Work assistance system, control method therefor, and program

Abstract

This work assistance system comprises: an inference unit that inputs work data pertaining to assistance to an inference model and infers assistance information corresponding to the work data; and a generation unit that inputs the assistance information and / or the work data to a generation model based on a large language model and generates generation data corresponding to the assistance.

Description

Business support system, its control method, and program

[0001] The present invention relates to a business support system, its control method, and a program.

[0002] In recent years, in regional support services, systems for quickly sharing information among organizations have been used. In regional support, for example, in dealing with child support, abuse response, child injuries and accident response, maternal and child health, DV (Domestic Violence) response, sexual violence response, delinquency and crime response, developmental disorder response, elderly welfare, and welfare for the disabled, it is necessary to respond based on various target persons and situations.

[0003] For example, in regional support services, the business content is diverse, including communication between support recipients and supporters, cooperation and information sharing inside and outside support organizations, and scrutiny of appropriate support timing and content, and the workload is high.

[0004] Also, it is necessary to consider appropriate information management assuming the past, present, and future, such as the cultivation of supporters' human resources, inheritance of know-how, and handling of personal information.

[0005] For example, in Patent Document 1, in child counseling services, a configuration is disclosed in which indicators related to abuse of children are estimated based on investigation results, and the estimated results are presented and shared.

[0006] Japanese Patent No. 7252688

[0007] Recently, AI-related technologies such as machine learning and generative AI (Artificial Intelligence) have developed and are being utilized in various fields. Such AI-related technologies each have their own characteristics, and by utilizing them based on those characteristics, they are also useful for solving the above-mentioned problems to be considered in regional support services. For example, it is desired to be utilized for early detection, prevention, intervention, and recurrence prevention of various problems related to children who are support recipients.

[0008] In view of the above problems, an object of the present invention is to enable the provision of a function that promotes the improvement of the work efficiency of supporters who provide regional support services and the quality of supporters' judgments.

[0009] To solve the above problems, one embodiment of the present invention has the following configuration. That is, the business support system includes an inference unit that inputs business data related to support into an inference model and infers support information corresponding to said business data, and a generation unit that inputs at least one of the support information and said business data into a generation model based on a large-scale language model and generates generated data corresponding to said support.

[0010] Another embodiment of the present invention has the following configuration: The control method for a business support system comprises: an inference step of inputting business data related to support into an inference model and inferring support information corresponding to said business data; and a generation step of inputting at least one of the support information and said business data into a generation model based on a large-scale language model and generating generated data corresponding to said support.

[0011] Another embodiment of the present invention has the following configuration: a program that causes a computer to perform an inference step of inputting business data related to support into an inference model and inferring support information corresponding to said business data, and a generation step of inputting at least one of the support information and said business data into a generation model based on a large-scale language model and generating generated data corresponding to said support.

[0012] According to the present invention, it is possible to provide functions that promote the efficiency of work and the quality of judgment of supporters providing community support services.

[0013] Configuration diagram showing an example of the configuration of the business support system according to the first embodiment of the present invention. Functional block configuration diagram showing the server terminal according to the first embodiment of the present invention. Functional block configuration diagram showing the user terminal according to the first embodiment of the present invention. Schematic diagram for explaining the data flow in the system according to the first embodiment of the present invention. Schematic diagram for explaining the input and output data of the generation model according to the first embodiment of the present invention. Flowchart of the processing according to the first embodiment of the present invention. Flowchart of the retraining process according to the first embodiment of the present invention.

[0014] Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings and other documents. The embodiments described below are merely examples for illustrating the present invention and are not intended to be interpreted as limiting the invention. Furthermore, not all configurations described in each embodiment are necessarily essential for solving the problems of the present invention. In each drawing, the same components are given the same reference numeral to indicate their correspondence. To avoid unnecessary redundancy and to facilitate understanding by those skilled in the art, some parts of the description may be omitted or simplified. For example, detailed explanations of already well-known matters or redundant explanations of substantially identical configurations may be omitted.

[0015] First, the embodiments for carrying out the present invention will be described assuming support services, which are a type of administrative service. It should be noted that some or all of the functions and components of the embodiments described later are not limited to support services, but may be applied to various fields and services. Furthermore, modifications, adjustments, and expansions may be made as appropriate depending on the field or service to which they are applied.

[0016] Furthermore, the term "personal information" as used in the following explanation includes data whether or not it was collected directly from an individual, and may include data consisting of one or more items. If other types of information are generated or derived based on personal information, such other types of information may also be treated as personal information. While personal information is generally managed under strict restrictions, the embodiments of the present invention assume the provision of support services to government agencies, etc., under management that adheres to such restrictions.

[0017] Furthermore, since the configuration according to the present invention assumes the use of sensitive data that requires careful handling, such as personal information, it assumes a network where user access is restricted. While the method of access restriction is not particularly limited, one example is the use of a closed network not connected to the internet.

[0018] In this embodiment, a "trained model" and "generated AI" are used, which are generated by performing a training process based on machine learning technology. In this specification, to distinguish between them, the trained model used for inference processing is also referred to as the "inference model." Similarly, the generative AI that generates new data based on input data is also referred to as the "generative model." The training algorithms, training methods, and training data used to generate the trained model and generative AI are not limited to those described later and may be adjusted as appropriate.

[0019] An inference model performs inference processing based on input data and outputs the desired output data. Inference models are trained to produce more appropriate output using machine learning, deep learning, and Bayesian modeling optimization. Depending on the level of training, such inference models can produce highly accurate output (inference results) for somewhat formalized data. However, inference models rely on fixed, standardized input data for their inferences. But, for example, users in various support settings consider and analyze data other than standardized input data, such as text data, video data, and family structure. Therefore, it is essential for the training model to utilize data other than the data used for inference. Consequently, methods using only inference models cannot adequately provide the necessary functions in support settings.

[0020] Generative models are Large Language Models (LLMs), built on large datasets, and capable of producing output that takes the context of the input data into account. However, generative models can sometimes produce and output plausible information even if it is not based on facts (hallucination). In other words, it is possible to encounter situations where it is impossible to recognize whether the output from a generative model is true or false.

[0021] The present invention aims to provide a system that improves the operational efficiency of support services, enhances the convenience of support workers' work, and improves the quality of support workers' judgments by combining inference models and generative models while taking the above characteristics into consideration, and by providing support functions specialized for specific domains.

