Data processing method, data processing system, and data processing program

Abstract

Improve the convenience of data processing. [Solution] A method for processing target data in which values ​​are input into multiple data items, the method vectorizing information identifying the multiple data items and attribute information of the values ​​of the multiple data items, preparing design information data including information on the relationships between data items, information identifying the vectorized multiple data items, and attribute information of the values ​​of the vectorized multiple data items, inputting information identifying the type of data processing and text data representing the processing content into a large-scale language model to output key words for the processing, inputting the output words into a vectorization model to generate key vectors, identifying the data items and processing content of the target data from the key vectors by referring to the design information data, and executing processing of the content identified for the identified data items or outputting information for executing the processing.

Description

[Technical Field]

[0001] The present invention relates to a technique for executing processing on data in which values ​​are input into each of a plurality of predetermined data items. [Background technology]

[0002] Databases containing customer information and the like are used in various places, such as companies and government offices. Relational databases are widely used as databases, where data items of the information to be stored are set in one or more columns, and multiple tables, each with records entered row-wise, are associated with each other. In relational databases, for example, a user can use a Structured Query Language (SQL) statement to specify the table and column to be searched, set search conditions, and execute a search to obtain information on the desired record (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 5-120100 [Patent Document 2] Japanese Patent Publication No. 2022-91986 [Non-patent literature]

[0004] [Non-Patent Document 1] "Getting an idea of ​​Embedding (Vectorization)!", [online], Classmethod Co., Ltd., Retrieved February 20, 2025, Internet<URL: https: / / zenn.dev / peishim / articles / c696ff85a539bd> [Non-patent document 2] "IBM Granite", [online], IBM Japan, Ltd., searched February 20, 2025, Internet<URL:https: / / www.ibm.com / jp-ja / topics / large-language-models> Summary of the Invention [Problem to be solved by the invention]

[0005] To extract desired records using the above database search, the user must accurately enter search conditions in the SQL statement, including the table name, the names of the data items set in each column, and the attributes of the values ​​of each data item (numeric, text, etc.). This poses a problem: users cannot extract desired records from a database without knowing the database structure. While database searches have been used as an example here, similar problems arise when inputting, updating, deleting, and other operations in a database, as the names of the tables and columns to be processed must be accurately specified. Similar problems also arise when processing data sets that do not take the so-called database format.

[0006] The problem to be solved by the present invention is to improve convenience when processing data in which values ​​are input for each of a plurality of predetermined items. [Means for solving the problem]

[0007] One aspect of the present invention, which has been made to solve the above problem, is a method used to execute processing on target data in which values ​​are input into each of a plurality of predetermined data items, the method comprising: A large-scale language model is prepared that receives text data as input and outputs key words for data processing, constructed using predetermined machine learning methods. A vectorization model is prepared that vectorizes and outputs input words, the vectorization model being constructed using predetermined machine learning techniques. vectorizing information identifying the plurality of data items and attribute information of values ​​of the plurality of data items using the vectorization model; preparing design information including information on the association between the plurality of data items, information identifying the plurality of vectorized data items, and attribute information of values ​​of the plurality of vectorized data items; prompting a user to input information specifying the type of processing to be performed on the target data and text data indicating the content of the processing; inputting text data representing the content of the processing into the large-scale language model, thereby outputting key words for the processing from the large-scale language model; Generate key vectors by inputting words output from the large-scale language model into the vectorization model; By referring to the design information data, a data item and a processing content of the target data are identified from the key vector; Execute the specified processing for the specified data item, or output information for executing the processing. It is characterized by:

[0008] Large-scale language models (LLMs) are generative AI specialized for natural language processing (NLP), which are constructed with a neural network whose parameters are adjusted by machine learning the vectors generated through tokenization, which divides various text data into the smallest units such as words and punctuation marks, and vectorization, which quantifies the tokenized data. By having such large-scale language models learn in advance the attributes of the words to be output, it is possible to derive the purpose of use (in the present invention) from the text data input as natural language. ,for exampleIt can output words corresponding to input words (database search). Also, a vectorization model (also called an embedding model) is a trained model constructed by deep learning using a large number of word-vector pairs. By using such a vectorization model, it is possible to output vectors having components that reflect the meaning of the input word.

[0009] The data processing method according to the present invention is typically used to process data (records) constituting a database, but can also be used to process other types of data. In the data processing method according to the present invention, a large-scale language model and a vectorized model are prepared in advance. Furthermore, information identifying data items and attribute information of the record values ​​input for each data item are vectorized, and design information data is created together with information on the relevance of the data items. The value attribute information here refers to, for example, numeric values, free text, and options (selectable using a pull-down menu, etc.). Note that the value of the target data may also be null. Furthermore, information on the relevance of data items refers to, for example, when multiple data items constitute a single table, information identifying a table common to those data items, and information for updating the value of another data item based on the value input to that data item.

[0010] A user inputs information identifying the type of data processing and text data representing the content of the processing. Examples of the type of data processing include data search, inputting, updating, and deleting values ​​of data items. The information on the type of data processing is not limited to information directly input by the user, but may also include information indirectly input (e.g., identifying the type of processing based on input through a specific screen or based on text data being sent to a specific email address). The input text data is input to a large-scale language model, which outputs key words for the data processing. Examples of key words for the data processing include words related to data items such as industry, address, and number of employees, as well as words related to search criteria such as "within one year," "within one month," and "Kyoto Prefecture." Furthermore, a vectorization model is used to vectorize the words output from the large-scale language model to generate key vectors. Then, by referencing design information data, the data item to be processed and the processing content are identified from the key vectors. This method of vectorizing the meanings of words and sentences and identifying those with similar meanings is called semantic search or vector search. By performing such a search, even if the words output from the large-scale language model do not completely match the information specifying the data items or the attribute information of the values ​​of each data item, it is possible to identify the target and content of the data processing based on the similarity in meaning between the two (approximation of vector components). For example, the target and content of the data processing can be identified based on having a vector that is closest to the processing key vector (smallest difference in components, shortest distance between vectors, smallest angle between vectors, etc.).

[0011] Once the data item to be processed and the content of the processing for that data item have been determined, the processing of the specified content is executed for the identified data item, or information for executing that processing is output.

