Information processing device, database generation method, and database generation program

Abstract

The system uses a generation AI to modify document data and generate a database. The information processing device includes: an identification unit that identifies a search range for the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed; a search unit that searches for document data corresponding to the document data to be processed from the identified search range; a modification unit that instructs the generation AI to modify the document data to be processed using prompts that designate the document data to be processed as the document data to be modified and the reference document data generated based on the searched document data as the document data to be referenced during modification; and a storage unit that updates the database by adding information based on the analysis results to the document data to be processed modified by the generation AI and storing it in an area within the database corresponding to the attributes of the document data to be processed.

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, a database generation method, and a database generation program.

Background Art

[0002] As a technology for digitizing handwritten documents and registering them in a database as document data, OCR (Optical Character Recognition) technology is known. The OCR technology is a technology for digitizing a handwritten document read by a scanner by performing character recognition processing.

[0003] On the other hand, a technology is known for correcting misrecognition results that occur during character recognition processing using generative AI (Artificial Intelligence) such as a large language model.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, there is a problem that correction of misrecognition results using generative AI cannot handle scenes that require specific technical fields and advanced expertise. Therefore, in such scenes, it is required to perform processing such as providing information on a specific technical field and information such as advanced expertise to the generative AI in advance.

[0006] The present disclosure provides a technology for correcting document data using generative AI and generating a database.

Means for Solving the Problems

[0007] The information processing device relating to the first aspect of this disclosure is: A specific unit identifies the search range of the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed, A search unit searches for document data corresponding to the document data to be processed from the specified search range, A modification unit instructs the generating AI to modify the document data to be processed, using prompts that designate the document data to be processed as the document data to be modified, and the reference document data generated based on the retrieved document data as the document data to be used as reference during modification. The system includes a storage unit that updates the database by adding information based on the analysis results to the document data to be processed, which has been modified by the generating AI, and storing it in an area of ​​the database corresponding to the attributes of the document data to be processed.

[0008] A second aspect of this disclosure is an information processing apparatus described in the first aspect, The specified unit analyzes at least one of the following: the technical field, technical content, technical name, method, and background of the document data to be processed.

[0009] A third aspect of this disclosure is an information processing device described in the first or second aspect, The attributes of the document data to be processed include, at a minimum, the department where the document data to be processed is stored, and the range of products handled by that department.

[0010] A fourth aspect of this disclosure is an information processing apparatus described in any of the first to third aspects, The modification unit instructs the document data to be processed to be modified on a page-by-page basis.

[0011] A fifth aspect of this disclosure is an information processing apparatus as described in the fourth aspect, When the modification unit instructs the AI ​​to modify the page to be processed in the document data to be processed, it also transmits the pages before and after the page to be processed to the generating AI.

[0012] A sixth aspect of this disclosure is an information processing device described in any of the first to fifth aspects, The system further includes an initial database generation unit that generates an initial database before the aforementioned database is updated, The initial database generation unit analyzes the contents of the document data to be stored in the initial database, adds information based on the analysis results, and stores the data in an area corresponding to the attributes of the document data.

[0013] A seventh aspect of this disclosure is a database generation method, A step of identifying the search range of the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed, The process of searching for document data corresponding to the document data to be processed from the specified search range, The process involves instructing the generating AI to modify the document data to be processed by using a prompt that designates the document data to be processed as the document data to be modified, and the reference document data generated based on the retrieved document data as the document data to be used as a reference during modification. The computer performs the following steps: adding information based on the analysis results to the document data to be processed, which has been corrected by the generating AI, and updating the database by storing it in an area of ​​the database corresponding to the attributes of the document data to be processed.

[0014] The eighth aspect of this disclosure is a database generation program, A step of identifying the search range of the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed, The process of searching for document data corresponding to the document data to be processed from the specified search range, Using a prompt that designates the document data to be processed as the document data to be corrected and the reference document data generated based on the retrieved document data as the document data to be referred to during correction, instruct the generative AI to correct the document data to be processed; Cause a computer to execute a step of updating the database by adding information based on the analysis result to the document data to be processed corrected by the generative AI and storing it in an area in the database according to the attributes of the document data to be processed.

