Information processing methods, information processing systems, and programs

Abstract

To provide a method for accurately understanding the contents of each document according to the needs of each business partner. [Solution] The information processing method includes a computer acquiring at least one form data, converting the form data into a text dataset containing multiple text data, acquiring customer identification information indicating a customer associated with the text dataset, and applying the text dataset to a trained model to generate a structured dataset associated with the customer, wherein the trained model is based on the relationship between the text data associated with the form data and the customer master data associated with the customer identification information.

Description

Technical Field

[0001] The present invention relates to an information processing method, an information processing system, and a program for assisting in generating a structured dataset.

Background Art

[0002] In recent years, from the perspective of promoting DX (Digital Transformation), the digitization of documents (transaction-related documents) such as order forms and estimates received from business partners has been progressing. At this time, as a method for converting a document into text data, OCR (Optical Character Recognition) processing is generally used as disclosed in Patent Document 1.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] On the other hand, the formats of documents vary for each business partner. Also, when digitizing handwritten documents, in conventional OCR processing, the information described in the document is diverse, and it is difficult to accurately understand the content of the document. Furthermore, in the case of handwritten documents, it is difficult to accurately determine the information described in the document.

[0005] In view of such problems, one object of the present invention is to provide a method for accurately understanding the content of a document according to each business partner. Another object of the present invention is to easily digitize handwritten documents. Another object of the present invention is to easily generate structured data from FAX documents and handwritten documents.

Means for Solving the Problems

[0006] According to one embodiment of the present invention, a computer provides an information processing method that includes acquiring at least one form data, converting the form data into a text dataset containing a plurality of text data, acquiring user identification information associated with the text data that can identify a user, acquiring user business partner identification information associated with the text dataset and the user identification information that indicates the user's business partners, and applying the text dataset to a trained model to generate a structured dataset associated with the user's business partners, wherein the trained model provides an information processing method based on the relationship between the text data associated with the previously acquired form data and the user business partner master data associated with the user business partner identification information.

[0007] In the above-described information processing method, generating the structured dataset may include determining the type of report corresponding to the text dataset and selecting a structured template for each of the multiple text data according to the type of report.

[0008] In the above information processing method, the first text data among the plurality of text data may be converted into a second text data characterized by the user business partner master data based on the trained model.

[0009] In the above information processing method, when the first text data contains multiple meanings, one meaning may be identified from the multiple meanings based on supporting information associated with the first text data and the trained model, which helps to identify one meaning from the multiple meanings.

[0010] In the above information processing method, the support information may be placed near the first text data.

[0011] In the above information processing method, the support information may be location information of the text data.

[0012] In the above-described information processing method, the report data may consist of image data or audio data.

[0013] In the above-described information processing method, the image data may include at least one of the following: email data, fax data, or PDF data of the form.

[0014] In the above-described information processing method, the at least one report data may include multiple report data.

[0015] According to one embodiment of the present invention, a program is provided for causing a computer to execute the above-described information processing method.

[0016] According to one embodiment of the present invention, an information processing system is provided which includes a control unit that acquires at least one document data, converts the document data into a text dataset containing a plurality of text data, acquires customer identification information indicating a customer associated with the text dataset, and applies the text dataset to a trained model to generate a structured dataset associated with the customer, wherein the trained model is based on the relationship between the text data associated with the document data and the customer master data associated with the customer identification information.

[0017] In the above-described information processing system, generating the structured dataset may include determining the type of report corresponding to the text dataset and selecting a structured template for each of the multiple text data according to the type of report.

[0018] In the above-described information processing system, the control unit may convert the first text data from the plurality of text data into a second text data characterized by the customer master data based on the trained model.

[0019] In the above information processing system, when the first text data includes multiple meanings, the control unit may specify one meaning from the multiple meanings based on the support information for assisting in specifying one meaning from the multiple meanings associated with the first text data and the learned model.

[0020] In the above information processing system, the support information may be arranged near the first text data.

[0021] In the above information processing system, the support information may be the position information of the text data.

[0022] In the above information processing system, the form may be composed of image data or audio data.

[0023] In the above information processing system, the image data may include at least one of form mail data, FAX data, and PDF data.

[0024] In the above information processing system, the at least one form data may include a plurality of form data.

Advantages of the Invention

[0025] According to an embodiment of the present invention, the content of the form can be accurately understood according to each business partner. Further, according to an embodiment of the present invention, a handwritten document can be easily digitized. Further, according to an embodiment of the present invention, information necessary for structured data can be easily generated from a FAX document or a handwritten document.