[0022] <First Embodiment> [System Configuration] Figure 1 is a schematic diagram showing an example configuration of a business support system (information processing system) according to the first embodiment of the present invention. The business support system 1 is a system whose main function is to support communication and decision-making with users, and it may be a system that enables input, viewing, and sharing of records inside and outside the organization that performs the work, and performs chat communication between supporters, registration and sharing of photos of people being supported, or simulation of past support trends, etc. Here, as an example of an organization to which this system can be applied, a child guidance center that performs work related to child support will be used as an example, but it is not limited to this. This system may be applied to or configured to link with systems of organizations such as municipal offices, maternal and child health centers, schools, kindergartens, nurseries, medical institutions, police, prosecutors, fire departments, child welfare organizations, NPOs (Non-Profit Organizations), etc. The business support system 1 is composed of a server device 100, a plurality of user terminals 200, and a linkage system 300. Each device constituting the business support system 1 is configured to communicate via a network (NW).

[0023] The server device 100 is a device for providing various functions provided by the business support system 1 to user terminals 200. The server device 100 provides applications for supporting business efficiency to each user terminal 200 and manages information registered through each user terminal 200. Note that some or all of the functions described later may be provided on the user terminal 200 side or on the linked system 300 side, which can be linked via a network NW. The server device 100 may be configured on-premises using a general-purpose computer such as a workstation or personal computer, or it may be logically implemented through cloud computing. In this embodiment, for the sake of explanation, one server device 100 is shown as an example, but it is not limited to this; there may be multiple devices, and there may be servers with different roles, such as an authentication server and a database server.

[0024] The user terminal 200 is an operating terminal used by users such as staff of the child guidance center. The user terminal 200 may consist of an information processing device such as a personal computer, tablet terminal, smartphone, or POS terminal. The user terminal 200 provides users with the functions of Web services provided by the server device 100 and applications installed and running on the user terminal 200. In the example in Figure 1, two user terminals 200a and 200b are shown, but many more user terminals may be used. The configurations of multiple user terminals 200 may be different or the same. Also, the available functions may differ depending on the role and authority of the user using the user terminal 200. For convenience, here they will be described collectively as user terminal 200. When it is necessary to explain each user terminal separately, subscripts (a, b, ...) will be added.

[0025] The linked system 300 is an external system that functions in conjunction with the server device 100 via a network NW. The linked system 300 may be configured, for example, as a child consultation record system, primarily providing functions such as recording child consultations and issuing administrative documents. The child consultation record system may perform functions such as issuing child numbers and reception numbers related to children, issuing medical examination tickets and temporary protection decision notices, managing family information and fees linked to administrative information, and managing the progress of procedures. The configuration of the linked system 300 is not particularly limited, and it may be configured to provide various functions related to business support, for example. Although only one linked system 300 is shown, it is not limited to this configuration and may be composed of one or more devices depending on the functions and services.

[0026] A network (NW) consists of the Internet, intranet, wireless LAN (Local Area Network), WAN (Wide Area Network), etc. The communication standards and wired / wireless connections related to the network (NW) are not particularly limited, and a network (NW) may be configured using a combination of multiple communication standards. As mentioned above, considering the confidentiality of the information handled, the network (NW) may be constructed as a closed network accessible only to designated users, rather than a public network such as the Internet.

[0027] Figure 2 is a block diagram showing an example of the functional configuration of the server device 100 according to this embodiment. The server device 100 is composed of a control unit 110, a communication unit 130, and a storage unit 140. Each part is configured to communicate with each other via an internal bus or the like.

[0028] The control unit 110 is responsible for controlling the operation of the server device 100. The control unit 110 is composed of, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an NPU (Neural Network Processing Unit), etc., and provides various functions by reading and executing various programs and data stored in the memory unit 140. The control unit 110 functions as, for example, a data collection unit 111, a data management unit 112, an inference processing unit 113, a data generation unit 114, a prompt control unit 115, an output data generation unit 116, a generated data evaluation unit 117, a learning data generation unit 118, a RAG data generation unit 119, a retraining processing unit 120, a RAG data update unit 121, a model management unit 122, a display control unit 123, and a communication control unit 124.

[0029] The data collection unit 111 collects data from various databases (hereinafter referred to as "DB") configured in the storage unit 140 based on user instructions. The data collected by the data collection unit 111 is not limited to data stored in the storage unit 140; it may also be collected by querying external systems (for example, the linked system 300).

[0030] The data management unit 112 manages the recording, retrieval, and updating of data for the various databases configured in the storage unit 140. Examples of data managed in each database will be described later.

[0031] The inference processing unit 113 outputs an inference result for the input using the trained model obtained through the learning process. As will be described in detail later, the inference model as a trained model in this embodiment takes standardized data from the business data as input and outputs an inference result as supporting information for said standardized data.

[0032] The data generation unit 114 takes various data and prompts as input to the Large Language Model (LLM) obtained through the learning process and outputs the desired generated data. As will be described in detail later, the generated model as a large language model in this embodiment uses the inference result data inferred by the inference processing unit 113 and business data as input, and generates and outputs data in a predetermined manner related to support work.

[0033] The prompt control unit 115 controls and selects the prompts to be used in the generation model according to this embodiment based on user instructions. The prompts used will differ depending on the type and content of the generated data output by the generation model. The prompts may be predetermined according to the generated data to be output, or multiple prompts may be combined according to the instructions. In this embodiment, RAG (Retrievable-Augmented Generation) technology is used when generating prompts for input to the generation model. An example of using RAG technology according to this embodiment will be described later. Note that RAG technology is just one example, and other technologies may be used as long as they can implement the functions to be realized in this embodiment.

[0034] The output data generation unit 116 generates output data converted into an output format using the inference result data from the inference processing unit 113 and the data generated by the data generation unit 114. For example, the output data may be in file format or a display screen for display in a browser or the like.

[0035] The generated data evaluation unit 117 receives evaluations from the user regarding the data generated by the data generation unit 114 and associates the content of these evaluations with the evaluation results. The evaluation method is not particularly limited, but for example, it may be configured to accept evaluations at multiple stages, or to accept comments regarding excesses, deficiencies, or truthfulness. It may also be configured to accept information regarding RAG data for generating prompts to be input into the generation model.