Advantages of the Invention

[0015] According to the present disclosure, it is possible to provide a technique for correcting document data using generative AI and generating a database.

Brief Description of the Drawings

[0016] [Figure 1] FIG. 1 is a diagram showing an example of the system configuration of a database generation system. [Figure 2] FIG. 2 is a diagram showing an example of document data for a DB. [Figure 3] FIG. 3 is a diagram showing an example of the hardware configuration of an information processing apparatus. [Figure 4] FIG. 4 is a diagram showing an example of the functional configuration of an information processing apparatus. [Figure 5] FIG. 5 is an example of a flowchart showing the flow of initial database generation processing. [Figure 6] FIG. 6 is a diagram showing an example of an initial database generation screen. [Figure 7] FIG. 7 is an example of a flowchart showing the flow of database generation processing. [Figure 8] FIG. 8 is a diagram showing an example of a database generation screen. [Figure 9A] FIG. 9A is a diagram showing an example of a system prompt. [Figure 9B] FIG. 9B is a diagram showing examples of an assistant prompt and a user prompt. [Figure 10A] Figure 10A shows an example of the target pages in OCR document data. [Figure 10B] Figure 10B shows an example of a summary of the target page, a list of search results, and reference document data. [Figure 10C] Figure 10C shows an example of corrected OCR document data. [Modes for carrying out the invention]

[0017] Each embodiment will be described below with reference to the attached drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0018] [First Embodiment] <System Configuration of the Database Generation System> First, the system configuration of the database generation system equipped with the information processing device according to the first embodiment will be described. Figure 1 is a diagram showing an example of the system configuration of the database generation system.

[0019] The database generation system 100 is applicable, for example, to a scenario where a company that manufactures product lines in multiple technological fields digitizes handwritten documents stored within its premises and stores them as document data in a database.

[0020] As shown in Figure 1, the database generation system 100 includes an information processing device 110, a reader 111, an information processing device 120, a reader 121, a server device 130, and an information processing device 140.

[0021] In the database generation system 100, the information processing device 110, the information processing device 120, the server device 130, and the information processing device 140 are interconnected via the network 160. The information processing device 140 is connected to the server device 150 via the network 170.

[0022] The information processing device 110 is installed in department α, which handles a portion of the manufactured product range. Before the database is generated, the information processing device 110 reads handwritten document data stored in department α using the reading device 111 and acquires OCR document data generated by character recognition processing. The information processing device 110 then transmits the acquired OCR document data to the information processing device 140.

[0023] When the database is used after it has been generated, the information processing device 110 accesses the server device 130 and displays the document data stored in the database to a viewer (not shown).

[0024] The information processing device 120 is installed in department β, which handles a portion of the manufactured product range. Before the database is generated, the information processing device 120 reads handwritten document data stored in department β using the reading device 121 and obtains OCR document data generated by character recognition processing. The information processing device 120 then transmits the obtained OCR document data to the information processing device 140.

[0025] When the database is used after it has been generated, the information processing device 120 accesses the server device 130 and displays the document data stored in the database to a viewer (not shown).

[0026] The server device 130 has a database for storing document data. In this embodiment, the product group is classified into multiple product ranges (four product ranges in the example of Figure 1), and the database is divided into areas corresponding to each product range to store the document data.

[0027] In Figure 1, the data storage unit 131 for product range A indicates an area where document data related to products classified as product range A from the above product group is stored. The data storage unit 132 for product range B indicates an area where document data related to products classified as product range B from the above product group is stored. The data storage unit 133 for product range C indicates an area where document data related to products classified as product range C from the above product group is stored. The data storage unit 134 for product range D indicates an area where document data related to products classified as product range D from the above product group is stored.

