the generation of document information. User terminal 1 is a terminal operated by user U. User terminal 1 communicates with server 2 via communication network 9.

[0027] User terminal 1, as shown in Figure 2(a), for example, comprises a housing 10, a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a storage unit 104, and I / Fs 105 to 107. Each component 101 to 107 is connected by an internal bus 110. As user terminal 1, known electronic devices such as laptop PCs, desktop PCs, tablet terminals, and smartphones can be used.

[0028] The CPU 101 controls the entire user terminal 1. The ROM 102 stores the operating code of the CPU 101. The RAM 103 is a working area used when the CPU 101 is operating. The storage unit 104 stores various information such as backups of the data stored in the ROM 102 and databases. As the storage unit 104, a data storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive) can be used. For example, the user terminal 1 may have a GPU (Graphics Processing Unit) which is not shown.

[0029] I / F105 is an interface for sending and receiving various information with server 2 via communication network 9 as needed. I / F106 is an interface for sending and receiving information with input unit 108. For example, a keyboard or mouse can be used as input unit 108, and user U inputs various information via input unit 108. I / F107 is an interface for sending and receiving various information with display unit 109. Display unit 109 displays various information stored in storage unit 104. A display can be used as display unit 109, and if it is a touch panel type, it is provided together with input unit 108. Display unit 109 displays various information generated by each component of the document generation system 100.

[0030] The display unit 109 displays various types of information, such as document information. The display unit 109 may display document information on a monitor, for example. The display unit 109 may also display the result of converting document information into a format that user U can understand, for example, using a display format pre-stored in the storage unit 104, etc.

[0031] User terminal 1, as shown in Figure 3 for example, comprises an acquisition unit 11, a setting unit 12 connected to the acquisition unit 11, a presentation unit 13 and a generation unit 16 connected to the setting unit 12, and an input acquisition unit 14 and a storage unit 15 connected to the generation unit 16. Each configuration of user terminal 1 is realized by the CPU 101 executing programs stored in ROM 102, storage unit 104, etc., using RAM 103 as a working area.

[0032] <Acquisition part 11> The acquisition unit 11 acquires various types of information. The acquisition unit 11 sends and receives information to, for example, the server 2. The acquisition unit 11 transmits information acquired, generated, evaluated, or stored by, for example, the user terminal 1 to the server 2 via the communication network 9. The acquisition unit 11 receives information acquired, generated, or stored by, for example, the server 2 via the communication network 9.

[0033] Furthermore, the acquisition unit 11 acquires business information related to the content of the work, for example, by receiving input from user U via the input unit 108.

[0034] <Settings section 12> The setting unit 12 sets condition information that indicates the conditions for the input information necessary for document creation, based on the business information acquired by the acquisition unit 11.

[0035] <Input acquisition unit 14> The input acquisition unit 14 acquires input information such as generation requests, which are queries (query data) that request the generation of document information from a model such as a large-scale language model. The acquisition unit 11 acquires input information, for example, by receiving input from user U via the input unit 108.

[0036] <Generation part 16> The generation unit 16 generates document information. The generation unit 16 executes a generation request obtained by the input acquisition unit 14 to a database 7 in which models such as large-scale language models for generating document information are pre-stored. The generation unit 16 also refers to the database 7 in which training data is stored and generates a generation model 71.

[0037] <Presentation part 13> The display unit 13 displays various types of information. The display unit 13 may, for example, display condition information and document information on the display unit 109.

[0038] <Storage section 15> The memory unit 15 stores various types of information. It may also store the database 7 and the generation model 71.

[0039] <Server 2> Server 2 stores a database containing pre-stored models, such as large-scale language models. Server 2 communicates with user terminal 1 via a communication network 9. Server 2 performs information generation, storage, and transmission / reception in response to processing requests received from user terminal 1 via the communication network 9.