[0036] The training data generation unit 118 generates training data for retraining the generation model based on the evaluation results of the generated data by the generated data evaluation unit 117. The training data here may consist of pairs of generated data and evaluation result data, or it may consist only of generated data that yielded a certain level of good results or better.

[0037] The RAG data generation unit 119 generates various types of information (hereinafter referred to as "RAG data") that are referenced when generating prompts to be input to the generation model. RAG data is used in combination with pre-registered prompt data, for example, to improve the accuracy of the output of the generation model. RAG data may be generated based on information input by the user based on the evaluation results of the generated data output by the generation model, or it may be generated based on newly collected information based on instructions from the user.

[0038] The retraining processing unit 120 performs a retraining process on the generated data using the training data generated by the training data generation unit 118, and updates the generated model. Note that the retraining is not limited to using only the training data generated by the training data generation unit 118, but may also include training data manually created by the user or others.

[0039] The RAG data update unit 121 updates the data by registering the newly generated RAG data from the RAG data generation unit 119 in the RAG DB 145. Furthermore, the RAG data update unit 121 may perform update processing such as deleting inaccurate or false information from the registered information based on user evaluations, or adding information indicating that the truthfulness of the information is unknown.

[0040] The model management unit 122 manages the inference model used by the inference processing unit 113 and the generation model used by the data generation unit 114. Each model may be used appropriately according to the content to be output. Also, each model may be managed in multiple versions according to the degree of learning. Further, when performing high-order inference by combining multiple generation models, the combination patterns and processing order of those generation models may be managed together.

[0041] The display control unit 123 performs control to display on the screen the display screen generated by the output data generation unit 116, generated data, etc. Also, the display control unit 123 displays a screen for receiving various instructions from the user. For example, the display control unit 123 may provide data of a UI (User Interface) screen for display to the user terminal 200.

[0042] The communication control unit 124 performs communication control with an external device (for example, the user terminal 200 or the cooperation system 300) and performs data transmission and reception. The communication control here may be executed according to, for example, the configuration of a closed network and access restrictions.

[0043] The communication unit 130 is a communication interface for communicating with an external device via the network NW. The communication unit 130 may be configured to be compatible with multiple communication standards according to the configuration of the network NW.

[0044] The storage unit 140 is a storage device for storing programs, data, etc. for executing various control processes and each function in the control unit 110. The storage unit 140 is composed of volatile / non-volatile storage devices such as a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard disk Drive), and a flash memory. The storage unit 140 includes a DB for managing data corresponding to the functions described later.

[0045] The business data DB 141 manages various business data related to support business operations. In the present embodiment, the business data is composed of standardized data and non-standardized data.

[0046] The inference model DB142 manages an inference model as a learned model used when the inference processing unit 113 performs inference processing. The learning process generally has a high processing load. Therefore, from the perspective of processing distribution, the learning process may be performed by a device other than the server device 100 (not shown), and the server device 100 may acquire an inference model in which learning has progressed to a certain extent and hold it in the inference model DB142. The configuration of the inference model and the learning algorithm are not particularly limited, and known methods may be used, but it is configured to be able to perform input and output as described later. As an example, the inference model may be a model based on the methods of Patent No. 7252688 and Patent No. 7261523 by the applicant of the present application.

[0047] The generation model DB143 manages a generation model as an LLM used when the data generation unit 114 performs data generation processing. The generation model holds a model in which learning has been performed to a certain extent and is managed so that re-learning can be performed by the re-learning processing unit 120.

[0048] The prompt DB144 manages prompt data for constructing a prompt used when the data generation unit 114 performs data generation processing. The prompt corresponds to an instruction or command used as the input of the generation model and is used appropriately according to the content to be output. For example, the format of the prompt assuming the output data may be defined in advance as the prompt data. The prompt data registered in the prompt DB144 may be configured to be directly input to the generation model, or may be used in combination with the RAG data obtained from the RAG DB145 described later.

[0049] The RAG DB145 manages RAG data used to generate a prompt input to the generation model. The type and method of the RAG data are not particularly limited, but may be registered according to the input and output by the generation model. The RAG data used according to a predetermined condition is extracted from the registered RAG data and input to the generation model in combination with the prompt data held in the prompt DB144.

[0050] The evaluation data DB 146 manages evaluation data that shows the evaluation results for the generated data generated by the data generation unit 114. The evaluation data is based on evaluations made by users who have reviewed the generated data. The evaluation data may include information entered according to the evaluation method, such as evaluation results evaluated on multiple levels or string information entered through free input. Furthermore, the evaluation data may also include information that can be used in the retraining process described later.

[0051] The inference result DB 147 stores the inference results from the inference processing unit 113. The inference result DB 147 may also store the data used as input during the inference process, associated with the estimation results.

[0052] The generated data DB 148 stores the generated data obtained as a result of the data generation process by the data generation unit 114. The generated data DB 148 may also store data used as input during data processing (including prompt data and RAG data) in association with the generation results.

[0053] The output data DB 149 holds output data that is constructed based on the processing results of the inference processing unit 113 and the data generation unit 114. The output data is configured to be output to the user, etc., on a UI screen or in a predetermined file format. The output data may be the data itself that is the processing result of the inference processing unit 113 and the data generation unit 114, or it may be data that has been formatted based on the report DB 151 described later.

[0054] The training data DB 150 holds training data created based on the generated data generated by the data generation unit 114 and evaluation data for said generated data. The structure of the training data may be defined, for example, according to the structure of the generation model.

[0055] The report database 151 holds various reports used, for example, when applying generated data generated by the data generation unit 114 to a predetermined format. The types and structure of the reports are not particularly limited and may be defined according to the nature of the business.

[0056] The usage log DB152 manages logs related to the use of inference and generative models. The logs may include historical information such as user information, usage date and time, usage frequency, versions of each model used, and input / output data. The logs may also include service login information and operation history.

[0057] Figure 3 is a block diagram showing an example of the functional configuration of the user terminal 200 according to this embodiment. The user terminal 200 is composed of a control unit 210, a storage unit 220, a communication unit 230, an operation unit 240, a display unit 250, and an external IF (Interface) 260.

[0058] The control unit 210 is responsible for controlling the operation of the user terminal 200. The control unit 210 is composed of, for example, a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit), and provides various functions by reading and executing various programs and data stored in the storage unit 220.

[0059] The memory unit 220 is a storage device for storing programs, data, and the like for executing various control processes and functions of the control unit 210. The memory unit 220 is composed of volatile / non-volatile storage devices such as RAM (Random Access Memory), ROM (Read Only Memory), HDD (Hard Disk Drive), and flash memory.