[0012] Another aspect of the present invention is a system used to execute processing on data in which values ​​are input for each of a plurality of predetermined items, the system comprising: a large-scale language model storage unit that stores a large-scale language model that is constructed by predetermined machine learning and receives input text data and outputs key words for data processing; a vectorization model storage unit that stores a vectorization model that is constructed by predetermined machine learning and that vectorizes and outputs input words; a design information data storage unit in which design information data is stored, the design information data including information on the association between the plurality of data items, information for identifying the plurality of data items vectorized by the vectorization model, and attribute information on the values ​​of the plurality of data items vectorized by the vectorization model; a processing key acquisition unit that allows a user to input information specifying the type of processing for the target data and text data representing the content of the processing, and that inputs the text data representing the content of the processing into the large-scale language model, thereby outputting a word that is a key to the processing from the large-scale language model; a key vector generation unit that generates a key vector by inputting words output from the large-scale language model into the vectorization model; a processing content specification unit that specifies a data item and processing content of the target data from the key vector by referring to the design information data; a processing execution unit that executes the specified processing for the specified data item, or outputs information for executing the processing; The present invention is characterized by comprising:

[0013] Yet another aspect of the present invention is a program used to execute processing on data in which values ​​are input for each of a plurality of predetermined items, the program comprising: a large-scale language model storage unit that stores a large-scale language model that is constructed by predetermined machine learning and receives input text data and outputs key words for data processing; a vectorization model storage unit that stores a vectorization model that is constructed by predetermined machine learning and that vectorizes and outputs input words; a design information data storage unit in which design information data is stored, the design information data including information on the association between the plurality of data items, information for identifying the plurality of data items vectorized by the vectorization model, and attribute information on the values ​​of the plurality of data items vectorized by the vectorization model; In a system having a computer, a processing key acquisition unit that allows a user to input information specifying the type of processing for the target data and text data representing the content of the processing, and that inputs the text data representing the content of the processing into the large-scale language model, thereby outputting a word that is a key to the processing from the large-scale language model; a key vector generation unit that generates a key vector by inputting words output from the large-scale language model into the vectorization model; a processing content specification unit that specifies a data item and processing content of the target data from the key vector by referring to the design information data; a processing execution unit that executes the specified processing for the specified data item, or outputs information for executing the processing; The present invention is characterized in that it functions as a [Effects of the Invention]

[0014] By using the data processing method, data processing system, or data processing program according to the present invention, it is possible to improve the convenience of performing processing on data in which values ​​have been entered for each of a plurality of specified items. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is a diagram showing the configuration of a main part of an embodiment of a data processing system according to the present invention; [Figure 2] FIG. 2 is a diagram for explaining the configuration of design information data related to data processing in this embodiment. [Figure 3] FIG. 2 is a diagram for explaining the configuration of design information data related to knowledge processing in this embodiment. [Figure 4] 10 shows an example of design information related to a column processing group (information group) and design information related to connection information in this embodiment. [Figure 5] 10 shows examples of design information related to basic information and design information related to item information in this embodiment. [Figure 6] 10 shows an example of design information related to option information in this embodiment. [Figure 7] 10 shows an example of design information related to sample value information in this embodiment. [Figure 8] 10 shows an example of design information related to trigger information in this embodiment. [Figure 9] 10 shows an example of design information related to detailed information updated by a trigger in this embodiment. [Figure 10] 10 shows examples of design information related to knowledge information and design information related to knowledge information-related documents in this embodiment. [Figure 11] 10 shows an example of processing for a database in this embodiment. [Figure 12] 10 shows an example of processing for data input to a screen or the like in this embodiment. [Figure 13] 10 shows an example of knowledge processing in this embodiment. [Figure 14] 1 is an image flow illustrating a flow of executing data processing by incorporating the data processing device of this embodiment into an existing business system. [Figure 15] 10 shows an example of request information (structure) in this embodiment. [Figure 16] 10 shows an example of decomposition result information (structure) in this embodiment. [Figure 17] 10 shows an example of return information (structure) in this embodiment. [Figure 18] 10 is an example of a screen of an input form for a customer visit report in this embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, an embodiment of a data processing method, a data processing system, and a data processing program according to the present invention will be described with reference to the drawings. In this embodiment, an example of processing information recorded in a business system (in-house database) provided in a company will be described.

[0017] 1 shows the configuration of the main parts of a data processing system 1 according to this embodiment. The data processing system 1 includes a business system 10 and a data processing device 20 that are installed within a private network of a company or the like.

[0018] The business system 10 includes a sales support system 11, a sales management system 12, a payroll calculation system 13, and a personnel management system 14. The sales support system 11 has a customer management database (DB) 11D, the sales management system 12 has a financial management database (DB) 12D, the payroll calculation system 13 has a payroll management database (DB) 13D, and the personnel management system 14 has a personnel management database (DB) 14D. Here, an example has been described in which each system has one database, but one system may have multiple databases.

[0019] The customer management database 11D, financial management database 12D, payroll management database 13D, and human resources management database 14D are all relational databases that associate multiple tables in which data items are arranged in the column direction and records are arranged in the row direction.

[0020] The data processing device 20 includes a storage unit 21. The storage unit 21 includes a design information data storage unit 211, a tuning data storage unit 212, a large-scale language model (LLM) storage unit 213, a vectorized model storage unit 214, and an input form storage unit 215.

[0021] The design information data storage unit 211 is one of the characteristic elements of this embodiment, and stores information relating to the structures of various databases that are the targets of processing by the data processing device 20 as design information data.

[0022] Design information data is broadly composed of information related to data processing (including database processing and other natural language processing) and knowledge information (data processing performed based on accumulated data).

[0023] The structure of design information data related to data processing is shown in Figure 2. Design information related to data processing includes (1) a processing group for specifying the content of the processing to be executed, (2) join information that indicates the join relationships between tables in a database and the join relationships between multiple input screens, (3) basic information about the database tables and screens, (4) item information about the columns in the tables and the items provided in the screens, (5) option information about the options presented in the database columns and the items in the screens, (6) value sample information about the values ​​to be entered in the database columns and the items in the screens, and (11) and (12) trigger information.

[0024] The structure of design information data related to knowledge processing is shown in Figure 3. The design information related to knowledge processing includes (7) knowledge information that accumulates documents such as daily reports, and (8) knowledge information-related documents (e.g., templates) for generating new information such as instructions and advice based on the knowledge information.