[0028] Note that while Figure 1 shows an example where document data is stored in areas corresponding to product ranges, the database configuration is not limited to this. Other configurations are acceptable as long as they facilitate the use of the database. Specifically, document data may be stored in areas corresponding to other attributes of the document data (attributes other than product range, such as department). This is because, considering that handwritten documents were originally stored by department, it may be easier to use if document data is stored by department in the database as well. Alternatively, document data may be subdivided and stored by product range and by department. This is because, even within the same product range, different departments will have different viewers, so it may be easier to use if the data is subdivided and stored separately.

[0029] The information processing device 140 is a device operated by the user 141. The information processing device 140 generates an initial database (referred to as the initial database) for the server device 130. For example, the information processing device 140 collects document data (error-free document data) managed by each department and adds information based on the analysis results obtained by analyzing the contents of the document data to the document data. In this way, the information processing device 140 generates document data for the database (hereinafter referred to as DB document data). The information processing device 140 generates the initial database by storing the generated DB document data in an area corresponding to the attributes of the document data (in the example in Figure 1, the product range).

[0030] The information processing device 140 corrects the misrecognition results of the OCR document data transmitted from the information processing devices 110 and 120, and then generates document data for the database. The information processing device 140 updates the initial database or database by storing the generated document data for the database in an area corresponding to the attributes of the OCR document data (product range in the example in Figure 1), and then generates the database.

[0031] In correcting misrecognition results contained in OCR document data, the information processing device 140 first determines the search range of the initial database based on the analysis results obtained by analyzing the attributes and contents of the OCR document data.

[0032] Next, the information processing device 140 searches for database document data corresponding to the OCR document data from the database document data included in the specified search range. Based on the retrieved database document data, the information processing device 140 generates reference document data.

[0033] Next, the information processing device 140 generates a prompt that designates the OCR document data as "document data to be corrected" and the generated reference document data as "document data to be referenced when correcting".

[0034] Next, the information processing device 140 uses the generated prompt to instruct the server device 150's generation AI, such as a large-scale language model, to correct the misrecognition result of the OCR document data.

[0035] Thus, in this embodiment, considering that in the case of generative AI such as large-scale language models, correcting misrecognition results may require information on specific technical fields, advanced expertise, etc., • Document data for the database containing information on specific technical fields, advanced expertise, etc., is prepared in advance. When correcting misrecognition results in OCR document data, the search range is narrowed based on attributes and analysis results. · Search for the document data for the DB corresponding to the OCR document data from the document data for the DB included in the narrowed search range, · Generate reference document data based on the searched document data for the DB, · Provide the reference document data to the generation AI.

[0036] Thus, according to the present embodiment, it becomes possible to provide the generation AI with information suitable for correcting the misrecognition result of the OCR document data to be processed. As a result, the generation AI can appropriately correct the misrecognition result.

[0037] That is, according to the present embodiment, it is possible to provide a technique for appropriately correcting document data using the generation AI and generating a database.

[0038] <Document data for the DB> Next, the initial database of the server device 130 or the document data for the DB stored in the database will be described. FIG. 2 is a diagram showing an example of the document data for the DB, and shows the relationship between each document data for the DB, the attributes of each document data for the DB, and the analysis result obtained by analyzing the content of each document data for the DB.

[0039] In the case of the example of FIG. 2, it is shown that the department α is a department that handles products classified into product range A, products classified into product range B, and products classified into product range D among the product groups to be manufactured.

[0040] In the case of the example of FIG. 2, it is shown that the department β is a department that handles products classified into product range B, products classified into product range C, and products classified into product range D among the product groups to be manufactured.

[0041] As shown in Figure 2, the DB document data related to products classified under product range A handled by department α includes document data A_1 to document data A_a, and this DB document data is stored in the product range A data storage unit 131. In addition to attribute information, each of document data A_1 to document data A_a has information based on the analysis results obtained by analyzing the contents of document data A_1 to document data A_a (in the example in Figure 2, the technical field and technical content) added to it.