[0060] The communication unit 230 is a communication interface for communicating with external devices via a network NW. The communication unit 230 may be configured to support multiple communication standards depending on the configuration of the network NW.

[0061] The operation unit 240 is an interface for receiving operations from the user of the user terminal 200. The operation unit 240 may consist of a mouse, keyboard, etc. The display unit 250 is an interface for displaying various screens and consists of a display, etc. A touch panel display that integrates the operation unit 240 and the display unit 250 may be used. The external IF 260 is an interface for connecting to various devices and may be a connection interface to, for example, an imaging unit (not shown) for taking images or sensors for acquiring predetermined information.

[0062] [Data Flow] Figure 4 is a schematic diagram illustrating the data flow in the business support system 1 according to this embodiment. The flow shown here is implemented in the server device 100 and focuses on the parts relevant to this embodiment. Therefore, the data flow is not limited to this, and further data flows (for example, data input by the user) may be added as needed.

[0063] The server device 100 collects business data 410 from stored data and input from new users (step S401). In this embodiment, the business data 410 consists of standardized data 411 and non-standardized data 412. Standardized data 411 is data that is entered into a format that has already been systematized in business operations. Examples of standardized data 411 include checklists consisting of predetermined check items, administrative management information used in administrative operations, income information of support recipients, family trees, and activity logs. On the other hand, non-standardized data 412 is data that has not been systematized. Examples of non-standardized data 412 include medical questionnaire data entered in a free-form entry area, image data of support recipients, and medical data. The classification of standardized data 411 and non-standardized data 412 included in the business data 410 is not particularly limited and may be set arbitrarily. Therefore, the items, granularity, and data format (text, still images, moving images, audio, etc.) included in standardized data 411 and non-standardized data 412 may differ depending on the field of application and the content of the service. In this embodiment, the business data 410 that can be used as input for the inference model is designated as standardized data 411, and the remaining data that can be used as input for the generative model is designated as unstandardized data 412.

[0064] The server device 100 uses standardized data 411 from the business data 410 as input to an inference model and obtains inference result data 420 (step S402). The inference model used in the inference process and the standardized data to be input differ depending on the output result desired by the user. Therefore, the target data is selected from the standardized data 411 based on the user's instructions. The options for the output desired by the user may be defined in advance, and the data items of the inference model and standardized data 411 may be defined in correspondence with these options.

[0065] The inference result data 420 obtained as a result of the inference process is displayed via a UI screen (not shown) (step S403). The UI screen may be configured to be shown or hidden based on user instructions.

[0066] The server device 100 uses the prompt data group 470 to set the prompt 430 to be used in the generation model based on the output result desired by the user (step S404). The prompt 430 differs depending on the content of the generated data 440 as the output result desired by the user. The prompt data group 470 includes prompt data held in the prompt DB 144 and RAG data held in the RAG DB 145. Multiple prompt data may be defined in advance depending on the content of the generated data, and an appropriate prompt data may be selected from among them or combined with RAG data. In this embodiment, RAG technology is used to search for information for prompt generation from the RAG data held in the RAG DB 145. The searched information is then combined with the prompt data held in the prompt DB 144 and used as the prompt 430.

[0067] The server device 100 uses the standardized data 411 and non-standardized data 412, inference result data 420, and prompts 430 from the business data 410 as inputs to a generation model and acquires generated data 440 (step S405). The generation model used in the data generation process, and the input standardized data 411, non-standardized data 412, inference result data 420, and prompts 430 differ depending on the output result desired by the user. Therefore, based on user instructions, the data to be used as input data is selected. The options for the output desired by the user may be defined in advance, and the data items of the generation model and various data may be defined in correspondence with these options.

[0068] Furthermore, in the data generation process (step S405), higher-order inference may be performed in which multiple generative models cooperate and interact with each other, repeating the inference process to improve the inference accuracy. In this case, the patterns of the multiple generative models that can be combined may be predetermined, or they may differ depending on the input and output data of the generative models. Also, the content of the prompt 430 obtained in the prompt control process (step S404) may be switched depending on the generative models that can be combined.

[0069] The server device 100 outputs the generated data 440 obtained as a result of the data generation process to the user in a predetermined format (step S406). This output may be in a predetermined file format or on a UI screen (not shown). The generated data 440 may be output in the same format as output from the generation model, or it may be output after processing to format it.

[0070] The server device 100 performs evaluation processing on the output generated data 440 (step S407). The generated data evaluation processing receives evaluations of the output generated data 440 from the user, aggregates the results, and stores them in association with other data. The evaluation items and evaluation methods here are not particularly limited, but may be predefined multi-stage evaluations or evaluations by free input. Furthermore, the server device may accept instructions on whether or not to adopt the data as training data 450 in the subsequent training data generation processing. In addition, it may accept information for updating the RAG DB 145 used in the prompt control processing.

[0071] The server device 100 performs the process of generating training data 450 based on the evaluation results of the generated data (step S408). The configuration of the training data 450 may differ depending on the configuration of the generation model and the learning algorithm. Furthermore, a user such as an administrator may be configured to perform part of the generation of the training data 450. For example, a user may intervene manually in the annotation and selection of the generated data.

[0072] The server device 100 performs a retraining process on the generative model using the generated training data 450 (step S409). The retraining process may be performed when a certain amount of training data 450 has been accumulated, or it may be performed based on user instructions. The generative model obtained through the retraining process may be used to replace an existing generative model, or the generative models before and after retraining may be managed separately and configured to allow switching between their use. Furthermore, if multiple generative models are used in combination in the data generation process (step S405), the retraining process may be performed on each of the multiple generative models.

[0073] The server device 100 performs the RAG data generation process 460 based on the evaluation results of the generated data (step S410). The configuration of the RAG data 460 may differ depending on the generation model and the configuration of the generated data 440. Also, the configuration of the generated RAG data 460 may differ in anticipation of the retraining process of the generation model (step S409).

[0074] The server device 100 performs update processing on the RAG DB 145 using the generated RAG data 460 (step S411). This update processing is not limited to adding newly generated RAG data; it may also involve deleting or updating low-accuracy or old data. As a result, the group of prompt data 470 for generating the prompt 430 is updated.