[0025] Figure 4 shows examples of (1) design information related to a processing group (information group) and (2) design information related to integration information. (1) A processing group (information group) does not directly handle database table or screen information, but refers to a label as a collection of information. Design information related to a processing group (information group) includes items related to system identification, function group, group code, object identification, name, meaning, vector value, and access rights, each of which has a value (character, numeric, vector array). (2) Integration information refers to connector information that treats several pieces of basic information as a single unit, such as which combination of tables in the database the processing group will use to execute processing, or in the case of a screen, which screen or object-based collection will be used to process. Design information related to integration information includes items related to system identification, function group, group code, integration order information, information ID, name, identifier, integration type, and integration source information conditions, each of which has a value (character, numeric).

[0026] Figure 5 shows examples of design information related to (3) basic information and (4) item information. (3) Basic information refers to a table in a database, the label for the entire screen in a screen, and the label for the entire process in a process. Design information related to basic information includes items related to system identification, function group, object identification, information ID, name, meaning, vector value, and access rights, each of which has a value (character, number, vector array). (4) Item information refers to a column in a database, an item object in a screen, and the label for the process details in a process. Design information related to item information includes items related to system identification, function group, object identification, information ID, item ID, option reference category, option reference item ID, name, meaning, vector value, attribute, number of digits, default value, item type, required item, auto-set target, and access rights, each of which has a value (character, number, vector array).

[0027] Figure 6 is an example of design information related to (5) option information. (5) option information refers to information for each selected value when the attribute is an option (such as in the case of a selection list or radio buttons). Design information related to option information includes items related to system identification, function group, object identification, information ID, item ID, option, name, meaning, vector value, attribute, and number of digits, and a value (character, numeric value, vector array) is set for each.

[0028] Figure 7 is an example of design information related to (6) sample value information. (6) Sample value information refers to information for each sample value that represents the characteristics of an item when the attribute of the item is other than a choice. Design information related to sample value information includes items related to system identification, function group, object identification, information ID, item ID, attribute, three types of sample values ​​(character, number, date), name, meaning, and vector value, each of which is set with a value (character, number, vector array). Note that choice information and sample information can also be used to register and use public information that is common within a system or within a function group, in addition to information specific to the item.

[0029] Figure 8 is an example of design information related to (11) trigger information. (11) trigger information refers to monitoring item information and the specification information for updating other items in conjunction with each other when an entry or change is made to that item. Design information related to (11) trigger information includes items related to system identification, function group, trigger management number, target information ID, target item ID, activation condition, activation specification value (character, number, date), action, and access right, and a value (character, number, date) is set for each.

[0030] Figure 9 is an example of design information related to (12) Trigger Detail Information. (12) Trigger Detail Information refers to the reference source information and setting destination information after the trigger is activated. Design information related to (12) Trigger Detail Information includes items related to system identification, function group, trigger management number, category identifying read source / update destination, branch number (row), target information ID, target item ID, read / update conditions, prompt, specified value (character, number, date), and operation, each of which has a value (character, number, date) set.

[0031] Figure 10 shows examples of (7) design information related to knowledge information and (8) design information related to knowledge information related documents. (7) Design information related to knowledge information has items related to system identification, function group, information unique number, related type, record document, vector value, keys 1 to 3, and name, each of which has a value (character, number, vector array). (8) Knowledge information related documents has items related to system identification, function group, information unique number, related document branch number, and related sentence, each of which has a value (character, number).

[0032] The tuning data storage unit 212 stores data for tuning a large-scale language model. Tuning of a large-scale language model is performed using learning methods such as prompt tuning and fine tuning. This results in a large-scale language model that has learned information specific to the industry (for example, industry terminology, its meaning, abbreviations, etc.). For example, in the case of the manufacturing industry, the tuning data used may include information such as the industry term "yield rate," which means the probability that no defective products will be produced, that the yield rate is expressed as a number between 0 and 1 (or a percentage between 0 and 100), and that the closer the yield rate is to 1, the higher the proportion of good products, and the closer it is to 0, the higher the proportion of defective products.

[0033] The storage unit 21 also includes a large-scale language model storage unit 213 that stores large-scale language models, which are trained models constructed by machine learning, such as receiving input text data and outputting key words for data processing, and a vectorization model storage unit 214 that stores vectorization models constructed by machine learning, which vectorize input words and output them.

[0034] The large-scale language models (LLMs) used in this embodiment are generative AI specialized for natural language processing (NLP), which are constructed with neural networks whose parameters are adjusted by machine learning the vectors generated through tokenization, which divides various text data into the smallest units such as words and punctuation marks, and vectorization, which quantifies the tokenized data. In this embodiment, such large-scale language models are trained in advance to learn the attributes of words to be output, thereby performing tokenization, cleansing, etc. on text data input in natural language, etc., and outputting words related to data processing.

[0035] In this embodiment, a large-scale language model is used that is stored in the large-scale language model storage unit 213 and that can be used only within a company (provided in a private network with no external connection). It is difficult for a single company to own a large-scale language model, and it is common for a large-scale language model provided on the cloud, for example, to be shared by an unspecified number of people (or a specified number of people). Although the present invention can use a large-scale language model that is open to the public (public), if a large-scale language model is trained with information related to confidential matters within a company, such as customer information, there is a risk that the information may be leaked to the outside.

[0036] For this reason, it is preferable to use a (private) large-scale language model that can only be used in a closed environment, as in this embodiment. In this embodiment, it is sufficient to extract words related to the processing of information stored in a database (search, data input, data update, etc.) from input natural language; it is not necessary to use a model that learns all kinds of words through machine learning. For example, IBM's Granite (Non-Patent Document 2) can be used as such a large-scale language model. Granite is a large-scale language model designed to be lightweight and with a low computational load, and can be owned and used by individual companies, even if they are not large companies. When using such a private large-scale language model, it is recommended that the above tuning be performed using the design information data stored in the design information data storage unit 211. This allows the large-scale language model to output processing keys that can more accurately identify tables, columns, etc.

[0037] The vectorization model used in this embodiment is a trained model constructed by deep learning using a large number of word-vector pairs, and constitutes part of a large-scale language model. In this embodiment, by using such a vectorization model, it is possible to output vectors having components that reflect the meaning of input words. The vector array (vector value) included in the above design information can be generated by using such a vectorization model. The vectorization model is also called an embedding model (e.g., Non-Patent Document 1), and is used to perform semantic search (also called vector search). The vectorization model may be made using suitable software that can be used over a network such as the Internet, or may be software owned by a company.