Brief Description of the Drawings

[0026] [Figure 1] It is a diagram for explaining an information processing system in an embodiment of the present invention. [Figure 2] It is a diagram for explaining the hardware configuration of each device constituting the information processing system in an embodiment of the present invention. [Figure 3]This diagram illustrates the software configuration of the control unit of an information processing server in one embodiment of the present invention. [Figure 4] This figure shows the user trading partner basic master data table in one embodiment of the present invention. [Figure 5] This figure shows the product master data table in one embodiment of the present invention. [Figure 6] This figure shows the selection candidate master data table in one embodiment of the present invention. [Figure 7] This is a flowchart illustrating the structured template generation process of an information processing system in one embodiment of the present invention. [Figure 8] This is a flowchart illustrating the user customer master data generation process of an information processing system in one embodiment of the present invention. [Figure 9] This is a flowchart illustrating the learning process of an information processing system in one embodiment of the present invention. [Figure 10] This is a flowchart illustrating the process of generating a structured dataset of report data in an information processing system according to one embodiment of the present invention. [Figure 11] This is a flowchart illustrating the process of generating a structured dataset of report data in an information processing system according to one embodiment of the present invention. [Figure 12] This is an example of a user interface displayed on a terminal device in one embodiment of the present invention. [Figure 13] This is an example of a user interface displayed on a terminal device in one embodiment of the present invention. [Figure 14] This is an example of report data in one embodiment of the present invention. [Figure 15] This is an example of an OCR-processed text dataset in one embodiment of the present invention. [Figure 16] This is an example of a text dataset generated in one embodiment of the present invention. [Figure 17] This is an example of a structured dataset generated in one embodiment of the present invention. [Figure 18]This is an example of report data in one embodiment of the present invention. [Figure 19] This figure shows the information estimation rule master data table in one embodiment of the present invention. [Figure 20] This is a flowchart illustrating the process of generating a structured dataset of report data in an information processing system according to one embodiment of the present invention. [Modes for carrying out the invention]

[0027] The embodiments described below are examples of embodiments of the present invention, and the present invention is not limited to these embodiments. In the drawings referenced in these embodiments, the same or similar reference numerals are used to denote identical parts or parts having similar functions, and repeated descriptions thereof may be omitted.

[0028] In this specification, "documents" refers to transaction documents used in transactions between business partners (companies), and includes quotations, purchase orders, order confirmations, delivery notes, invoices, receipts, contracts, business cards, etc.

[0029] In this specification, "business partner" includes companies, organizations, and individuals that conduct business with the user (or the company or organization to which the user belongs).

[0030] <First Embodiment> The information processing system in this embodiment will be described in detail with reference to the drawings.

[0031] (1-1. Overall Configuration of the Information Processing System) Figure 1 is a diagram illustrating the information processing system 1 in this embodiment. The information processing system 1 includes an information processing server 10, a database 20, and a terminal device 50.

[0032] The database 20 is connected to the information processing server 10. In Figure 1, the information processing server 10 and the terminal device 50 are connected to a network NW. The network NW is a communication network such as the internet or an intranet, and the appropriate network is used depending on the communication environment.

[0033] In this embodiment, the information processing system 1 may be built using a cloud platform. The form of the cloud platform is not particularly limited, and Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS) can be applied. Specific examples of cloud platforms include Microsoft Azure (manufactured by Microsoft), Amazon Web Services (manufactured by Amazon), and Google Cloud Platform (manufactured by Google). However, the information processing system 1 may be built without being limited to a cloud platform.

[0034] According to Information Processing System 1, it is possible to generate structured datasets tailored to each user's business partners by interpreting the context of the report data acquired from the terminal device 50. The generated structured datasets are displayed on the terminal device 50. The configuration of Information Processing System 1 for realizing this process will be described below.

[0035] (1-2. Hardware Configuration) Figure 2 is a diagram illustrating the hardware configuration of each device that makes up the information processing system 1.

[0036] (1-2-1. Information Processing Server 10) The information processing server 10 includes a control unit 11, a storage unit 12, and a communication unit 13. The control unit 11, storage unit 12, and communication unit 13 may be connected by a communication bus. In this example, a cloud server is used for the information processing server 10. Depending on the system environment, the information processing server 10 may also be an on-premise server or other processing unit. The control unit 11 is an example of a computer that includes arithmetic processing circuits such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), and FPGA (Field Programmable Gate Array). The control unit 11 executes programs stored in the storage unit 12 to realize various functions in the information processing server 10. This enables the information processing server 10 to perform various information processing in the information processing system 1.