[0040] Server 2 comprises, for example, a chassis 20, a CPU 201, a ROM 202, a RAM 203, a storage unit 204, and an I / F 205, as shown in Figure 2(b). Each component 201-205 is connected by an internal bus 210. As Server 2, a well-known electronic device such as a laptop PC or a desktop PC can be used.

[0041] The CPU 201 controls the entire server 2. The ROM 202 stores the operating code for the CPU 201. The RAM 203 is a working area used when the CPU 201 is operating. The storage unit 204 stores various information such as backups of the data stored in the ROM 202 and databases. As the storage unit 204, a data storage device such as an HDD or SSD can be used. For example, the server 2 may also have a GPU (not shown). The I / F 205 is an interface for sending and receiving various information with the user terminal 1 via the communication network 9 as needed.

[0042] <Communication Network 9> The communication network 9 is, for example, the Internet network to which user terminal 1 and server 2 are connected via a communication circuit. The communication network 9 may consist of a so-called optical fiber communication network. In addition, the communication network 9 may be implemented using known communication technologies such as wired communication networks or wireless communication networks.

[0043] <Database 7> Database 7 is a collection of data pre-stored in the document generation system 100. Database 7 is pre-stored, for example, in a storage unit 204 of server 2. Database 7 stores a generation model 71 for generating information in advance.

[0044] Next, the generation model 71 used in the document generation system 100 in this embodiment will be described.

[0045] <Generative Model 71> The generative model 71 is a foundational model that outputs document information by taking input information as input. The generative model 71 is a type of natural language processing model used, for example, by a document generation system 100 to automatically generate document information. The generative model 71 may be, for example, a known large-scale language model (LLM) that has been pre-machine-trained using a large amount of text data.

[0046] A large-scale language model (LLM) is a deep learning model that pre-trains a language model, which models human spoken language based on its probability of occurrence, using a vast amount of data. In other words, a large-scale language model is a natural language processing model trained using a large amount of text data. For example, it takes sentences as input information and outputs sentences as document information. When a large-scale language model is applied to a question-and-answer system, a question is input to the LLM, and the LLM outputs an answer as document information. The large-scale language model statistically estimates the probability of generating the next word from the sentence contained in the received prompt and sends the estimation result to the requester.

[0047] A large-scale language model, upon receiving text data (prompts), statistically estimates the probability of generating the next word from the text contained in the received prompt and outputs a response based on the estimation result. As a large-scale language model, publicly known technologies can be employed, such as those described on internet sites like "https: / / chatgpt-lab.com / n / n418d3aa56f0b" and "https: / / agirobots.com / chatgpt-mechanism-and-problem / ".

[0048] Specifically, the generative model 71 may include GPT-3, GPT-3.5, or GPT-4 related to "GPT (Generative Pre-trained Transformer) (registered trademark)", "BERT (Bidirectional Encoder Representations from Transformers)", "LaMDA (Language Model for Dialogue Applications)", "PaLM (Scaling Language Modeling with Pathways)", "LLaMA (Large Language Model Meta AI)", etc.

[0049] Furthermore, the generative model 71 may be a base model. The base model may be a natural language model. The base model may be a model generated by unsupervised learning. The base model may also be a generative AI. The natural language model may be an interactive, chat-type or conversational model that alternately receives input information such as instructions and generates response information such as responses.

[0050] Furthermore, the generative model 71 may be generated using machine learning, for example, a neural network model. The generative model 71 is, for example, an AI neural network. The generative model 71 is trained using machine learning, for example, a neural network model such as a CNN (Convolutional Neural Network), or any other model may be used. In addition, the generative model 71 may be generated using methods such as Retrieval-Augmented Generation (RAG), Seq2Seq (Sequence To Sequence), Linear Discriminant Analysis, Support Vector Machines, k-Nearest Neighbors, Random Forest, Deep Learning, etc.