[0038] The storage unit 21 also includes an input form storage unit 215 that stores input forms for inputting various types of data. For example, input forms related to the customer management database 11D include a form for registering a new customer, a form for reporting the results of a visit to a customer, and a form for inputting the results of a transaction with a customer. These input forms include data items corresponding to columns and display fields for inputting values ​​of the columns. Furthermore, for specific data items, input of a value triggers processing, such as changing the value of another column (stored as trigger information; see FIGS. 8 and 9). Furthermore, when text data, such as natural language data, is input into a specific input / display field of each input form, processing is performed using a large-scale language model and / or a vectorization model (this is also stored as trigger information).

[0039] The data processing device 20 includes, as functional blocks, a design information data creation unit 31, a tuning execution unit 32, a processing key acquisition unit 33, a key vector generation unit 34, a processing content identification unit 35, a processing command generation unit 36, and a processing execution unit 37. The data processing device 20 is configured, for example, by a general-purpose personal computer or workstation, and these functional blocks are realized by executing a pre-installed data processing program on a processor. Also connected to the data processing device 20 are an input unit 51 including a keyboard, a mouse, etc., and a display unit 52 including a liquid crystal display, etc.

[0040] 11 to 13 show an overview of data processing performed in the data processing device of this embodiment. Fig. 11 relates to processing for a database, Fig. 12 relates to processing for data input to a screen or the like, and Fig. 13 relates to knowledge processing.

[0041] As shown in Figure 11, there are four types of database processing: data search (SELECT), data update (UPDATE), data creation (INSERT), and data deletion (DELETE). When executing these data processing operations, text data consisting of natural language is input, and processing using a large-scale language model (LLM) (LLM language processing 1) and vector database (DB) processing (Vector DB processing 1) are performed to obtain the respective outputs.

[0042] As shown in Figure 12, there are three types of processing for data entered on a screen, etc.: processing to generate a summary from a group of items, processing to generate a group of items from a sentence, and processing to generate a group of conditions from a sentence. In processing to generate a summary from a group of items, information on the structure of the group of items and their values ​​is input, and a summary document is obtained through vector database (DB) processing (vector DB processing 1) and processing using a large-scale language model (LLM) (LLM language processing 2). In processing to generate a group of items from a sentence and processing to generate a group of conditions from a sentence, natural language and information on the structure of the group of items to be set are input, and processing using a large-scale language model (LLM) (LLM language processing 1) and vector database (DB) processing (vector DB processing 1) is performed to obtain an array of the group of items and setting values.

[0043] As shown in Figure 13, there are three types of knowledge processing: accumulation (vector DB), summarization, and proposal. In accumulation processing, the structure information of the items and values ​​to be summarized, related types, and related documents are input, and data is accumulated through processing using a large-scale language model (LLM) (LLM-based language processing 1), vector database (DB) processing (vector DB processing 1), large-scale language model (LLM) processing (LLM-based language processing 2), and vector database (DB) processing (vector DB processing 2). (No output is generated during accumulation processing.) In summarization processing, the structure information of the items and values ​​to be summarized is input, and data is accumulated through processing using a large-scale language model (LLM-based language processing 1), vector database (DB) processing (vector DB processing 1), and large-scale language model (LLM) processing (LLM-based language processing 2), resulting in a summary document as output. In the proposed process, the structure information of the items and values ​​to be summarized and the return related type are input, and processing is performed using a large-scale language model (LLM) (language processing 1 using LLM), vector database (DB) processing (vector DB processing 1), processing using a large-scale language model (LLM) (language processing 2 using LLM), and vector database (DB) processing (vector DB processing 2), and related documents are obtained as output.

[0044] Next, the operation of the data processing system of this embodiment (one embodiment of a data processing method according to the present invention) will be described.

[0045] In the data processing system 1 of this embodiment, , processing For each of the databases of the business system 10 to be managed, the design information described with reference to FIGS. 2 to 10 is used to execute the processing described with reference to FIGS.

[0046] The data processing device 20 of this embodiment can be incorporated into an existing business system 10. Fig. 14 is an image flow illustrating the flow of executing the above-mentioned processing in a data processing system 1 in which the data processing device 20 is incorporated into the existing business system 10. In Fig. 14, the part surrounded by a dashed line indicates the processing and operations in the business system 10, and the part surrounded by a dashed line indicates the processing and operations by the data processing device 20. The business system 10 and the data processing device 20 are configured to send and receive information via an application programming interface (API).

[0047] 2 to 10 is created in advance by the design information data creation unit 31, and the created design information is stored as a vector database (DB) in the design information data storage unit 211 of the data processing device 20. Furthermore, the tuning execution unit 32 performs learning such as prompt tuning and fine tuning of the large-scale language model using tuning data stored in the tuning data storage unit 212, and the large-scale language model after learning is stored in the large-scale language model storage unit 213.

[0048] Generally, a database is associated with a program for processing data in that database, and a user can execute various data processing operations on the database through that program. In conventional data processing, a user processes data by inputting commands into the program in a language that the program can process.

[0049] On the other hand, when the data processing device 20 of this embodiment is incorporated, when a user inputs text data (request information) such as natural language, the data is sent to the data processing device 20 via the API. In the data processing device 20, the processing key acquisition unit 33 inputs the text data to a large-scale language model as request information (structure), and obtains the decomposition result information (structure) that is decomposed and output by the large-scale language model as a processing key (language processing 1 using LLM).

[0050] As shown in Figure 15, the request information (structure) may include API basic information, target identification, requested action, requested natural language, additional prompt, information ID, and input information array. Of these, the API basic information and target identification can also be identified, for example, from the screen on which the user inputs or the email address to which the user sent text data. Furthermore, as shown in Figure 16, the decomposition result information (structure) may include API basic information, target identification, information ID, input information array, and decomposition information array.

[0051] Next, the key vector generation unit 34 vectorizes the processing key using the vectorization model. Then, the processing content identification unit 35 uses the vector to perform a semantic search of the specification information, values, and options in the design information database (vector DB) to identify the content of the data processing requested by the text data entered by the user (vector DB processing 1).

[0052] Here, if the processing content is processing on a database (FIG. 11), the processing command generation unit 36 ​​generates a command statement (a command statement written in a programming language associated with the target database) describing the content of the identified data processing as return information (structure) and sends it to the business system 10 via the API. As shown in FIG. 17, the return information (structure) may include result status information and a result information array.