[0037] The storage unit 12 includes a memory device and stores an information processing program. The storage unit 12 also stores various data used when the information processing program is executed. This program may be provided recorded on a computer-readable recording medium such as a magnetic recording medium, optical recording medium, magneto-optical recording medium, or semiconductor memory. In this case, the information processing server 10 only needs to include an interface for connecting the recording medium. Here, the storage medium may be defined as a medium separate from the storage unit 12 included in the information processing server 10, or it may be a medium used in the storage unit 12.

[0038] The communication unit 13 includes a communication module and, under the control of the control unit 11, connects to the network NW to send and receive information with other devices connected to the network NW. In this example, the communication unit 13 also connects to the database 20 (also called a database server) to send and receive information. The information registered in the database 20 will be described later. The communication unit 13 may also connect to the database 20 via the network NW. Furthermore, the information registered in the database 20 may be stored in the storage unit 12. In this case, the database 20 does not need to exist.

[0039] In addition to the above configuration, the information processing server 10 may also have other configurations (input / output interfaces) such as a display unit, an audio output unit, and a light-emitting unit.

[0040] (1-2-2. Terminal device 50) The terminal device 50 consists of at least one of the following: a desktop PC (personal computer), a notebook PC, a mobile phone, a smartphone, a tablet terminal, or other electronic application device. The terminal device 50 can function as a client. The terminal device includes a control unit 51, a storage unit 52, a communication unit 53, a display unit 54, and an operation unit 55. The control unit 51 has a configuration that is basically the same as the control unit 11 described above and realizes various functions in the terminal device 50. It executes processing in the terminal device 50 in the information processing system 1.

[0041] The memory unit 52 has a configuration that is basically the same as the memory unit 12, with the only difference being the content of the program instructions it stores; therefore, its explanation is omitted. The communication unit 53 has a configuration that is basically the same as the communication unit 13, with the only difference being the network it can connect to; therefore, its explanation is omitted.

[0042] The display unit 54 includes a display device whose display content is controlled by the control unit 51. The operation unit 55 may include a keyboard, switches, a handle, etc., and outputs information corresponding to the operation to the control unit 51. The operation unit 55 may also include a touch sensor and output information corresponding to the position operated by the user to the control unit 51. The touch sensor may be provided on the display area of ​​the display unit 54. In other words, the display unit 54 and the operation unit 55 may constitute a touch panel.

[0043] (1-3. Software configuration of the information processing server) Figure 3 shows the software configuration of the control unit 11 of the information processing server 10. The control unit 11 includes an acquisition unit 11a, a preprocessing unit 11b, a machine learning unit 11c, an inference processing unit 11d, a generation unit 11e, and a display control unit 11f.

[0044] The acquisition unit 11a has the function of acquiring various types of information from each device. For example, the acquisition unit 11a acquires various types of information such as user identification information, user business partner identification information, user business partner information, report data, and user business partner master data. The acquired information is stored in the database 20. User business partner identification information refers to information that can identify the business partners of the user (the company to which the user belongs). User business partner information refers to various types of information that the user has regarding each of the user's business partners. User business partner master data refers to master data that the user has regarding each of the user's business partners.

[0045] The preprocessing unit 11b performs preprocessing on the acquired information. In this example, the preprocessing unit 11b preprocesses the user customer master data. Techniques such as chunking and vectorization may be used for preprocessing. These preprocessing steps allow for optimization in specific data retrieval processes. The preprocessed information is stored in the database 20 as appropriate. The preprocessing unit 11b can also perform OCR processing on customer request datasets, such as order form data, to generate text data.

[0046] The machine learning unit 11c has the function of performing machine learning based on acquired information. In this example, the machine learning unit 11c learns company-specific master data patterns, such as product names and product codes, stored in the acquired user business partner master data. The machine learning unit 11c also performs machine learning by associating text data (text dataset) obtained by OCR processing of previously acquired form data with user business partner master data that characterizes the user's business partners. This makes it possible to learn the content (context) written in the form based on different interpretation criteria for each business partner.

[0047] The inference processing unit 11d has the function of performing inference processing on OCR-extracted text data based on a trained model. During this process, normalization processing may be performed according to patterns obtained through machine learning. Furthermore, vector search or LLM (Large Language Model) can be used for the inference processing.

[0048] The generation unit 11e has the function of generating various types of information. In this example, the generation unit 11e generates a structured dataset.

[0049] The display control unit 11f has the function of controlling the display of various generated information on the terminal device 50.