[0042] As shown in Figure 2, the DB document data related to products classified under product range B handled by department α includes document data B_1 to document data B_b, and this DB document data is stored in the product range B data storage unit 132. In addition to attribute information, document data B_1 to document data B_b each have information based on analysis results obtained by analyzing the contents of document data B_1 to document data B_b (in the example in Figure 2, technical field and technical content) added to them.

[0043] As shown in Figure 2, the DB document data related to products classified under product range D handled by department α includes document data D_1 to D_d, and this DB document data is stored in the product range D data storage unit 134. In addition to attribute information, each of the document data D_1 to D_d has information based on the analysis results obtained by analyzing the contents of document data D_1 to D_d (in the example in Figure 2, the technical field and technical content) added to it.

[0044] As shown in Figure 2, the DB document data related to products classified under product range B handled by department β includes document data B_b+1 to document data B_b+n, and this DB document data is stored in the product range B data storage unit 132. In addition to information indicating attributes, each of the document data B_b+1 to document data B_b+n has information based on the analysis results obtained by analyzing the contents of document data B_b+1 to document data B_b+n (in the example in Figure 2, the technical field and technical content) added to it.

[0045] As shown in Figure 2, the DB document data related to products classified under product range C handled by department β includes document data C_1 to C_c, and this DB document data is stored in the product range C data storage unit 133. In addition to attribute information, each of the document data C_1 to C_c has information based on the analysis results obtained by analyzing the contents of document data C_1 to C_c (in the example in Figure 2, the technical field and technical content) added to it.

[0046] As shown in Figure 2, the DB document data related to products classified under product range D handled by department β includes document data D_d+1 to D_d+m, and this DB document data is stored in the product range D data storage unit 134. In addition to attribute information, each of the document data D_d+1 to D_d+m has information based on the analysis results obtained by analyzing the contents of document data D_d+1 to D_d+m (in the example in Figure 2, the technical field and technical content) added to it.

[0047] <Hardware configuration of the information processing device> Next, the hardware configuration of the information processing device 140 will be described. Figure 3 shows an example of the hardware configuration of the information processing device. As shown in Figure 3, the information processing device 140 includes a processor 301, memory 302, auxiliary storage device 303, interface device 304, communication device 305, and drive device 306. The hardware components of the information processing device 110 are interconnected via a bus 307.

[0048] The processor 301 has various computing devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 301 executes various programs (for example, a database generation program) by reading them into the memory 302.

[0049] Memory 302 has main memory devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 301 and memory 302 form a so-called computer, and the computer realizes various functions by the processor 301 executing various programs read into memory 302.

[0050] The auxiliary storage device 303 stores various programs and various data used when these programs are executed by the processor 301.

[0051] Interface device 304 is a connection device for connecting an operating device 311 and a display device 312, which are examples of user interface devices. Communication device 305 is a communication device for communicating with information processing devices 110 and 120 and server devices 130 and 150 via networks 160 and 170.

[0052] The drive device 306 is a device for setting the recording medium 313. The recording medium 313 here includes media for recording information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 313 may also include semiconductor memory such as ROM and flash memory for recording information electrically.

[0053] The various programs to be installed on the auxiliary storage device 303 are installed, for example, when the distributed recording medium 313 is set in the drive device 306 and the various programs recorded on the recording medium 313 are read by the drive device 306. Alternatively, the various programs to be installed on the auxiliary storage device 303 may be installed when they are downloaded from the network via the communication device 305.

[0054] Note that only the hardware configuration of the information processing unit 140 is described here, and the hardware configurations of the information processing units 110 and 120, and the server units 130 and 150 are omitted. However, the hardware configurations of the information processing units 110 and 120, and the server units 130 and 150 are basically the same as those of the information processing unit 140.

[0055] <Functional Configuration of Information Processing Devices> Next, the functional configuration of the information processing device 140 will be described. As mentioned above, the information processing device 140 has a database generation program installed, and when this program is executed, the information processing device 140 functions as an initial database generation unit 410 and a database generation unit 420.