[0053] If the processing content is a process for generating a summary from a group of items (FIG. 12) or knowledge processing (FIG. 13), the processing execution unit 37 again executes processing using a large-scale language model (such as creating a summary sentence) (LLM-based language processing 2). Also, if the processing content is a summary process or a proposal process (FIG. 13), the processing execution unit 37 again executes a search of a vector DB using a vectorized model (vector DB processing 2). After these processes, the processing command generation unit 36 ​​generates a command statement (a command statement written in a programming language associated with the target database) describing the content of the identified data processing as return information (structure), and sends it to the business system 10 via the API.

[0054] After the above processing, the data processing device 20 outputs information in a format that is previously associated with the content of each data processing.

[0055] The information output from the data processing device 20 is input to the business system 10 via an API. In the business system 10, a program associated with each database processes the information. This allows data processing based on request information input by the user using text data such as natural language.

[0056] In conventional data processing systems, in order for a user to perform desired data processing (such as extracting records), the user had to accurately enter the name of the table, the names of the data items set in each column, the value attributes of each data item (numeric, text, etc.), and the processing content in the SQL statement.This created a problem in that the user could not extract the desired records from the database unless they knew the database structure or the programming language.

[0057] In contrast, in this embodiment, even if the wording used as the processing key does not match the names set in the table information, column information, etc. that make up the database, the table and column to be searched can be identified based on the semantic similarity determined by comparing vector values, and the processing content for the value of that column can be identified. This type of processing is called a semantic search or vector search. By performing a semantic search in this way, desired information can be extracted even if the user does not have knowledge of the structure of the database to be processed. Furthermore, the user only needs to enter the desired data processing content in text data such as natural language, and no knowledge of programming languages ​​is required.

[0058] Furthermore, in business systems, it is common for various employees to add records and change or update the values ​​of data items. However, not all employees know the input format for data items. As a result, values ​​with the same meaning can be mixed in different formats. Examples of different formats include a mix of Arabic numerals and Chinese numerals, or the presence or absence of units (entering units causes data to be recognized as text data and excluded from statistical processing of numeric data). In conventional data processing systems, when values ​​in various formats are mixed, the input values ​​are reflected as is, creating the problem that, for example, when performing statistical processing on numeric data, values ​​entered in Chinese numerals or values ​​entered with units are not processed statistically.

[0059] In contrast, in the data processing system 1 of this embodiment, data processing is performed in a format defined in the design information from text data input by the user in various formats. It is possible to accurately perform statistical processing on numerical data and extract data.

[0060] The data processing system 1 of this embodiment can be configured to implement the following functions.

[0061] Support functions: For example, based on some input data such as natural language requirements, the remaining items are automatically entered. Individual input values ​​are collected and summarized to create a summary document (for example, the body of a daily report). The content is analyzed from the natural language contained in emails, etc., and the processing flow to be performed is then invoked.

[0062] Automatic creation function: Daily reports and other reports can be automatically created by attaching audio, memo photos, etc. to an email while on the go and sending it to a specified email address with the required information, such as API basic information, target identification, and requested actions, pre-assigned.

[0063] Response function: Searches for past inquiries that have a similar meaning to the inquiry received via the inquiry form or the email address set as the contact point, and suggests a response that embeds the answer in a pre-prepared template.

[0064] Data extraction (search) function: Automatically sets appropriate search conditions according to the requested text. Accumulation function: Data on success and failure patterns is accumulated and used as knowledge. For example, summary reports and proposals are made from accumulated data that has a similar meaning to data such as input daily reports. Application: Read text from business card images and automatically extract customers and contact persons.

[0065] Below, some specific examples of data processing in the data processing system 1 of this embodiment will be described.

[0066] <Database operation: SELECT> When a user inputs text data "List of customers not recently visited" from a screen previously associated with the customer management database 11D, the large-scale language model outputs the words "customer," "list," and "not recently visited," and a semantic search of the design information data identifies that the word "customer" represents the target table, the word "list" represents the target column group, and the word "not recently visited" represents the search condition. Then, the processing command generator 36 generates a command statement for executing this data search and sends it to the business system 10 as return information.

[0067] <Database operation: UPDATE> When a user inputs text data such as "Enable direct mail delivery to prospective customers within a 5-kilometer radius who are currently in negotiations," from a screen previously associated with the customer management database 11D, the large-scale language model outputs the phrases "currently in negotiations; customer," "direct mail delivery; enabled," and "prospective customer within a 5-kilometer radius." Then, a semantic search of the design information data identifies that the phrase "currently in negotiations; customer" refers to the target table. The system also searches for column information corresponding to the phrase "direct mail delivery; enabled" and searches for specifications related to "enabled," which refers to processing for that column. The system then identifies the option information with the closest meaning. Furthermore, even if a specific numeric value, such as "within a 5-kilometer radius," a semantic search of the design information identifies the option information with the closest meaning. As a result, for customers who meet the search criteria of "prospective customer within a 5-kilometer radius," the option for the data item "direct mail delivery" is updated to the value "enabled."

[0068] <Processing for database: INSERT & Processing for items on screen etc: Condition group from sentences> When the user accesses the screen for adding a "customer visit report" to the customer management database 11D, the processing key acquisition unit 33 reads out the customer visit report input form stored in the storage unit 21 and displays it on the screen of the display unit 52.

[0069] An example of the screen for the customer visit report input form is shown in Figure 18. This screen has columns for customer code, customer name, status report, ABC rank, and accuracy. Of these, the user enters text data in the customer name and status report columns (the columns indicated by solid lines in Figure 18). When the user enters the customer name, the code for the customer name (customer code) is read from the database and displayed in the customer code column.

[0070] Natural language text data is input into the status report field. When the user inputs natural language text data into the status report field, the processing key acquisition unit 33 inputs the input natural language into the large-scale language model. The large-scale language model outputs, from the input natural language, terms related to data processing (processing keys) such as "order amount," "high price," "promising," "competitor," "slightly unfavorable," and "situation."

[0071] The key vector generation unit 34 inputs the text output from the large-scale language model into a vectorized model to generate a key vector. The processing content identification unit 35 references the design information data and identifies the content of data processing corresponding to the generated vector value. In this example, trigger information is set as the design information, which generates values ​​for the priority and accuracy data items when a value for the status report data item is input. The processing content identification unit 35 references this design information and attribute information for the values ​​input for the priority and accuracy, and determines input values ​​for the priority and accuracy data items from the output word group (in the example of FIG. 18, the priority value is "A" and the accuracy value is "50%)).