[0050] (1-4. Various Data Tables) Next, we will describe the various data tables used in the information processing system 1. In one embodiment of the present invention, all or some of the various data tables are associated with user business partner identification information that indicates the user's business partners, and can therefore be collectively referred to as user business partner master data (or user business partner master data group).

[0051] (1-4-1. User Business Partner Basic Master Data Table) Figure 4 shows the user customer basic master data table 100. The user customer basic master data table 100 contains information about the user's customers. In this example, the user customer basic master data table 100 includes customer identification information 101, customer name 102, and address information 103, but it may also include other information. The user customer basic master data table 100 is stored in the database 20 of the information processing server 10.

[0052] (1-4-2. Product Master Data Table) Figure 5 shows the product master data table 200. The product master data table 200 contains information about products sold by the user's trading partners. In this example, the product master data table 200 includes trading partner identification information 201, product code information 202, product (color) information 203, design information 204, and price information 205. The product master data table 200 is stored in the database 20 of the information processing server 10. In addition, the product master data table 200 may also include various other information such as category, product name, JAN code, supplier, cost / selling price, and weight.

[0053] (1-4-3. Selection Candidate Master Data Table) Figure 6 shows the selection candidate master data table 300. The selection candidate master data table 300 contains selection candidates for identifying the meaning of a single text data. In this example, the selection candidate master data table 300 includes user customer identification information 301, product model number information 302, selection candidate 303, and remarks information 304. The selection candidate master data table 300 is stored in the database 20. In this embodiment, appropriate information is selected from the selection candidates contained in the selection candidate master table 300.

[0054] (1-5. Information Processing Methods) Next, the information processing method implemented in the information processing system of this embodiment will be described. Figures 7 to 10 are flowcharts showing the information processing of the information processing system 1 of this embodiment.

[0055] (1-5-1. Structured Template User Business Partner Master Data Generation Process) Figures 7-8 are flowcharts showing the generation process of structured templates and user business partner master data in this embodiment. First, a login user interface is displayed on the terminal device 50 of the information processing system (step S101). At this time, the user, such as the business partner's information manager, enters user identification information into the login user interface (step S103). User identification information refers to information that can identify the user (the company to which the user belongs). In this example, a username and password are entered, but the input content can be changed as appropriate, such as a company code or employee number. In addition, information other than text data, such as facial recognition data, may also be entered. The entered user identification information is sent to the information processing server 10 (step S105), and the information processing server 10 acquires the user identification information (step S107). At this time, the information processing server 10 performs user authentication processing based on the user identification information (step S109). At this time, along with the user identification information, information for identifying the user's business partners (user business partner identification information) is authenticated (acquired). User business partner identification information may include multiple pieces of information to identify the business partner.

[0056] Once user authentication is complete, the information processing server 10 generates instruction information to display a screen for configuring a structured template (step S111) and sends the instruction information to display the structured template configuration screen to the terminal device 50 (step S113). Upon receiving the instruction information, the terminal device 50 displays the user interface for configuring the structured template on the display unit 54 (step S115). The structured template is a format for generating a structured dataset.

[0057] Figure 12 shows a user interface 400 for setting structured templates to generate structured data. In this example, the user interface 400 for setting structured templates includes a category selection input section 401 and a subject input section 403. More specifically, the subject input section 403 is used to input various information such as the report title, creation date and time, numerical information, product information, delivery date, delivery location, and payment terms. The user inputs the various information (step S117). The input information is sent to the information processing server 10 (step S119). The information processing server 10 can generate (set) structured templates for each type of report data by obtaining the structured template setting information (step S121) (step S123).

[0058] The information processing server 10 generates instruction information for displaying a screen for inputting various prompt information (and user business partner master data) to generate structured data (step S131), and sends the prompt information input screen display instruction information to the terminal device 50 (step S133). When the terminal device 50 receives the instruction information, it displays a user interface for inputting prompt information (user business partner master data) on the display unit 54 (step S135).

[0059] Figure 13 shows a user interface 500 for inputting prompt information (user customer master data). The input user interface 500 includes a prompt name 501, a prompt input section 503, and a save button 505. In one embodiment of the present invention, the user inputs various prompt information such as text data, table data, and graph data into the prompt input section 503 (step S137). For example, text data such as "Please determine the description of the item as the product name" is input as prompt information. The input prompt information is sent to the information processing server 10 (step S139), and the information processing server 10 acquires the prompt information (S141). At this time, the information processing server 10 can generate user customer master data based on the acquired prompt information (step S143).