[0051] In such cases, the generative model 71 stores a correlation between the input data (input information) and the output data (document information), as shown in Figure 4, for example. Scoring information may also be used as input data. In such cases, morphemes or words included in the input information may also be used as input data. The correlation degree indicates the degree of connection between the input data and the output data; for example, a higher correlation degree indicates a stronger connection between the data. The correlation degree can be expressed as a percentage or other value with three or more levels, or as a binary value or in two levels. The input information and document information used to learn the correlation degree are, for example, information acquired in advance for use as training data, but are not limited to this, and information acquired at any time may be used. Such information including the correlation degree becomes what is called training data in artificial intelligence. Such training data is learned in advance, and then new input information is actually input into the generative model 71 to output document information.

[0052] The input information is information necessary for document creation and may be in text or audio format. The input information may also be prompts. Furthermore, the input information may be images or videos entered by the user. The input information may also be information obtained by morphological analysis of a question. The input information may be audio or image format converted to text format using speech recognition or image recognition. The input information may be image format converted to text format, such as the names of objects or places, based on features calculated using R-CNN (Region Based Convolutional Neural Networks), YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), etc. The input information may also include queries requesting processing. Furthermore, the input information may be information about a company. Company information may include, for example, the company name, product name, address, history, employee information, and business activities. Company information may also include management information such as sales and performance. The input information may include both internal and external information. Internal information may include internal company performance data, financial information, and information indicating management philosophy. External information includes information such as external market data, economic indicators, competitive information, and industry trends.

[0053] Document information is information about a document. For example, document information may be a document introducing a company. Document information may also be advertising information for a company or product. Document information may also be information from a company's website. Document information may be in text, image and video, or audio format. Furthermore, document information may include table of contents information, which shows the table of contents for multiple sections of the document, such as company information, management philosophy, product introduction, greeting, purpose, challenges, results, countermeasures, and summary. Table of contents information is information that shows the table of contents, which is the structure of the document to be generated.

[0054] Next, we will describe the details of the process for generating the generative model 71. Figure 5 is a flowchart showing the process for generating the generative model 71.

[0055] First, in step S1, the acquisition unit 11 acquires training data. Training data is information used for training. Training data is, for example, past input information. Training data may also be, for example, past statements by managers, presentation materials, meeting minutes, emails, etc. Training data is text data, but is not limited to this, and may also be image or audio data. Furthermore, training data may include past document information. In addition, the acquisition unit 11 may acquire training data to which qualitative information including one or more of the following is attached: time-series information, performance indicator information, philosophy information, and presenter information. Furthermore, the acquisition unit 11 may acquire information indicating the management status of user U. In addition, training data may also be information that is a combination of input data and output data. For example, training data may be information in which input information, which is input data, and words or semantic features of this input information are linked to document information, which is output data, and words or semantic features of this document information, according to the degree of correlation.

[0056] Qualitative information includes one or more of the following: time-series information, performance indicator information, philosophy information, and presenter information. Time-series information is information that shows the time series in which the training data was created. Time-series information may also be information that shows the amount of change in the characteristics of training data from different time series. For example, time-series information may include the date and time when the content shown in the training data was created, the date and time of the manager's statement, the date and time of the meeting minutes, etc. Presenter information is information about the person and company that presented the training data, or the person and company that created the training data. For example, presenter information may be information about the person and company that made the statement shown in the training data. Performance indicator information is an indicator that shows the effectiveness of the content shown in the training data. Performance indicator information is determined based on the business results based on the content shown in the training data. Business results include the success or failure of past business decisions, financial indicators, and achievements in performance improvement, etc. For example, performance indicator information may be determined based on a relationship table that shows the relationship between business results and effectiveness. Philosophy information is information that shows the degree of agreement with User U's management philosophy. Philosophy information is, for example, an index that shows the degree of agreement between the pre-entered vision, mission, and strategic goals of user U and the content shown in the learning data. Alternatively, philosophy information may also show the degree of agreement with the presenter's attributes. These attributes may include the presenter's job title, industry, emotions, and tendencies.