[0072] The processing command generation unit 36 ​​generates a command statement for adding the processing content (a set of a data item and its value) identified by the processing content identification unit 35 as a record to the customer management database 11D, and transmits it as return information (structure) to the business system 10. The business system 10 executes the processing command statement generated by the processing command generation unit 36, and adds the record to the database specified by the user (here, the customer management database 11D).

[0073] In conventional data processing systems, information entered by a user in natural language, such as the situation report in the above example, is simply entered as a database value, and meaningful statistical processing is not possible. In contrast, the data processing system 1 of this embodiment determines values ​​for predetermined data items (here, priority and accuracy) from the natural language entered by the user, making it possible to perform meaningful statistical processing using these values. Furthermore, when entering natural language, even if the content is the same, the length and expression of the text vary depending on the inputter, making it difficult to correctly understand the situation without carefully reading the input natural language sentence. In contrast, the data processing system 1 of this embodiment uses a large-scale language model to evaluate natural language written by various users using a common standard and register the evaluated values ​​in the database.

[0074] As a technology for creating SQL statements from natural language, Patent Document 2 describes a technology that pre-matches question information to be processed with table contents of a target table to obtain a pre-matching result, extracts word segments that match the question most closely, determines target attribute values ​​semantically associated with the extracted word segments, and generates an SQL query statement using the target attribute values, etc. Patent Document 2 states that by using this technology, it is possible to generate an SQL statement even if the question information submitted by the user does not perfectly match the column names or attribute values ​​of the table.

[0075] Patent Document 2 only presents the above technology conceptually, and ofAlthough no specific application examples are given, this technology creates SQL statements in response to interactive human requests and outputs the resulting data by executing them, and search methods other than SQL statements cannot be used. Therefore, it is only possible to perform database searches that conform to the input format of the SQL statements. Furthermore, it is not easy to create SQL statements by accurately outputting the elements necessary for data processing from a large-scale language model. Furthermore, it is not clear what the term "attribute" means in Patent Document 2. In addition, it is unclear how this technology should be applied when there are multiple tables in a database (such as in the case of a so-called relational database).

[0076] On the other hand, while the data processing system 1 of this embodiment can use SQL statements, it essentially outputs elements necessary for a database search using a semantic search, and can be used for search methods other than database searches using SQL statements. In addition, in this embodiment, as described with reference to FIGS. 2 to 10, attributes of each item are defined in the database design information data, thereby enabling accurate data processing in line with the user's intentions. Furthermore, the data processing system 1 of this embodiment uses large-scale language models and vectorized models not only for database searches, but also for registering records in the database, updating and deleting values ​​of data items, and for processing inputs to items on a screen, etc. (FIG. 12) and knowledge processing (FIG. 13), thereby enabling a variety of data processing.

[0077] As described above, the data processing method, data processing system, and data processing program of the present embodiment combine artificial intelligence (AI) with a business system (including a large number of existing business systems). By using the data processing method, data processing system, and data processing program of the present embodiment, existing business systems can be improved into AI-compatible, flexible, and efficient business systems with minor modifications, without making design changes to the existing business systems or redeveloping the systems. Furthermore, the data processing method, data processing system, and data processing program of the present embodiment can accumulate more design information as it is used, thereby improving the AI's understanding of the design information. As a result, in the future, it is expected that AI will be able to generate various data input / output screens and reports tailored to user needs, eliminating the need for the programs responsible for such processing.

[0078] The above embodiment is merely an example and can be modified as appropriate in accordance with the spirit of the present invention.

[0079] In the above embodiment, the case where the user specifies the database to be processed and inputs the details of the data processing from an entry (a specific screen, a specific email address) corresponding to the content of the data processing has been described. However, a common entry may be used for executing all databases and all types of data processing. In this case, the processing content specification unit 35 refers to the design information data of all databases and specifies the database to be processed and the type and content of the data processing in that database based on the key vector generated by the key vector generation unit 34 from text data, such as natural language, entered by the user. However, by providing separate entry points, such as screens, through which the user inputs instructions for at least each database, the load associated with semantic search can be reduced and more accurate data processing can be performed.

[0080] There may be cases where it is not possible to extract all of the elements necessary for a command statement to execute data processing from the text data entered by the user. In such cases, the processing content specification unit 35 may notify the user by displaying the missing elements on the screen of the display unit 52, or by outputting an error message.

[0081] In the above embodiment, an example has been described in which all of the components necessary for searching for information in a database and inputting data into the database are provided within the data processing device 20. However, some of these components may be provided in another device (including another device within a private network, a cloud server provided on an open network with access restricted by encryption or a password, etc.). For example, in the above embodiment, the large-scale language model storage unit 213 and the vectorization model storage unit 214 are provided within the data processing device 20. However, a large-scale language model or a vectorization model provided in another device (same as above) connected to the data processing device 20 via a network may also be used. Furthermore, when a public large-scale language model or a vectorization model is used, it may be provided in another device (such as a cloud computer) connected via an open network (such as the Internet). Furthermore, a functional block related to database search and a functional block related to data input may each be provided in a different device (same as above), and these devices may be connected to the internal network of the data processing system 1.

[0082] In the above embodiment, an example of data processing for a customer management database 11D provided in a company's business system 10 has been described, but a similar configuration can also be adopted when processing information in various types of databases provided in public institutions such as government agencies. The specific examples of data processing described in the above embodiment are only a portion of what can be executed in the data processing system 1 of this embodiment, and can be modified as appropriate depending on the contents of the database and the purpose of use. Furthermore, in the above embodiment, a configuration has been described in which the business system 10 and the data processing device 20 are provided within a single private network, but they may also be connected via an open network (such as the Internet).

[0083] In the above embodiment, the case where access to all data in the database is permitted has been described, but it is also possible to configure the database so that only those with access rights can access a portion of the data (for example, data related to management, financial data, or personnel data). For example, when using the data processing system 1, a user is required to log in using a user ID and password, and the accessible range is determined according to the authenticated user ID.

[0084] In the above embodiment, an example of processing data in a database has been described, but a similar configuration can also be used when processing data other than a database. For example, the process of adding a customer visit report described in the above embodiment can also be used for purposes such as submitting a customer visit report in writing (i.e., for purposes that do not involve adding the report as a database record). Furthermore, in the above embodiment, values ​​for the data items priority and accuracy are generated from natural language input by the user, but conversely, a configuration is also possible in which the user is prompted to input priority and accuracy, and natural language is generated from those values ​​using a large-scale language model.