[0060] In this example, the user customer master data includes customer-specific information (such as customer-specific terminology or text data interpretation rules) characterized for each user customer. In this example, the user customer master data includes the user customer basic master data table shown in Figure 4 and the product master data table shown in Figure 5. The user customer master data may also include a selection candidate master data table 300 specific to each customer, as shown in Figure 6. The selection candidate master data table 300 may include abbreviation data and specific business term data used within the customer. More specifically, in the selection candidate master data table 300, if the customer is "Sample Co., Ltd.", and the text data includes the selection candidates 481-SHN-SM (candidate 1), 〃-UM (candidate 2), and UM (candidate 3), it is set to recognize the product as having the model number "481-SHN-SM". The user customer master data is used so that the learning results (interpretations) differ for each customer. The generated user customer master data may be stored in the database 20.

[0061] (1-5-2. Learning Process) Figure 9 is a flowchart of the learning process. First, the information processing server 10 performs preprocessing on the user business partner master data before machine learning processing (step S201). In this example, the information processing server 10 performs a partitioning (vectorization) process on the user business partner master data. The vectorization method is Bag of Words (BoW), TF-IDF (Term Methods such as Frequency-Inverse Document Frequency (FHz), Word2Vec, and BERT (Bidirectional Encoder Representations from Transformers) are used. By converting user customer master data (e.g., text data) into vectors, the similarity between different sentences can be compared numerically. Vectorization enables numerical comparison and manipulation of data, making it easier to learn user customer-specific information.

[0062] Next, the information processing server prepares a text dataset by performing OCR processing on the previously acquired forms (step S203). The previously acquired text dataset is stored in the database 20. The previously acquired text data may be selected according to the type of form.

[0063] Next, machine learning is performed using user customer master data and a pre-acquired text dataset (step S205). Known learning methods such as backpropagation and genetic algorithms (GA) may be used for machine learning. By repeatedly performing machine learning, a trained model is generated (step S207). This trained model is based on the relationship between text data associated with report data and user customer master data associated with user customer identification information. At this time, training data may be applied in which text data is input and text data corresponding to a part (or all) of the correct user customer master data is output. By using one embodiment of the present invention, it is possible to generate results (structured datasets) based on customer-specific information (knowledge) based on customer-specific business terms and interpretation rules (rules) understood by machine learning for input report data. Furthermore, the above trained model can accurately understand customer-specific terms that users include in report data.

[0064] Furthermore, the generated trained model may be fine-tuned using new data as needed. This allows for optimization of the trained model's performance.

[0065] (1-5-3. Structured Dataset Generation Process) Figures 10 and 11 are flowcharts illustrating the structured dataset generation process. First, the user enters user identification information into a login user interface based on the information processing program displayed on the terminal device 50 (steps S301, S303). The entered information is sent to the information processing server 10 (step S305), and the information processing server 10 obtains the user identification information (step S307). At this time, the information processing server 10 performs user authentication processing based on the user identification information (step S309). This authenticates the user identification information.

[0066] Once user authentication is complete, the information processing server 10 generates instruction information to display a screen for entering form data (step S311) and transmits this instruction information to the terminal device 50 (step S313). Upon receiving this instruction information, the terminal device 50 displays a user interface (UI) for entering form data on the display unit 54 (step S315). The user enters the form data on the input screen displayed on the terminal device 50 (step S317). Figure 14 shows the user interface 600 for displaying form data entered by the user and displayed on the display unit 54. In this example, the form data is image data consisting of a faxed document of a form (purchase order) handwritten by a representative of a business partner.

[0067] The entered form data is sent to the information processing server 10 by pressing (clicking) the send button provided on the input screen (step S319), and the information processing server 10 retrieves the form data (step S321). In this example, purchase order data is retrieved as the form data.

[0068] Next, the information processing server 10 generates a structured dataset based on the acquired report data (step S323). Figure 11 is a flowchart of the structured dataset generation process. As shown in Figure 11, first the information processing server 10 recognizes (determines) the type of report corresponding to the text dataset (step S32301), and selects a structured template corresponding to the type of report (step S32303).

[0069] Next, the information processing server 10 performs OCR processing on the form data (step S32305). This generates (converts) a text dataset 700 containing multiple text data as shown in Figure 15. The text dataset is processed as prompt information. Next, the information processing server 10 inputs (temporarily inputs) each text data into a predetermined part of the selected structured template (step S32307).

[0070] At this time, the information processing server 10 can obtain user business partner identification information that indicates the user's business partners associated with the generated text dataset. At this time, user business partner identification information (business partner number or business partner name) contained in the text data may be extracted. Alternatively, the user business partner identification information may be entered by the user. By obtaining user business partner identification information, it is possible to conform to interpretation rules more appropriately and quickly.