[0057] Furthermore, the training data may also be information relating to natural science. The training data may also be a tool that performs the aforementioned simulations relating to natural science based on the input parameters. As for simulations relating to natural science, tools capable of simulating physical phenomena, chemical phenomena, and even all kinds of natural sciences may be used, such as FEM simulators that analyze stress fields and strains based on the finite element method, chemical reaction simulations, chemical engineering simulations including plants, electrical circuit simulations, fluid dynamics simulations, and quantum mechanics simulations.

[0058] Furthermore, the learning data may consist of keyword information indicating keywords and related information related to those keywords. Keyword information may, for example, be information indicating a word, but it may not be limited to words and may include information indicating any word such as a person, place name, or event. Related information is information related to the word indicated by the keyword information. Related information may, for example, be an explanation and description of the word indicated by the keyword information, or questions and answers related to the word indicated by the keyword information. The acquisition unit 11 may store the acquired learning data in any memory such as the storage unit 16.

[0059] Furthermore, the training data may also be information related to marketability. The training data may also be a tool that performs the market simulations described above based on the input parameters. The market simulator is a tool that performs the market simulations described above based on the input parameters. This market simulator may utilize any existing tool. Examples of information implemented in the market simulator include financial statements of listed companies, expert comments and opinions on futures forecasts and reasons for stock price increases and decreases published in analyst reports and newspaper articles. Other examples of information implemented in the market simulator may include expert opinions on the Nikkei 225 futures as a whole, or opinions on specific segments or industries, or even opinions on individual futures, as well as comments and forecasts of increases or decreases from experts (analysts) published on the internet. Furthermore, examples of information implemented in the market simulator may include price movements of interest rates, futures, exchange rates, stock prices of individual securities, crude oil, precious metals, Bitcoin, etc., as well as sequential charts and line graphs, Bollinger Bands, trading volume, MACD, moving averages, and other information. Examples of information implemented in a market simulator may include various data related to politics, economics, society, and technology, such as GDP, employment statistics, industrial production index, capital investment, labor force survey, business cycle index, consumer spending, new car sales, and consumer price index. Furthermore, examples of information implemented in a market simulator may include information reflecting changes in the market environment, such as new technologies of interest to companies, digital transformation, and the latest industry trends.

[0060] Next, in step S2, the setting unit 12 determines classification information, including the classification of the training data. The classification information is information that includes the classification of the training data, and may be information such as fields, people, companies, events, etc., used to identify the classification of the training data. The classification information may also be words indicating classifications, such as "customer relationships" or "financial strategy." The classification information may also be information that shows the classification vector of the training data. The classification vector of the training data may be, for example, the appropriateness of the classification of the training data, or coordinates on the classification division. The setting unit 12 may determine the classification information based on the input information, but is not limited to this, and may also determine the classification information based on words included in the training data, for example. The setting unit 12 may, for example, refer to a correspondence table showing the relationship between words and classification information and determine the classification information for a word.

[0061] Furthermore, classification information may include classifications such as financial strategy, marketing strategy, risk management, organizational management, human resource development, technological innovation, corporate culture, management philosophy, market trends, business succession, and business continuity. Financial strategy is a classification related to a company's profitability, cost management, and investment plans. Marketing strategy is a classification related to a company's market development, such as market research, promotion, and brand strategy. Risk management is a classification related to risks faced by a company, such as market risk, credit risk, and operational risk. Organizational management is a classification related to a company's internal operations, such as organizational structure, internal control, personnel strategy, and business processes. Human resource development is a classification related to the growth and development of human resources, such as employee training, leadership development, and successor training. Technological innovation is a classification related to technological transformation and progress, such as the adoption of new technologies, research and development, and innovation strategies. Corporate culture and management philosophy is a classification related to a company's values, vision, mission, and management philosophy. Market trends are a classification related to overall market trends, such as industry trends, competitive analysis, and changes in the economic environment. Business succession and business continuity are classifications related to the long-term operation of a company, including successor issues, business succession plans, and business continuity strategies.