[0085] Furthermore, it is also possible to combine multiple pieces of trigger information in the above embodiment to execute a process similar to the flowchart. For example, when the number of records with priority A in the customer management database 11D exceeds 10, these records are extracted and the records are output as a list, and various processes can be combined and executed.

[0086] [Aspect] It will be apparent to those skilled in the art that the above-described exemplary embodiments are examples of the following aspects.

[0087] (Section 1) One aspect of the present invention is a method used to execute processing on target data in which values ​​are input into each of a plurality of predetermined data items, the method comprising: A large-scale language model is prepared that receives text data as input and outputs key words for data processing, constructed using predetermined machine learning methods. A vectorization model is prepared that vectorizes and outputs input words, the vectorization model being constructed using predetermined machine learning techniques. vectorizing information identifying the plurality of data items and attribute information of values ​​of the plurality of data items using the vectorization model; preparing design information data including information on the association between the plurality of data items, information for identifying the plurality of vectorized data items, and attribute information on the values ​​of the plurality of vectorized data items; prompting a user to input information specifying the type of processing to be performed on the target data and text data indicating the content of the processing; inputting text data representing the content of the processing into the large-scale language model, thereby outputting key words for the processing from the large-scale language model; Generate key vectors by inputting words output from the large-scale language model into the vectorization model; By referring to the design information data, a data item and a processing content of the target data are identified from the key vector; Execute the specified processing for the specified data item, or output information for executing the processing. It is characterized by:

[0088] (Section 7) Another aspect of the present invention is a system used to execute processing on data in which values ​​are input for each of a plurality of predetermined items, the system comprising: a large-scale language model storage unit that stores a large-scale language model that is constructed by predetermined machine learning and receives input text data and outputs key words for data processing; a vectorization model storage unit that stores a vectorization model that is constructed by predetermined machine learning and that vectorizes and outputs input words; a design information data storage unit in which design information data is stored, the design information data including information on the association between the plurality of data items, information for identifying the plurality of data items vectorized by the vectorization model, and attribute information on the values ​​of the plurality of data items vectorized by the vectorization model; a processing key acquisition unit that allows a user to input information specifying the type of processing for the target data and text data representing the content of the processing, and that inputs the text data representing the content of the processing into the large-scale language model, thereby outputting a word that is a key to the processing from the large-scale language model; a key vector generation unit that generates a key vector by inputting words output from the large-scale language model into the vectorization model; a processing content specification unit that specifies a data item and processing content of the target data from the key vector by referring to the design information data; a processing execution unit that executes the specified processing for the specified data item, or outputs information for executing the processing; The present invention is characterized by comprising:

[0089] (Section 9) Yet another aspect of the present invention is a program used to execute processing on data in which values ​​are input for each of a plurality of predetermined items, the program comprising: a large-scale language model storage unit that stores a large-scale language model that is constructed by predetermined machine learning and receives input text data and outputs key words for data processing; a vectorization model storage unit that stores a vectorization model that is constructed by predetermined machine learning and that vectorizes and outputs input words; a design information data storage unit in which design information data is stored, the design information data including information on the association between the plurality of data items, information for identifying the plurality of data items vectorized by the vectorization model, and attribute information on the values ​​of the plurality of data items vectorized by the vectorization model; In a system having a computer, a processing key acquisition unit that allows a user to input information specifying the type of processing for the target data and text data representing the content of the processing, and that inputs the text data representing the content of the processing into the large-scale language model, thereby outputting a word that is a key to the processing from the large-scale language model; a key vector generation unit that generates a key vector by inputting words output from the large-scale language model into the vectorization model; a processing content specification unit that specifies a data item and processing content of the target data from the key vector by referring to the design information data; a processing execution unit that executes the specified processing for the specified data item, or outputs information for executing the processing; The present invention is characterized in that it functions as a

[0090] The data processing method according to paragraph 1, the data processing system according to paragraph 7, and the data processing program according to paragraph 9 are typically used to process data (records) constituting a database, but may also be used to process other data. In the data processing method according to the present invention, a large-scale language model and a vectorized model are prepared in advance. Furthermore, information identifying data items and attribute information of the values ​​of records input for each data item are vectorized, and design information data is created together with information on the relevance of the data items. The value attribute information here refers to, for example, numeric values, free text, and options (selectable in a pull-down menu, etc.). Note that the value of the target data may also be null. Furthermore, information on the relevance of data items refers to, for example, when multiple data items constitute a single table, information identifying a table common to those data items, and information for updating the value of another data item based on the value input to that data item.

[0091] A user inputs information identifying the type of data processing and text data representing the content of the processing. Examples of the type of data processing include data search, inputting, updating, and deleting values ​​of data items. The information on the type of data processing is not limited to information directly input by the user, but may also include information indirectly input (e.g., identifying the type of processing based on input through a specific screen or based on text data being sent to a specific email address). The input text data is input to a large-scale language model, which outputs key words for the data processing. Examples of key words for the data processing include words related to data items such as industry, address, and number of employees, as well as words related to search criteria such as "within one year," "within one month," and "Kyoto Prefecture." Furthermore, a vectorization model is used to vectorize the words output from the large-scale language model to generate key vectors. Then, by referencing design information data, the data item to be processed and the processing content are identified from the key vectors. This method of vectorizing the meanings of words and sentences and identifying those with similar meanings is called semantic search or vector search. By performing such a search, even if the words output from the large-scale language model do not completely match the information specifying the data items or the attribute information of the values ​​of each data item, it is possible to identify the target and content of the data processing based on the similarity in meaning between the two (approximation of vector components). For example, the target and content of the data processing can be identified based on having a vector that is closest to the processing key vector (smallest difference in components, shortest distance between vectors, smallest angle between vectors, etc.).

[0092] Once the data item to be processed and the details of the processing for that data item are determined, the specified processing is executed for the specified data item, or information for executing that processing is output. This improves the convenience of executing processing on data in which values ​​are entered for each of multiple specified items.

[0093] (Section 2) The data processing method according to paragraph 2 is the data processing method according to paragraph 1, The large-scale language model has learned predetermined words and their meanings, which are stored in a library prepared in advance.

[0094] The data processing method according to paragraph 2 makes it possible to correctly process abbreviations, slang terms, and the like commonly used in the industry using a large-scale language model.

[0095] (Section 3) The data processing method according to paragraph 3 is a data processing method according to paragraph 1 or 2, The large-scale language model is located within a private network along with the database.

[0096] The data processing method according to paragraph 3 can prevent confidential information held within a company or other entity from being leaked to the outside.