[0071] Next, the information processing server 10 applies the generated text dataset to a trained model and performs calculations (step S32309). At this time, the information processing server 10 performs inference processing using the trained model so that the acquired report data conforms to interpretation rules characterized for each user's business partner (step S32311). Vector search or LLM (Large Language Model) can be used for the inference processing.

[0072] In this embodiment, before performing the inference process, multiple similar text data may be generated by referring to the user trading partner master data. LLM, Python, or other programming languages ​​can be used to generate the similar text. In this case, normalization processing may be performed on each data point of the previously acquired user trading partner master data (e.g., product master data) according to a pattern generated or set by the information processing program (e.g., a Python library). If multiple similar texts are generated, a matching search may be performed against the user trading partner master data using the multiple similar texts as keywords. This reduces the amount of information to be processed for inference and allows for faster retrieval of the correct data.

[0073] Furthermore, when performing inference processing, the data included in the generated dataset and the data included in the user business partner master data are each vectorized and searched (vector search processing). When performing vector searches, a vector search may be performed by product name, or a composite vector may be used, which combines the vector of the product name, the vector of the product code, and the vector relating the positional relationship between the product name and the product code. In addition, during the inference processing, multiple similar texts as described above may be vectorized and a matching search may be performed against the vector in the user business partner master data.

[0074] In this embodiment, ambiguous text data (also referred to as "first text data") is converted into specific text data (also referred to as "second text data") characterized by user customer master data. For example, when form data is sent from a customer called "Sample Co., Ltd." and is a text dataset containing multiple text data, and as shown in Figure 15, the product information (in this example, the product model number) includes ambiguous text data 701 ("〃-UM"), the system is configured to recognize the product as having the model number "481-SHN-UM" based on the trained model. As a result, a text dataset 800 is generated that includes text data 801 composed of accurate information, as shown in Figure 16, based on the recognized content. Consequently, a structured dataset associated with customer identification information is generated and output (step S32313). Furthermore, the information processing server 10 can perform CSV (Comma-Separated Values) processing on the text dataset (step S32315) to generate text data with respective meanings.

[0075] Furthermore, when generating structured datasets, generative models may be applied as appropriate. Common large-scale language models (LLMs) such as ChatGPT (OpenAI), BERT, T5, LLaMa, and Claude can be used as generative models. Additionally, Graph RAG (Graph Retrieval-Augmented Generation) processing may be performed when generating structured datasets. Graph RAG refers to the process of applying a graph structure to search and generation processes. By performing Graph RAG processing, it becomes possible to leverage the relationships and structures between data, enabling more advanced applications.

[0076] Let's return to Figure 10 for explanation. Next, the information processing server 10 generates instruction information for displaying the acquired structured dataset (step S325). The generated instruction information for displaying the structured dataset is sent to the terminal device 50 (step S327), and the terminal device 50 acquires the structured dataset of the report data and displays it on the display unit 54 (step S329). Figure 17 is an example of a user interface 900 for displaying the generated structured dataset of the report data. In this example, the user interface 900 for displaying the structured dataset includes subject 901, delivery date 903, delivery location 905, payment terms 907, subtotal 909, consumption tax 911, total 913, description 915, quantity 917, unit 919, unit price 921, and amount 923. The structured dataset can include all the items written on the handwritten purchase order. The structured dataset may be a tabular data dataset, text data, or image data. This concludes the information processing according to one embodiment of the present invention.

[0077] By using one embodiment of the present invention, the system can pre-train the user's customer master data and then perform OCR processing on the document data, thereby accurately understanding the document data and providing digitized data.

[0078] Furthermore, by using this embodiment, even document data that is difficult to understand, such as handwritten documents, can be easily understood and easily digitized. Therefore, structured data can be easily generated from fax documents and handwritten documents.

[0079] <Second Embodiment> This embodiment describes an information processing method different from that of the first embodiment. Specifically, it describes an example in which a single text data (first text data) has multiple meanings. Parts that overlap with the first embodiment will be omitted from the explanation as appropriate.

[0080] Figure 18 shows an example of a form data 1000 consisting of acquired handwritten information. The form data 1000 includes information 1001 that has multiple meanings, such as "6 / 1".

[0081] Figure 19 shows the candidate selection dataset 1100. In this example, the candidate selection dataset 1100 includes term information 1101, rule number information 1102, support information 1103, and semantic information 1104.