[0062] Furthermore, in step S2, the configuration unit 12 extracts semantic features from the training data. Semantic features are information that indicates, for example, the meaning of words or word usage, and their trends, frequencies, etc. The configuration unit 12 generates feature vectors from the text data included in the training data using language analysis methods such as TF-IDF (term frequency-inverse document frequency), Word2Vec, or BERT (Bidirectional Encoder Representations from Transformers), and extracts semantic features for each document or utterance. The configuration unit 12 also adds keyword information related to the qualitative information attached to the training data. The configuration unit 12 may also extract keywords using a custom dictionary or a dedicated embedding model to emphasize specialized terms and phrases related to management philosophy and tacit knowledge.

[0063] Furthermore, in step S2, the setting unit 12 may cluster the training data and the multidimensional data indicated by the various information attached to the training data. The setting unit 12 may apply, for example, K-means, hierarchical clustering, or dynamic clustering methods such as DBSCAN (Density-based spatial clustering of applications with noise) to cluster the multidimensional data based on its similarity. Alternatively, similarity calculations such as cosine similarity and vector distance may be used to determine the similarity of the data.

[0064] Furthermore, in step S2, the configuration unit 12 may perform normalization and cleansing of the training data as preprocessing for classification. The configuration unit 12 may, for example, remove unnecessary symbols and noise contained in the training data and unify and normalize the writing style. The configuration unit 12 may also perform morphological analysis of the text contained in the training data. The configuration unit 12 may, for example, use any tool to perform word segmentation, part-of-speech tagging, stop word removal, etc., to prepare the text in a format that can be analyzed.

[0065] Next, in step S3, the setting unit 12 determines scoring information including training data and classification information. Scoring information is information including training data and classification information. Scoring information may include information regarding the importance of the training data. Scoring information may include time-series information showing the time series of the training data. Scoring information may be determined based on presenter information regarding the presenter of the training data. Scoring information may also be determined based on time-series information, performance indicator information, philosophy information, or presenter information. The setting unit 12 may determine the scoring information based on the input information, but is not limited to this; for example, it may determine it based on information linked to the training data. This makes it possible to respond to changes in key points over time.

[0066] The setting unit 12 may determine scoring information by assigning classification information such as "financial strategy" or "customer relationships" to the training data. Alternatively, the setting unit 12 may determine scoring information by assigning position or vector information in a two-dimensional space consisting of time series and fields to the training data. The setting unit 12 may also determine scoring information by assigning a score indicating importance to the training data. The setting unit 12 stores the scoring information in the storage unit 15.

[0067] Furthermore, the setting unit 12 may refer to a classification model trained using pre-trained data that takes reference training data as input and scoring information as output, and determine the classification information or scoring information for the training data obtained in step S1. The reference training data is training data that has been acquired in advance for use in training the classification model.

[0068] Furthermore, the setting unit 12 may determine scoring information based on qualitative information attached to the training data. In such a case, the setting unit 12 may, for example, refer to a relationship table between presenters indicated by the presenter information and scores indicating importance, and determine scoring information by attaching scores indicating importance to the training data. Alternatively, the setting unit 12 may, for example, refer to a classification model trained using pre-trained data that takes reference training data and reference presenter information as inputs and scores information as output, and determine scoring information for the acquired training data and presenter information in step S1. Reference presenter information is presenter information acquired in advance for use in training the classification model.