[0075] (Generation Model) Figure 5 is a block diagram showing an overview of the input and output of the generation model 500 used in the data generation process according to this embodiment (step S405 in Figure 4). The generation model 500 is controlled so that its inputs differ depending on the type, content, and structure of the generated data 440 desired by the user. The correspondence between inputs and outputs may be predetermined. Furthermore, as described above, when multiple generation models work together to generate data, the inputs and outputs and processing order are defined for each of them.

[0076] The generative model 500 uses at least one of the following as input data: standardized data 411, unstandardized data 412, and inference result data 420. Furthermore, the LLM-based generative model 500 receives a prompt 430 as an instruction to output the desired generative data 440. The prompt 430 is composed of prompt data 471 and RAG data 472, which constitute the prompt data group 470. The prompt data 471 is managed in the prompt DB 144. The RAG data 472 is managed in the RAG DB 145.

[0077] The business data 410, which consists of standardized data 411 and non-standardized data 412, may be used as input in the form of the collected data itself, or it may be preprocessed for input into the generative model 500. Similarly, the inference result data 420 may be used as input in the form of the data inferred by the inference model, or it may be preprocessed for input into the generative model 500. Examples of input and output data for the generative model 500 include the following data.

[0078] (Standardized Data) - Checklists consisting of prescribed items - Administrative management information used in administrative operations - Information on the person being supported: - Family register information - Family structure (including graph data such as family trees) - Income information - Medical information - Insurance information - Pension information - Behavioral log data - SNS data (including SNS consultations from the consultation hotline "189") ...

[0079] (Non-standardized data) • Natural language – Questionnaire data (results of questionnaires conducted by visitors) – Survey records (input values ​​for survey items) • Still images – Images of bruises (images of injuries, etc., of the person receiving support) – Images of living environment (images of the person receiving support's living environment) • Videos, audio...

[0080] (Inference result data) ・Severity (priority of support for the person being supported) ・Prediction of abuse occurrence (prediction of the type of abuse occurring) ・Type of response (type of response to be taken for the abuse occurring) ・Prediction of response outcome (prediction of the outcome if a response is taken) ...

[0081] (Generated data) ・Support plan (list of actions to be taken for the person receiving support) ・Support effect (predicted value) ・Case records and legal documents (documents in prescribed format) ・Preparation documents for responses (documents for reporting on support) ・Data for knowledge accumulation (data for sharing know-how regarding support, etc.) ・Deductive explanations and inferences regarding the results of the inference model...

[0082] While the above examples illustrate standardized and non-standardized data according to this embodiment, some or all of the following information items may also be used. The choice of which information items to use may be arbitrarily adjusted depending on the service or case to which they are applied.

[0083] (Resident Basic Information Data) Resident basic information managed by the national and local governments, standardized by the Digital Agency of Japan: Resident Basic Register, Family Register, Family Register Supplementary Register, Individual Resident Tax, Corporate Resident Tax, Fixed Asset Tax, Light Vehicle Tax, Collection Management (Tax System), Delinquency Management (Tax System), Local Taxes (Common), School Register Compilation, School Attendance Assistance, Health Management, Child Allowance, Public Assistance, Disability Welfare, Long-Term Care Insurance, National Health Insurance, Medical Care for the Elderly, National Pension, Child Allowance, Children, Childcare Support, Application Management, Address Number Management for Non-Residents, Integrated Addressing within Organizations, Medical Claim Management (Public Assistance), Integrated Collection Management, Integrated Delinquency Management, Population Dynamics Survey, Specific Health Checkups, etc.

[0084] (Education-related data) ・Basic school information - Name of school attended (elementary, junior high, high school, special needs school, etc.) - Grade and class - Homeroom teacher / grade level head information - Learning status - Academic achievement test results (National Academic Achievement and Learning Status Survey, in-school tests) - Class attendance status (number of absences, tardiness, early departures) - Report card and internal evaluation - Evaluation of learning attitude - Participation in club activities - Use of after-school learning support, etc. ・Life and behavior - History of bullying consultation and guidance - Presence or absence of truancy and number of days attended - Records of school troubles (violence, problematic behavior) - History of career guidance and career counseling, etc. ・Health and health room related - History of visits to the health room (frequency of visits, reasons) - Results of health checkups (vision and hearing tests, internal medicine examinations, dental examinations, nutritional status) - Mental state (student's self-reported mental health assessment), etc.

[0085] (Welfare-related data) ・Family circumstances - Status of receipt of public assistance - Status of receipt of child allowance - Status of certification as a single-parent family - Housing stability (number of moves, history of homeless support), etc. ・Use of child guidance centers and welfare services - History of consultations and reports to child guidance centers - History of reports and responses to child abuse - History of use of foster care systems and child welfare institutions - Use of support for hikikomori (socially withdrawn individuals) and support for people in poverty - Use of child welfare facilities (nursery schools, children's centers, etc.) - Use of single-parent support services, etc.

[0086] (Health and medical data) - Regular health checkup data - Infant and toddler health checkup results (developmental assessment, height and weight, nutritional status) - Vaccination history (date of vaccination, type of vaccine) - School health checkup (height, weight, vision, hearing, dental, internal medicine, motor skills) - Medical institution visit history - Medical institution visit history (medical department, diagnosis, treatment status) - Mental health related (history of visits to psychiatry and psychosomatic medicine, medication status) - Allergy and chronic illness information (asthma, food allergies, atopic dermatitis, etc.), etc. - Others - Oral health checkup information such as the number of cavities - Motor skills test results (endurance, muscle strength, flexibility) - Nutrition and dietary habits (eating of school lunches, presence or absence of picky eating), etc.

[0087] (Related to child support programs) - History of participation in programs supporting social withdrawal / school refusal and developmental support - History of admission to child welfare facilities and group homes - History of using school counseling - History of using programs supporting further education and employment - Status of use of self-reliance support programs for people in poverty, etc.

[0088] [Processing Flow] Figure 6 is a flowchart showing the overall processing flow related to the business according to this embodiment. This processing flow is realized, for example, by the control unit 110 of the server device 100 reading and executing programs and various data stored in the storage unit 140. For the sake of simplicity, the processing entity is described here as the server device 100.

[0089] Before this processing flow begins, various types of information are registered and configured to be accessible in each database of the server device 100. Furthermore, the server device 100 is configured to allow the addition, editing, and deletion of various types of information managed in each database.