[0082] Figure 20 is a flowchart of the context recognition process in the generation of a structured dataset. As shown in Figure 20, the information processing server 10 recognizes the context of the text dataset based on a trained model (step S32311A1). At this time, the information processing server 10 determines whether the text data (first text data) contained in the text dataset has multiple meanings (step S32311A2). If the text data does not have multiple meanings (step S32311A2; No), the information processing server 10 identifies the meaning of the text data (step S32311A5). At this time, the text data is transformed to correspond to a specific (single) meaning.

[0083] On the other hand, if the text data has multiple meanings (step S32311A2; Yes), the information processing server 10 detects supporting information from within the text dataset (step S32311A3). Supporting information is information that helps identify the meaning of text data that has multiple meanings. Supporting information may be text data adjacent to the text data in question, location information (coordinate information) where a term is placed, header information, or type information of the report data. Next, the information processing server 10 identifies one meaning from the multiple meanings based on the supporting information and the trained model. Specifically, the information processing server 10 recognizes the relationship between the text data with multiple meanings and the supporting information (step S32311A4). Specifically, when the term "○○ / △△" satisfies rule 1 (placed in the upper right (supporting information is the location information of the text data)), it is determined to be a date (○○ month △△ day). When the term "○○ / △△" satisfies Rule 2 (placed adjacent to the product name or model number (support information is placed near the text data information)), it is determined that ○○ is the model number and △△ is the quantity. When the term "○○ / △△" satisfies Rule 3 (placed in the lower right), it is determined that it is the page (currently ○○ pages / total △△ pages). As a result, the meaning of the text data is identified even when it has multiple meanings (step S32311A5).

[0084] This allows for accurate understanding of the content of report data, even when the report data contains terms with multiple meanings.

[0085] <Variation> This disclosure is not limited to the embodiments described above, but includes various other modifications. For example, the embodiments described above are described in detail for the purpose of explaining this disclosure clearly, and are not necessarily limited to those having all the configurations described. Furthermore, it is possible to replace parts of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add configurations from other embodiments to the configuration of one embodiment. In addition, it is possible to add, delete, or replace parts of the configuration of each embodiment with other configurations. Some modifications are described below. Note that examples of modifications of each embodiment can also be applied as examples of modifications of other embodiments.

[0086] (1) In one embodiment of the present invention, an example was shown in which information for each business partner (user business partner master data) is obtained from a terminal device 50, but the present invention is not limited thereto. For example, it can be applied not only to business partners but also to sole proprietors, other corporations, hospitals, research institutes and other organizations.

[0087] (2) One embodiment of the present invention has shown an example of generating one structured data from one report data, but the present invention is not limited thereto. For example, multiple text datasets may be generated from multiple report data, the context of the multiple text datasets may be understood, and the meaning of each text data may be identified (text data transformation process).

[0088] (3) In one embodiment of the present invention, an example was shown in which information entered into the terminal device 50 is acquired by the information processing server 10 and a learning process is executed. However, learning may also be performed based on user customer master data existing in the information processing server 10 or the database 20, or the user customer master data may be uploaded as CSV data, or the answer information to the question information sent from the information processing server 10 (or pre-set in the program) may be used as the user customer master data. Alternatively, the learning process may be performed based on the information (answer information) entered in response to the question information sent from the information processing server 10 (or pre-set in the program). In this case, order number information, product name information, quantity information, and amount information may be entered. The user enters various information and presses (clicks) the send button to send the user customer master data to the information processing server 10. Alternatively, the user may directly enter the user customer master data into the terminal device 50.

[0089] (4) Although one embodiment of the present invention has been described using a form, the present invention is not limited thereto. The present invention can also be applied to documents other than forms (for example, internal company documents) and handwritten documents.

[0090] (5) In the first embodiment of the present invention, an example was shown in which information processing is performed by the information processing server 10, but the present invention is not limited thereto. In one embodiment of the present invention, the information processing server 10 may be composed of a plurality of servers according to their functions, or different servers (such as a web server and an application server) may be provided to perform their respective processes related to the information processing system of one embodiment of the present invention.

[0091] (6) In the first embodiment of the present invention, an example was shown in which preprocessing (vectorization) is performed in the learning process, but the present invention is not limited thereto. Vectorization processing may also be performed at the time when a structured dataset of report data is generated.