[0069] Furthermore, the setting unit 12 may calculate a score for each data point within the clustered clusters of the multidimensional data. The setting unit 12 may also calculate a score by integrating multiple indicators such as the frequency of occurrence of each data point, the freshness of the statement, the reliability of the presenter, and the degree of relevance to the management philosophy using a weighted average or regression model. The setting unit 12 may calculate the score such that, for example, the more recent the time series indicated by the time series information, the higher the score. The setting unit 12 may also calculate the score such that, for example, the score is higher according to the attributes and achievements of the presenter indicated by the presenter information. The setting unit 12 may also calculate the score such that, for example, the score is higher according to the degree of agreement between the user U's statements and the management philosophy indicated by the philosophy information.

[0070] Next, in step S4, the generation unit 16 refers to the database 7 in which the scoring information generated in step S3 is stored and generates a generation model 71. The generation unit 16 stores the generated generation model 71 in the database 7, for example.

[0071] The above steps complete the process of generating the generative model 71. Furthermore, the generative model 71 is not limited to the above configuration and may be generated using any equipment. The generative model 71 can generate appropriate document information according to the importance of the scoring information, for example, by setting a high probability of generating words in the training data with high scores, or by setting a high degree of correlation between words. This makes it possible to generate a generative model 71 that enables appropriate output according to the purpose using the scoring information. Moreover, it becomes possible to generate a generative model 71 that provides answers that are not merely statistical, but are based on past experience and principles, and can flexibly adapt to changing times.

[0072] Furthermore, by integrating diverse data such as time-series information, presenter information, and qualitative management philosophy, and combining dynamic clustering and priority scoring, it is possible to generate a generative model 71 with technical novelty and high effectiveness. This makes it possible to reflect manager-specific judgment criteria and tacit knowledge, which was difficult with conventional methods based on a single training dataset.

[0073] Next, as a method for generating documents, an example of the operation of the document generation system 100 to generate document information will be described. The document generation system 100 is executed, for example, via a document generation program installed in the user terminal 1 and server 2.

[0074] As shown in Figure 6, in step S11, the acquisition unit 11 acquires business information related to the content of the work. The acquisition unit 11 acquires business information entered by, for example, user U. The business information is information related to the content of the work, such as information indicating tasks such as website creation, daily report creation, presentation material creation, report creation, daily report creation, and management strategy document creation. In addition, in step S11, the acquisition unit 11 may acquire history information related to user U's operation history. The history information may be, for example, information such as document information previously generated by user U, operations and processes performed in the past, input information previously entered, and the frequency of these.

[0075] Next, in step S12, the setting unit 12 sets condition information indicating the conditions for input information necessary for generating business-related documents, based on the business information obtained in step S11. The condition information may, for example, be information indicating questions about the input information. The condition information may, for example, be information indicating company information and questions about its strengths or weaknesses. The setting unit 12 outputs the set condition information to the presentation unit 13 and presents it to the user U.

[0076] Furthermore, in step S12, the setting unit 12 may, for example, refer to a relationship table between previously acquired business information and condition information based on business information, and set the condition information based on the business information acquired in step S11. In such a case, the setting unit 12 may set condition information linked to business information that matches or is similar to the business information acquired in step S11 from the relationship table stored in the storage unit 15.

[0077] Furthermore, in step S12, the setting unit 12 may refer to a condition model trained using training data that takes business information as input data and conditions as output data, and set condition information based on the business information obtained in step S11. The condition model may be trained using training data that takes business information as input data and conditions as output data, and may be generated in the same way as the generation model 71.

[0078] Furthermore, in step S12, the setting unit 12 may perform a table of contents setting process to set table of contents information based on business information. In step S12, the setting unit 12 may, for example, refer to a relationship table between previously acquired business information and table of contents information and set the table of contents information based on the business information acquired in step S11. In such a case, if the business indicated by the business information is the creation of a business report, the table of contents information to be set will include table of contents information such as "1. Business Overview", "2. Implementation Details", and "3. Issues and Countermeasures". In such a case, the condition information may also include input conditions set for each section of the table of contents. Furthermore, in step S12, the setting unit 12 may refer to scoring information and set the table of contents information based on business information. Furthermore, in step S12, the setting unit 12 may set the table of contents information based on business information and condition information. In step S12, the setting unit 12 may, for example, refer to a relationship table between previously acquired business information, condition information, and table of contents information and set the table of contents information based on the business information and condition information acquired in step S11. The setting unit 12 may also present the configured table of contents information to the user U.