[0090] In step S601, the server device 100 receives instructions from the user regarding the data to be output as generated data. These instructions may be given by presenting options on a UI screen (not shown) and accepting the user's selection from those options. For example, the selection may be made from the candidates for generated data 440 shown in Figure 5.

[0091] In step S602, the server device 100 collects the necessary business data as input data from various databases based on the instructions received in step S601.

[0092] In step S603, the server device 100 extracts the target standardized data from the business data collected in step S602. The standardized data extracted here is identified based on the instructions received in step S601. Furthermore, the server device 100 determines the inference model to be used for inference processing based on the instructions received in step S601. Then, the server device 100 obtains inference result data by inputting the standardized data into the determined inference model. This enables inference processing using the standardized data.

[0093] In step S604, the server device 100 determines whether or not to display the inference result data obtained in step S603. Whether or not to display the data can be determined by receiving a display instruction in step S601, or by switching according to the type of generated data 440 instructed in step S601. If the inference result data is to be displayed (step S604: YES), the server device 100 proceeds to step S605. On the other hand, if the inference result data is not to be displayed (step S604: NO), the server device 100 proceeds to step S607.

[0094] In step S605, the server device 100 generates display data for the inference result data obtained in step S603. The display data may be, for example, a UI screen (not shown) that can be displayed on a browser. The display data may also be generated in a predetermined format.

[0095] In step S606, the server device 100 displays the display data generated in step S605. For example, the server device 100 displays the display data by providing it to the user terminal 200.

[0096] In step S607, the server device 100 preprocesses the inference result data obtained in step S603 for use as input data to the generation model. If the inference result data obtained in step S603 is to be input directly to the generation model, the preprocessing may be omitted. Furthermore, the content of the preprocessing is not particularly limited and may differ depending on the type of generation data specified in step S601.

[0097] In step S608, the server device 100 determines the prompt to be input to the generation model according to the type of generation data 440 instructed in step S601. Multiple prompts may be predetermined and selected or combined according to the type of generation data 440 to be output. In this embodiment, the server device 100 performs a search operation on the RAG data held in the RAG DB 145 according to the generation data 440 to be output. Furthermore, the server device 100 identifies the prompt data to be used from the prompt DB 144 according to the generation data to be output. The server device 100 then combines these prompt data with the RAG data to determine the prompt to be input to the generation model.

[0098] In step S609, the server device 100 extracts the target standardized and non-standardized data from the business data collected in step S602. The standardized and non-standardized data extracted here are identified based on the instructions received in step S601. Furthermore, the server device 100 determines a generation model based on the instructions received in step S601. Then, the server device 100 obtains the desired generation data as output by inputting the inference result data, standardized data, non-standardized data, and prompts into the determined generation model.

[0099] In step S610, the server device 100 outputs the generated data obtained in step S609. The output here may be the generated data obtained in step S609 as is, or it may be converted to an output file format or manner before output. Then, this processing flow ends.

[0100] (Retraining Process) Figure 7 is a flowchart of the retraining process for the generative model according to this embodiment. This process flow is realized, for example, by the control unit 110 of the server device 100 reading and executing programs and various data stored in the storage unit 140. For the sake of simplicity, the server device 100 is described as the processing entity here.

[0101] This processing flow may be performed each time generated data is obtained in the process shown in Figure 6, or it may be performed all at once after a certain number of generated data have been generated and are to be evaluated. Furthermore, this processing flow shows an example in which the retraining of the generative model and the updating of the prompt data set used to generate prompts as input to the generative model are performed in a series of steps. However, these do not necessarily have to be performed simultaneously and may be performed as separate processes.

[0102] In step S701, the server device 100 generates an evaluation screen for inputting evaluations of the generated data to be evaluated. The evaluation screen may be, for example, a UI screen that can be displayed in a browser (not shown). The evaluation items here may differ depending on the type of generated data to be evaluated. The evaluation items may be multiple-choice or free-form.

[0103] In step S702, the server device 100 displays the evaluation screen generated in step S701. For example, the server device 100 displays the evaluation screen by providing it to the user terminal 200.

[0104] In step S703, the server device 100 accepts the evaluation of the generated data to be evaluated via the evaluation screen displayed in step S702.

[0105] In step S704, the server device 100 generates training data based on the evaluation content and target generated data received in step S703. The method of generating training data is not particularly limited, but for example, the generated data may be classified according to the evaluation, or generated data that will not be used as training data may be identified.

[0106] In step S705, the server device 100 records the training data generated in step S704.

[0107] In step S706, the server device 100 generates RAG data based on the evaluation content and target generated data received in step S703.

[0108] In step S707, the server device 100 updates the prompt data set using the RAG data generated in step S706. More specifically, it updates the data held in the RAG DB 145.

[0109] In step S708, the server device 100 determines whether the accumulated training data exceeds a certain amount. The training data subject to this determination is training data that has not been used for retraining. The threshold here is not particularly limited and may be either the amount of training data or the number of training data. If the accumulated training data exceeds a certain amount (step S708: YES), the server device 100 proceeds to step S709. On the other hand, if the accumulated training data does not exceed a certain amount (step S708: NO), this processing flow ends.

[0110] In step S709, the server device 100 performs retraining of the generative model using the accumulated training data. The method of retraining here is not particularly limited and may differ depending on the configuration and algorithm of the generative model. For example, continuous learning is a known method for improving the accuracy of LLM. Generally, the training process is computationally intensive, so it may be configured so that it is performed by another device instead of the server device 100. In addition, the training data used for retraining may be discarded so as not to be reused, or it may be managed as used data.

[0111] In step S710, the server device 100 records the generative model obtained as a result of the retraining performed in step S709. Here, only the retrained generative model may be retained, or multiple generative models may be managed depending on the degree of training. Then, this processing flow is terminated.

[0112] [Examples] Two examples of applications of the method according to this embodiment are shown. In child welfare and maternal and child health, the output results (generated data) according to this embodiment can be used from the perspective of preventing the occurrence and recurrence of problems in assistance guidelines. In addition, in educational settings, the output results (generated data) according to this embodiment can be used from the perspective of preventing the occurrence and recurrence of problems in curriculum guidelines.