[0097] (Section 4) The data processing method according to paragraph 4 is the data processing method according to paragraph 3, The large-scale language model is further trained on the design information data.

[0098] In the data processing method according to paragraph 4, a large-scale language model that has learned the structure and attributes of a database is used, thereby making it possible to output a processing key that can more accurately identify tables and columns.

[0099] (Section 5) The data processing method according to paragraph 5 is a data processing method according to any one of paragraphs 1 to 4, further comprising: The design information includes trigger information for executing data processing for a predetermined second data item when an input is made for a predetermined first data item.

[0100] In the data processing method according to paragraph 5, the information input by the user into the first data item can be automatically reflected in the second data item.

[0101] (Section 6) The data processing method according to paragraph 6 is the data processing method according to paragraph 5, further comprising: the design information data includes conversion information for converting a value input for the first data item into an input value for the second data item in accordance with a predetermined criterion; When information is input to the first data item, the input value is converted based on the conversion information and the converted value is input as the value of the second data item.

[0102] In the data processing method according to paragraph 6, the user inputs natural language and the words output from the large-scale language model can be used to automatically determine rankings of importance, accuracy values, etc.

[0103] (Section 8) The data processing system according to paragraph 8 is the data processing system according to paragraph 7, The data is stored in a database that is provided independently of the data processing system and operates according to a pre-prepared program, The processing execution unit outputs a command statement for executing the processing of the specified content in a language executable by the program. It is characterized by:

[0104] The data processing system according to paragraph 8 may be incorporated into an existing database. [Explanation of symbols]

[0105] 1...Data processing system 10. Business Systems 11...Sales support system 11D...Customer management database 12...Sales management system 13...Payroll system 14...Human Resources Management System 20...Data processing device 21...Storage section 211...design information data storage unit 212... Tuning data storage unit 213...Large-scale language model memory unit 214...Vectorized model storage unit 215...Input form memory section 31...Design Information Data Creation Department 32...Tuning execution unit 33...Processing key acquisition unit 34...Key vector generation unit 35...Processing content specification unit 36...Processing instruction generation unit 37...Processing execution unit 51...Input section 52...Display section

Claims

1. A data processing method in which a computer executes processing on target data in which a value has been entered into at least one data item among a plurality of data items in a database, comprising: A large-scale language model is prepared that receives text data as input and outputs key words for data processing, constructed using predetermined machine learning methods. A vectorization model is prepared that vectorizes and outputs input words, the vectorization model being constructed using predetermined machine learning techniques. vectorizing information identifying each of the plurality of data items and attribute information of values ​​of each of the plurality of data items using the vectorization model; storing design information data including information for identifying each of the plurality of vectorized data items and attribute information of the values ​​of each of the plurality of vectorized data items in a design information data storage unit; having a user input text data representing the content of processing to be performed on the target data; inputting text data representing the content of the processing into the large-scale language model, thereby outputting key words for the processing from the large-scale language model; Generate key vectors by inputting words output from the large-scale language model into the vectorization model; comparing the key vector with a vector of information identifying each of the plurality of data items and a vector of attribute information of the values ​​of each of the plurality of data items, and identifying the data item and processing content of the target data based on the similarity with the key vector; Execute the specified content processing for the specified data item. A data processing method comprising:

2. 2. The data processing method according to claim 1, wherein the large-scale language model is a model that has learned predetermined words stored in a library and the meanings of the words.

3. The data processing method of claim 1 , wherein the large-scale language model is provided within a private network.

4. The data processing method according to claim 3 , wherein the large-scale language model is further trained on the design information data.

5. moreover, 2. The data processing method according to claim 1, wherein the design information data includes trigger information for executing data processing for a predetermined second data item when input is made for a predetermined first data item.

6. moreover, the design information data includes conversion information for converting a value input for a predetermined first data item into an input value for a predetermined second data item in accordance with a predetermined criterion; 2. The data processing method according to claim 1, wherein when a value is input to the first data item, the input value converted based on the conversion information is input as the value of the second data item.

7. A data processing system used to execute processing on target data in which a value has been input into at least one data item among a plurality of data items in a database, comprising: a large-scale language model storage unit that stores a large-scale language model that is constructed by predetermined machine learning and receives input text data and outputs key words for data processing; a vectorization model storage unit that stores a vectorization model that is constructed by predetermined machine learning and that vectorizes and outputs input words; a design information data storage unit in which design information data including information identifying each of the plurality of data items vectorized using the vectorization model and attribute information of values ​​of each of the plurality of data items vectorized using the vectorization model is stored; a processing key acquisition unit that allows a user to input text data representing the content of processing for the target data, and that inputs the text data representing the content of processing into the large-scale language model, thereby outputting a word that is a key for the processing from the large-scale language model; a key vector generation unit that generates a key vector by inputting words output from the large-scale language model into the vectorization model; a processing content identification unit that compares the key vector with a vector of information identifying each of the plurality of data items and a vector of attribute information of the values ​​of each of the plurality of data items, and identifies the data item and processing content of the target data based on the similarity with the key vector; a processing execution unit that executes the processing of the specified content with respect to the specified data item; A data processing system comprising:

8. The processing execution unit outputs a command statement written in a predetermined programming language that executes the processing of the specified content.

8. The data processing system of claim 7.

9. A data processing program for causing a computer to execute processing on target data in which a value has been input into at least one data item among a plurality of data items in a database, comprising: a large-scale language model storage unit that stores a large-scale language model that is constructed by predetermined machine learning and receives input text data and outputs key words for data processing; a vectorization model storage unit that stores a vectorization model that is constructed by predetermined machine learning and that vectorizes and outputs input words; a design information data storage unit in which design information data including information identifying each of the plurality of data items vectorized using the vectorization model and attribute information of values ​​of each of the plurality of data items vectorized using the vectorization model is stored; A computer in a system having a processing key acquisition unit that allows a user to input text data representing the content of processing for the target data, and that inputs the text data representing the content of processing into the large-scale language model, thereby outputting a word that is a key for the processing from the large-scale language model; a key vector generation unit that generates a key vector by inputting words output from the large-scale language model into the vectorization model; a processing content identification unit that compares the key vector with a vector of information identifying each of the plurality of data items and a vector of attribute information of the values ​​of each of the plurality of data items, and identifies the data item and processing content of the target data based on the similarity with the key vector; a processing execution unit that executes the processing of the specified content for the specified data item; A data processing program characterized by causing the program to function as:

Images