[0079] Next, in step S13, the input acquisition unit 14 acquires input information based on the conditions set in step S12, which is input from user U. The input acquisition unit 14 outputs the input information to the generation unit 16.

[0080] Furthermore, in step S13, the input acquisition unit 14 may perform a correction process to correct the input information based on the conditions set in step S12, which have been input by the user U. This allows the input information to be corrected based on the format structure even if the user inputs irregular sentences.

[0081] Next, in step S14, the generation unit 16 refers to the generation model and generates document information based on the input information obtained in step S13.

[0082] Furthermore, in step S14, the generation unit 16 may refer to a generation model trained using training data in which input information is used as input data and document information is used as output data, and set the document information based on the input information obtained in step S13.

[0083] Furthermore, in step S14, the generation unit 16 may refer to a generation model trained using training data that takes input information and table of contents information as input data and document information as output data, and set document information based on the input information obtained in step S13 and the table of contents information set in step S12. Also, in step S14, the generation unit 16 may use input information based on the conditions set in step S12, for example, input from user U, as search information and perform a search process described later. The generation unit 16 may also generate new document information based on past document information retrieved by the search process. For example, if "2023 Customer Presentation Materials" is retrieved by the search process, the generation unit 16 will supplement the differences with the input information based on the structure of the materials, such as the customer name, purpose, and proposal content, and generate a "Presentation Material Draft that can be reused as the 2025 version". In such a case, in step S14, the generation unit 16 searches for document information stored in the storage unit 15 based on the business information and input information. The generation unit 16 searches for document information that is highly relevant to the business information and the input information, and generates new document information based on the input information and the searched document information. This makes it possible to generate specific and useful documents based on the search information, such as FAQ answers, contract drafts, and meeting minutes summaries.

[0084] Next, in step S15, the presentation unit 13 presents the document information generated in step S14 to the user U.

[0085] Furthermore, the storage unit 15 may store the generated document information. The storage unit 15 may also store the generated document information linked to attribute information such as category, generation date and time, and creator name. The storage unit 15 may perform a search process to search for document information stored by the storage unit 15 based on search information indicating search information entered by user U. The storage unit 15 may search for document information that contains content that matches or is similar to the search content indicated by the search information, or document information linked to attribute information that matches or is similar to the search content indicated by the search information. The storage unit 15 may search for document information by means of semantic search, for example. Furthermore, the document information may be shareable and editable.

[0086] Furthermore, in step S12, the setting unit 12 may set the processing steps based on the history information acquired in step S11. The processing steps indicate whether or not to perform processes such as table of contents setting, correction, and search. The setting unit 12 may also refer to a processing model learned using training data that takes history information and business information as input data and processing steps as output data, and set the processing steps based on the history information and business information acquired in step S11.

[0087] Furthermore, the generation unit 16 retrains the generation model 71 based on evaluation information indicating an evaluation of the document information generated in step S14. The generation unit 16 retrains the generation model 71 based on, for example, feedback of evaluation information for the document information input by user U. The evaluation information may be, for example, an evaluation score for each item. The generation unit 16 may, for example, assign weights to items that received low scores. The generation unit 16 may learn using, for example, focal loss, hard negative mining, etc. The generation unit 16 may also incorporate the document information generated in step S14 as additional context to the generation model.

[0088] Furthermore, the generation unit 16 may retrain the generation model 71 if, for example, the evaluation of the document information generated in step S14 falls below a certain value, or if a specific error frequently occurs in each step. The generation unit 16 may also retrain at a fixed interval, such as one day or one week.