[0092] (7) In the first embodiment of the present invention, the document data was shown as image data consisting of a faxed document of a handwritten document (order form), but the present invention is not limited thereto. For example, email data or PDF data of the document may be used. In addition to image data, the document data may also be audio data. In the case of audio data, it may be audio data input to the terminal device 50, or it may be recorded audio data. [Explanation of Symbols]

[0093] 10...Information processing server, 11...Control unit, 11a...Acquisition unit, 11b...Preprocessing unit, 11c...Machine learning unit, 11d...Inference processing unit, 11e...Generation unit, 11f...Display control unit, 12...Storage unit, 13...Communication unit, 20...Appropriate database, 20...Database, 50...Terminal device, 51...Control unit, 52...Storage unit, 53...Communication unit, 54...Display unit, 55...Operation unit, 100...User trading partner basic master data table, 1 01...Customer Identification Information, 102...Customer Name, 103...Address Information, 200...Product Master Data Table, 201...Customer Identification Information, 202...Product Code Information, 203...Product (Color) Information, 204...Design Information, 205...Price Information, 300...Selection Candidate Master Table, 300...Selection Candidate Master Data Table, 301...User Customer Identification Information, 302...Product Model Number Information, 303...Selection Candidate, 304...Remarks Information, 400... ··User interface for setting structured templates, 401···Category selection input section, 403···Subject input section, 500···User interface for inputting prompt information, 501···Prompt name, 503···Prompt input section, 505···Save button, 600···User interface for displaying report data, 700···Text dataset, 701···Text data, 800···Text dataset, 801···Text data, 900···Structure User interface for displaying the creation dataset, 901...Subject, 903...Delivery date, 905...Delivery location, 909...Subtotal, 911...Consumption tax, 913...Total, 915...Description, 917...Quantity, 919...Unit, 921...Unit price, 923...Amount, 1000...Report data, 1001...Information, 1100...Selection candidate dataset, 1101...Term information, 1102...Rule number information, 1103...Support information, 1104...Semantic information

Claims

1. Computers To obtain at least one report data, Converting the aforementioned report data into a text dataset containing multiple text data, This involves obtaining user business partner identification information that indicates the user's business partners, associated with the aforementioned text dataset, Applying the aforementioned text dataset to the trained model, the first text data among the multiple text data is transformed into a second text data characterized by the user business partner master data associated with the user business partner identification information. This includes generating a structured dataset associated with the user's business partners using the second text data, The aforementioned trained model is a model that has been trained to take text data associated with the report data as input and output text data corresponding to at least a portion of the user customer master data. Information processing methods.

2. Generating the aforementioned structured dataset is Determine the type of report corresponding to the aforementioned text dataset. This includes selecting a structured template for each of the aforementioned multiple text data according to the type of report, The information processing method according to claim 1.

3. When the first text data contains multiple meanings, the system identifies one meaning from the multiple meanings based on supporting information associated with the first text data and the trained model. The information processing method according to claim 1.

4. The support information is located near the first text data. The information processing method according to claim 3.

5. The aforementioned support information is the location information of the text data. The information processing method according to claim 3.

6. The aforementioned report data consists of image data or audio data. The information processing method according to claim 1.

7. The aforementioned image data includes at least one of the following: email data, fax data, or PDF data of the document. The information processing method according to claim 6.

8. The aforementioned at least one report data includes multiple report data, The information processing method according to claim 1.

9. A program for causing a computer to execute the information processing method described in any one of claims 1 to 8.

10. Obtain at least one report data, The aforementioned report data is converted into a text dataset containing multiple text data, Associated with the aforementioned text dataset, user business partner identification information indicating the user's business partners is obtained, The text dataset is applied to the trained model to convert the first text data among the multiple text data into a second text data characterized by the user business partner master data associated with the user business partner identification information. The system includes a control unit that generates a structured dataset associated with the business partners using the second text data, The aforementioned trained model is a model that has been trained to take text data associated with the report data as input and output text data corresponding to at least a portion of the user customer master data. Information processing system.

11. Generating the aforementioned structured dataset is Determine the type of report corresponding to the aforementioned text dataset. This includes selecting a structured template for each of the aforementioned multiple text data according to the type of report, The information processing system according to claim 10.

12. The control unit, When the first text data contains multiple meanings, the system identifies one meaning from the multiple meanings based on supporting information associated with the first text data and the trained model. The information processing system according to claim 10.

13. The support information is located near the first text data. The information processing system according to claim 12.

14. The aforementioned support information is the location information of the text data. The information processing system according to claim 12.

15. The aforementioned report data consists of image data or audio data. The information processing system according to claim 10.

16. The aforementioned image data includes at least one of the following: email data, fax data, or PDF data of the document. The information processing system according to claim 15.

17. The aforementioned at least one report data includes multiple report data, The information processing system according to claim 10.

Images