[0113] (Recurrence) In a designated application area, for "recurrence" of a problem or event, the following can be performed, for example, based on the configuration shown in Figure 4: - Prediction (calculate and display the recurrence probability value) - Background of the recurrence (explanation using LLM or RAG) - Measures to reduce the recurrence probability (explanation using LLM or RAG)

[0114] (Requires Support) In a designated application area, for "requires support" regarding problems or events, the following may be performed based on the configuration shown in Figure 4: - Prediction (calculate and display the probability value of requiring support) - Background of the suspected need for support (explanation using LLM or RAG) - Measures to reduce the probability of requiring support (explanation using LLM or RAG)

[0115] The above is merely an example, and the content executed in the inference process (step S402) and data generation process (step S405) may be adjusted as appropriate depending on the domain and the problem being considered.

[0116] As described above, this embodiment makes it possible to improve the work efficiency of supporters providing community support services. In particular, it becomes possible to easily generate various data necessary for community support services. Furthermore, it becomes possible to accumulate knowledge in specific work areas and to pass on various kinds of knowledge. In addition, by using the functions of this embodiment, it becomes possible to improve the quality of on-site judgment by supporters. Moreover, by considering the characteristics of the inference model and the generative model and combining them, it becomes possible to refine the prompts that serve as input data for obtaining more appropriate inference results in the generative model.

[0117] <Other Embodiments> In addition, the present invention can also be realized by supplying a program or application for realizing the functions of one or more embodiments described above to a system or device using a network or storage medium, and having one or more processors in the computer of that system or device read and execute the program.

[0118] Alternatively, one or more functions may be implemented using a circuit (for example, an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array)).

[0119] Although various embodiments have been described above with reference to the drawings, it goes without saying that this disclosure is not limited to these examples. It will be clear to those skilled in the art that various modifications, alterations, substitutions, additions, deletions, and equivalents can occur within the scope of the claims, and these will naturally fall within the technical scope of this disclosure. Furthermore, the components of the various embodiments described above can be combined arbitrarily without departing from the spirit of the invention.

[0120] Thus, the present invention is not limited to the embodiments described above. It is also intended and within the scope of protection to be provided for the combination of each configuration of the embodiments, as well as for modifications and applications by those skilled in the art based on the description in the specification and well-known technology.

[0121] As described above, the following matters are disclosed in this specification:

[0122] (Technology 1) A business support system (e.g., 100) comprising: an inference unit (e.g., 113) that inputs business data related to support (e.g., 410) into an inference model and infers support information (e.g., 420) corresponding to the business data; and a generation unit (e.g., 114) that inputs at least one of the support information and the business data into a generation model (e.g., 500) based on a large-scale language model and generates generated data (e.g., 440) corresponding to the support. This configuration makes it possible to improve the work efficiency of supporters who provide community support services. In particular, it makes it possible to easily generate various data necessary for community support services. It also makes it possible to accumulate knowledge in specific business areas and to pass on various kinds of knowledge. Furthermore, by using the functions according to this embodiment, it is possible to promote the improvement of the quality of on-site judgment by supporters. Furthermore, by considering the characteristics of the inference model and the generation model and combining them, it is possible to refine the prompts that become input data for obtaining more appropriate inference results in the generation model.

[0123] (Technology 2) The business support system according to Technology 1, wherein the business data comprises standardized data (e.g., 411) with predetermined items and non-standardized data (e.g., 412), the inference model is input to the standardized data, and the generation model is input to at least one of the standardized data and the non-standardized data. With this configuration, for example, by using a combination of data having different characteristics from standardized data and non-standardized data in the business data as input, it becomes possible to generate data for improving business efficiency.

[0124] (Technology 3) The business support system described in Technology 2, wherein the standardized data includes at least one of the following: checklists, administrative management information, and revenue information. This configuration makes it possible to use various types of information as standardized data, for example.

[0125] (Technology 4) The business support system described in Technology 2 or 3, wherein the non-standardized data includes at least one of the following: interview data, survey records, and images of the person being supported. This configuration makes it possible to use various types of information as non-standardized data, for example.

[0126] (Technology 5) A business support system according to any one of Technology 1 to 4, further comprising an output unit (for example, 116, 123) that outputs the generated data generated by the generation unit. With this configuration, for example, it becomes possible to output generated data by a generation model to the user.

[0127] (Technical 6) The business support system described in Technical 5, wherein the output unit further outputs support information inferred by the inference unit. This configuration makes it possible to output support information as an inference result from the inference model, in addition to the generated data from the generative model.

[0128] (Technology 7) A business support system according to any one of Technology 1 to Technology 6, further comprising: a receiving unit (e.g., 121) for receiving instructions on the type of generated data to be generated by the generation unit; and an instruction unit (e.g., 115) for switching prompts to control the output from the generation model based on the instructions. With this configuration, for example, it becomes possible to control the switching of prompts for the generation model according to the output content desired by the user, based on user instructions.

[0129] (Technical 8) The business support system according to Technical 7, wherein the instruction unit generates prompts to be input to the generation model using information identified by RAG (Retrievable-Augmented Generation) based on the instruction. With this configuration, for example, it becomes possible to improve the accuracy of the generated data generated by the generation model and output more appropriate user-desired generated data.

[0130] (Technical 9) The business support system according to Technical 8, further comprising: a second generation unit (e.g., 119) that generates update information for a database used in the RAG in accordance with the evaluation of the generated data; and an update unit (e.g., 121) that updates the database when the instruction unit generates a prompt using the update information. With this configuration, for example, the database can be updated based on an evaluation of the output result as generated data, and the quality of subsequent generated data can be further improved.

[0131] (Technology 10) A business support system according to any one of Technologies 7 to 9, wherein the type of generated data can be specified as any of the following: support plan, predicted value of support effect, case record document, legal document, response preparation material, or data for knowledge accumulation. With this configuration, for example, it becomes possible to obtain generated data for various types of support as output.

[0132] (Technology 11) A business support system according to any one of Technology 1 to Technology 10, further comprising: an evaluation unit (e.g., 117) that receives an evaluation of the generated data; and a learning data generation unit (e.g., 118) that uses the generated data and the evaluation of the generated data to generate learning data for retraining the generation model. With this configuration, for example, it becomes possible to receive feedback on the generated data and generate learning data for retraining the generation model.

[0133] (Technical 12) The business support system according to Technical 11, further comprising a learning unit (e.g., 120) that performs a learning process of the generation model using the learning data generated by the learning data generation unit. With this configuration, for example, the accuracy of the generation model that generates support data can be improved by performing retraining using newly generated learning data based on feedback to the generation data.