[0089] Furthermore, the generation unit 16 may perform batch updates or fine-tuning during the retraining of the generative model 71. Alternatively, the generation unit 16 may use a hybrid method combining online learning and batch updates during the retraining of the generative model 71, for example, performing online updates at the end of each user U session while simultaneously performing batch updates when a certain number of feedback items are reached. Additionally, the generation unit 16 may set a small learning rate during the initial online updates to suppress rapid parameter fluctuations, and optimize the learning rate using validation data during periodic batch updates. Furthermore, the generation unit 16 may assign higher weights to samples that receive low ratings in the feedback during the retraining of the generative model 71, focusing on correcting their errors.

[0090] After performing each of the steps described above, the document generation operation of the document generation system 100 in this embodiment is completed. The document generation system 100 and the program set the conditions for the input information necessary for document creation based on the business information, and generate document information based on the input information based on the conditions entered by the user U. As a result, by setting appropriate input information conditions, documents can be generated with high accuracy.

[0091] Furthermore, the document generation system 100 in this embodiment may seamlessly link the modules for collecting, analyzing, and retraining feedback data via an API (Application Programming Interface) or a message queue such as RabbitMQ or Kafka. This makes it possible for the model to retrain in real time based on the response from the user U.

[0092] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of Symbols]

[0093] 100 Document Generation Systems 1 User terminal 10 cabinets 101 CPU 102 ROM 103 RAM 104 Preservation Department 105~107 I / F 108 Input section 109 Display section 110 Internal bus 11 Acquisition Department 12. Settings section 13 Presentation section 14 Input acquisition unit 15 Storage section 16 Generation part 2 servers 20 cabinets 201 CPU 202 ROM 203 RAM 204 Preservation Department 205 I / F 210 Internal bus 7 Databases 71 Generative Models 9. Communication Network U User

Claims

1. Means of obtaining business information related to the content of work, A setting means for setting condition information that indicates the conditions for input information necessary for generating documents related to the business, based on the business information acquired by the acquisition means, An input means for acquiring input information based on condition information set by the setting means entered by the user, The system includes a generation means that references a generation model trained using training data, which takes input information as input data and document information about the document as output data, and generates document information based on the input information acquired by the input means, The setting means sets table of contents information indicating the table of contents of a document based on the business information acquired by the acquisition means. The generation means refers to a generative model trained using training data that takes input information and table of contents information as input data and document information as output data, and generates the document information based on the input information and table of contents information obtained by the input means. A document generation system characterized by the following:

2. A storage means for storing document information, The system further comprises a search means for searching document information stored by the storage means based on the aforementioned business information and the aforementioned input information. A document generation system according to claim 1, characterized by the following:

3. The input means corrects the input information entered by the user. A document generation system according to claim 1, characterized by the following:

4. Based on the business information obtained by the aforementioned acquisition means, the condition information is set by referring to a table showing the relationship between the previously acquired business information and the conditions of the input information. A document generation system according to claim 1, characterized by the following:

5. The condition model is trained using training data in which business information is input data and conditions are output data, and the condition information is set based on the business information acquired by the acquisition means. A document generation system according to claim 1, characterized by the following:

6. Steps for obtaining business information related to the content of work, A setting step in which condition information indicating the conditions for input information necessary for generating documents related to the business is set based on the business information obtained in the acquisition step, An input step that obtains input information based on condition information set by the setting step entered by the user, The computer is instructed to perform a generation step in which it generates document information based on the input information obtained in the input step, by referring to a generative model trained using training data that takes input information as input data and document information about the document as output data, and by referring to a generative model trained using the training data that takes input information as input data and generates document information based on the input information obtained in the input step. The setting step sets table of contents information indicating the table of contents of the document based on the business information obtained in the acquisition step. The generation step involves referencing a generative model trained using training data that takes input information and table of contents information as input data and document information as output data, and generating the document information based on the input information and table of contents information obtained in the input step. A document generation program characterized by the following:

Images