[0056] The initial database generation unit 410 generates the initial database by collecting digitized arbitrary document data (error-free document data), adding information based on the analysis results obtained by analyzing the contents of the document data, and thereby generating document data for the database.

[0057] Specifically, the initial database generation unit 410 collects, for example, document data (error-free document data) managed by each department. The initial database generation unit 410 generates DB document data by adding information (technical field, technical content) based on the analysis results obtained by analyzing the content of the collected document data (error-free document data) to the collected document data. It is assumed that the collected document data already has attribute information attached as metadata. The analysis method used by the initial database generation unit 410 when analyzing the content of the document data is arbitrary, and for example, a generation AI such as tf-idf (term frequency-inverse document frequency) or a large-scale language model may be used. An example of a large-scale language model is ChatGPT.

[0058] The initial database generation unit 410 stores the generated DB document data in areas within the initial database that correspond to the attributes of the collected document data (e.g., product range).

[0059] As shown in Figure 4, the database generation unit 420 further includes an OCR document data acquisition unit 421, a search range identification unit 422, a search unit 423, a modification unit 424, and a storage unit 425.

[0060] The OCR document data acquisition unit 421 acquires the OCR document data transmitted from the information processing device 110 and the information processing device 120, and notifies the search range identification unit 422 and the correction unit 424.

[0061] The search range identification unit 422 is an example of an identification unit, and it analyzes the contents of the OCR document data notified by the OCR document data acquisition unit 421 and obtains the analysis results (e.g., technical field, technical content). The analysis method used by the search range identification unit 422 when analyzing the contents of the OCR document data is arbitrary, and for example, tf-idf, ChatGPT, etc., may be used.

[0062] The search range identification unit 422 identifies the search range based on the attributes of the OCR document data notified by the OCR document data acquisition unit 421 (e.g., department, product range) and the acquired analysis results (e.g., technical field, technical content), and notifies the search unit 423.

[0063] The search unit 423 searches for DB document data that corresponds to the OCR document data among the DB document data stored in the initial database or in each area within the database, within the search range notified by the search range specification unit 422. The search unit 423 performs the search for DB document data corresponding to the OCR document data on a page-by-page basis.

[0064] To search for database document data corresponding to OCR document data on a page-by-page basis, • Search for pages in the database document data that are similar to the target page in the OCR document data. • Searching for pages in the database document data that are similar to the summary of the target page in the OCR document data. • Search for pages in the database document data that contain keywords extracted from the target pages within the OCR document data. The data must include one of the following. For example, tf-idf, ChatGPT, etc., may be used to generate a summary of the target page or to extract keywords from the target page. The search method used by the search unit 423 is arbitrary; for example, vector search, fuzzy search, semantic search, etc., may be used.

[0065] The search unit 423 generates reference document data based on the pages of the retrieved DB document data. For example, if only one page of data from one DB document is retrieved, the search unit 423 uses the data from that one page as reference document data. For example, if one page of data is retrieved from each of multiple DB document data sets, the search unit 423 generates reference document data by combining the data from each of the top three DB document data sets with the highest similarity. The search unit 423 notifies the modification unit 424 of the generated reference document data.

[0066] The correction unit 424 generates prompts that designate the target page of the OCR document data notified by the OCR document data acquisition unit 421 as "document data to be corrected" and the reference document data notified by the search unit 423 as "document data to be referenced when correcting".

[0067] Furthermore, the "document data to be referenced when making corrections" may include not only reference document data, but also the pages before and after the target page of the OCR document data that is to be corrected. For example, if the target page of the OCR document data contains sentences that span across the preceding and succeeding pages, a generative AI such as a large-scale language model can make contextually appropriate corrections by referring to the preceding and succeeding pages.

[0068] The correction unit 424 uses the generated prompt to instruct the generation AI, such as the large-scale language model, of the server device 150 to correct the misrecognition result of the OCR document data.

[0069] As a result, the correction unit 424 acquires corrected OCR document data in which the misrecognition results of the OCR document data have been corrected. The correction unit 424 then notifies the storage unit 425 of the acquired corrected OCR document data.