[0134] (Technology 13) A business support system described in any of Technologies 1 to 12, wherein the business data related to the support includes personal information relating to one of the following: child support, abuse response, child injury / accident response, maternal and child health, domestic violence (DV) response, sexual violence response, delinquency / crime response, developmental disability response, elderly welfare, or disability welfare. This configuration makes it possible to generate support-related data by including personal information in business data related to various types of support.

[0135] (Technical 14) A control method for a business support system comprising: an inference step (e.g., S402) in which business data related to support (e.g., 410) is input into an inference model to infer support information (e.g., 420) corresponding to the business data; and a generation step (e.g., S405) in which at least one of the support information and the business data is input into a generation model based on a large-scale language model to generate generated data corresponding to the support. This configuration makes it possible to improve the work efficiency of supporters who provide community support services. In particular, it makes it possible to easily generate various data necessary for community support services. It also makes it possible to accumulate knowledge in specific business areas and to pass on various kinds of knowledge. Furthermore, by using the functions according to this embodiment, it is possible to promote the improvement of the quality of on-site judgment by supporters. Furthermore, by considering the characteristics of the inference model and the generation model and combining them, it is possible to refine the prompts that become input data for obtaining more appropriate inference results in the generation model.

[0136] (Technical 15) A program for causing a computer (e.g., 100) to execute: an inference step (e.g., S402) in which business data related to support (e.g., 410) is input into an inference model to infer support information (e.g., 420) corresponding to the business data; and a generation step (e.g., S405) in which at least one of the support information and the business data is input into a generation model based on a large-scale language model to generate generated data corresponding to the support. This configuration makes it possible to improve the work efficiency of supporters who provide community support services. In particular, it makes it possible to easily generate various data necessary for community support services. It also makes it possible to accumulate knowledge in specific business areas and to pass on various kinds of knowledge. Furthermore, by using the functions according to this embodiment, it is possible to promote the improvement of the quality of on-site judgment by supporters. Furthermore, by considering the characteristics of the inference model and the generation model and combining them, it is possible to refine the prompts that become input data for obtaining more appropriate inference results in the generation model.

[0137] Although various embodiments have been described above with reference to the drawings, it goes without saying that the present invention is not limited to these examples. It is clear to those skilled in the art that various modifications or alterations can be conceived within the scope of the claims, and these will naturally also fall within the technical scope of the present invention. Furthermore, the components of the above embodiments may be combined in any way without departing from the spirit of the invention.

[0138] This application is based on a Japanese patent application (JP 2024-230930) filed on December 26, 2024, the contents of which are incorporated by reference within this application.

[0139] The present invention is useful, for example, as a device, system, or method for improving operational efficiency, information sharing, and the quality of decision-making by support workers at support sites in organizations and institutions that provide community support services targeting a variety of individuals.

[0140] 1...Business support system 100...Server device 110...Control unit 111...Data collection unit 112...Data management unit 113...Inference processing unit 114...Data generation unit 115...Prompto control unit 116...Output data generation unit 117...Generated data evaluation unit 118...Training data generation unit 119...RAG data generation unit 120...Retraining processing unit 121...RAG data update unit 122...Model management unit 123...Display control unit 124...Communication control unit 130...Communication unit 140...Storage unit 141...Business data DB 142...Trained model DB 143...Generated model DB 144...Prompto DB 145...RAG DB 146...Evaluation data DB 147...Inference result DB 148...Generated data DB 149...Output data DB 150...Training data DB 151...Document DB 152...Usage Log DB 200...User Terminal 210...Control Unit 220...Storage Unit 230...Communication Unit 240...Operation Unit 250...Display Unit 260...External IF 300...Integration System NW...Network

Claims

1. A business support system comprising: an inference unit that inputs business data related to support into an inference model and infers support information corresponding to said business data; and a generation unit that inputs at least one of the support information and said business data into a generation model based on a large-scale language model and generates generated data corresponding to said support.

2. The business support system according to claim 1, wherein the business data comprises standardized data with predetermined items and non-standardized data, the standardized data is input to the inference model, and at least one of the standardized data and the non-standardized data is input to the generation model.

3. The business support system according to claim 2, wherein the standardized data includes at least one of a checklist, administrative management information, and revenue information.

4. The business support system according to claim 2, wherein the non-standardized data includes at least one of the following: medical interview data, survey records, and images of the person being supported.

5. The business support system according to claim 1, further comprising an output unit that outputs the generated data generated by the generation unit.

6. The business support system according to claim 5, wherein the output unit further outputs support information inferred by the inference unit.

7. The business support system according to claim 1, further comprising: a receiving unit for receiving instructions on the type of generated data to be generated by the generation unit; and an instruction unit for switching prompts for controlling the output from the generation model based on the instructions.

8. The business support system according to claim 7, wherein the instruction unit generates prompts to be input to the generation model using information identified by RAG (Retrievable Augmented Generation) based on the instruction.

9. The business support system according to claim 8, further comprising: a second generation unit that generates update information for a database used in the RAG in accordance with the evaluation of the generated data; and an update unit that updates the database when the instruction unit generates a prompt using the update information.

10. The business support system according to claim 7, wherein the type of generated data can be specified to be any of the following: support plan, predicted value of support effect, case record document, legal document, response preparation material, or data for knowledge accumulation.

11. The business support system according to claim 1, further comprising: an evaluation unit that receives an evaluation of the generated data; and a learning data generation unit that generates learning data for retraining the generation model using the generated data and the evaluation of the generated data.

12. The business support system according to claim 11, further comprising a learning unit that performs a learning process of the generated model using the learning data generated by the learning data generation unit.

13. The business support system according to claim 1, wherein the business data related to the support includes personal information relating to any of the following: child support, abuse response, child injury, accident response, maternal and child health, domestic violence (DV) response, sexual violence response, delinquency and crime response, developmental disability response, elderly welfare, and disability welfare.

14. A control method for a business support system, comprising: an inference step of inputting business data related to support into an inference model and inferring support information corresponding to said business data; and a generation step of inputting at least one of the support information and said business data into a generation model based on a large-scale language model and generating generated data corresponding to said support.

15. A program for causing a computer to perform the following steps: an inference step of inputting business data related to support into an inference model and inferring support information corresponding to said business data; and a generation step of inputting at least one of the support information and said business data into a generation model based on a large-scale language model and generating generated data corresponding to said support.

Images