[0070] The storage unit 425 generates DB document data by adding information (technical field, technical content) based on the analysis results notified by the search range identification unit 422 to the corrected OCR document data notified by the correction unit 424. The storage unit 425 stores the generated DB document data in the initial database or an area within the database, according to the attributes (e.g., product range) that have been pre-added to the corrected OCR document data.

[0071] <Details of processing by the initial database generation unit> Next, we will explain the details of the processing performed by the initial database generation unit 410.

[0072] (1) Flow of the initial database generation process by the initial database generation unit 410 First, we will explain the flow of the initial database generation process performed by the initial database generation unit 410. Figure 5 is an example of a flowchart showing the flow of the initial database generation process.

[0073] In step S501, the initial database generation unit 410 collects digitized document data.

[0074] In step S502, the initial database generation unit 410 analyzes the contents of the collected document data on a page-by-page basis.

[0075] In step S503, the initial database generation unit 410 adds information based on the analysis results (e.g., technical field, technical content) to the document data on a page-by-page basis to generate document data for the database.

[0076] In step S504, the initial database generation unit 410 stores the generated DB document data in an area within the initial database corresponding to the attributes of the collected document data.

[0077] In step S505, the initial database generation unit 410 determines whether or not there is the next document data. If it determines that there is the next document data (i.e., if the answer in step S505 is YES), it returns to step S501.

[0078] On the other hand, if it is determined in step S505 that there is no further document data (i.e., the answer is NO in step S505), the initial database generation process is terminated.

[0079] (2) Initial database generation screen Next, we will describe the initial database generation screen displayed on the display device 312 when the initial database generation unit 410 executes the initial database generation process. Figure 6 shows an example of the initial database generation screen.

[0080] As shown in Figure 6, the initial database generation screen 600 includes a document data read button 610, an analysis button 620, an edit button 622, and a register button 630.

[0081] The document data read button 610 is a button for reading document data used to generate the initial database. User 141 reads the document data used to generate the initial database by pressing the document data read button 610 and accessing the document data storage location. The read document data is displayed in the document data display area 611. The document data used to generate the initial database may also be read by dragging and dropping it onto the initial database generation screen 600.

[0082] The analysis button 620 is used to identify the attributes of the read document data and to analyze the content of the read document data. When the user 141 presses the analysis button 620, the attributes of the read document data are identified, the content of the read document data is analyzed, and information based on the attributes and analysis results is displayed in the attribute and analysis result display area 621.

[0083] The example in Figure 6 shows how the attributes of the retrieved document data are identified, resulting in the display of "Department 'Department α'" and "Product Range 'Product Range B'". It is assumed that the attributes of the retrieved document data are included as metadata within the retrieved document data itself.

[0084] The example in Figure 6 shows how, after the content of each page of the retrieved document data is analyzed, the information displayed in the document data display area 611, based on the analysis results for the page, displays "Technical field: Chemistry" and "Technical content: Alumina recovery".

[0085] As described above, since the analysis of the retrieved document data is performed on a page-by-page basis, the information based on the attributes and analysis results displayed in the attribute and analysis result display area 621 changes in accordance with the scrolling of the document data pages displayed in the document data display area 611.

[0086] The edit button 622 is used by user 141 to manually correct information based on attributes and analysis results displayed in the attribute and analysis results display area 621. User 141 views each page of the document data displayed in the document data display area 611. If user 141 determines that there is an error in the information based on attributes or analysis results displayed in the attribute and analysis results display area 621, user 141 presses the edit button 622 to make the information based on attributes and analysis results editable. This allows user 141 to arbitrarily correct the information based on attributes or analysis results.

[0087] The registration button 630 is used to generate DB document data by adding information based on the analysis results to the read document data, and to store it in the initial database. When user 141 presses the registration button 630, the DB document data is stored in the initial database, and the stored information is displayed in the stored information display area 631.

[0088] The example in Figure 6 shows that the retrieved document data was stored in the initial database as DB document data with the file name "Document Data B_1". The example in Figure 6 shows that the DB document data was stored in the area corresponding to product range "Product Range B" ("Data Storage Area for Product Range B") among the various areas in the initial database.

[0089] <Details of processing by the database generation unit> Next, we will explain the details of the processing performed by the database generation unit 420.

[0090] (1) Flow of database generation process by database generation unit 420 First, we will explain the flow of the database generation process performed by the database generation unit 420. Figure 7 is an example of a flowchart showing the flow of the database generation process.

[0091] In step S701, the database generation unit 420 collects OCR document data.

[0092] In step S702, the database generation unit 420 analyzes the contents of the collected OCR document data on a page-by-page basis.

[0093] In step S703, the database generation unit 420 determines the search range based on the attributes of the collected OCR document data and the analysis results.

[0094] In step S704, the database generation unit 420 searches for DB document data corresponding to each page of the OCR document data on a page-by-page basis from the DB document data included in the specified search range, either from the initial database or from each area within the database. Based on the pages of the searched DB document data, the database generation unit 420 generates reference document data.

[0095] In step S705, the database generation unit 420 generates prompts that include the "document data to be corrected" and the "document data to be referenced when correcting". More specifically, the correction unit 424 of the database generation unit 420 includes the rules for correcting typographical errors in the system prompt. The correction unit 424 of the database generation unit 420 includes the reference document data and the pages before and after the target page of the OCR document data in the assistant prompt as the "document data to be referenced when correcting". The correction unit 424 of the database generation unit 420 includes the target page of the OCR document data in the user prompt as the "document data to be corrected".

[0096] In step S706, the database generation unit 420 uses the generated prompt to instruct the generation AI, such as the large-scale language model, of the server device 150 to correct the misrecognition result for the target page of the OCR document data.

[0097] In step S707, the database generation unit 420 obtains corrected OCR document data in which the misrecognition results have been corrected for the target pages of the OCR document data. The database generation unit 420 performs the same process (steps S702 to S707) for all pages of the OCR document data. As a result, the database generation unit 420 obtains corrected OCR document data in which the misrecognition results have been corrected for all pages of the OCR document data.

[0098] In step S708, the database generation unit 420 generates DB document data by adding information based on the analysis results to each corrected OCR document data for which the misrecognition results have been corrected for all pages of the OCR document data.

[0099] In step S709, the database generation unit 420 stores the generated DB document data in the initial database or an area within the database, according to the attributes of the OCR document data.

[0100] In step S710, the database generation unit 420 determines whether or not there is the next OCR document data. If it is determined in step S710 that there is the next OCR document data (if the answer is YES in step S710), the process returns to step S701.

[0101] On the other hand, if it is determined in step S710 that there is no further OCR document data (i.e., the answer is NO in step S710), the database generation process is terminated.

[0102] (2) Database generation screen Next, we will describe the database generation screen displayed on the display device 312 when the database generation unit 420 executes the database generation process. Figure 8 shows an example of the database generation screen.

[0103] As shown in Figure 8, the database generation screen 800 includes an OCR document data read button 810, a DB document data search button 820, a modification instruction button 830, and a registration button 840.

[0104] The OCR document data read button 810 is a button for reading OCR document data used to generate the database. User 141 presses the OCR document data read button 810 and accesses the storage location of the OCR document data to read the OCR document data used to generate the database. The read OCR document data is displayed in the OCR document data display area 811. In addition, the file name of the read OCR document data, attributes (department, product range), and information based on the analysis results obtained by analyzing the contents of the read OCR document data (technical field, technical content) are displayed in the attribute and analysis result display area 812. The OCR data used to generate the database may also be read by dragging and dropping it onto the database generation screen 800.

[0105] As described above, the analysis of the read OCR document data is performed on a page-by-page basis. Therefore, the "Technical Field" and "Technical Content" displayed in the attribute and analysis result display area 812 change according to the scrolling of the OCR document data displayed in the OCR document data display area 811.

[0106] The example in Figure 8 shows that the file name of the read OCR document data is "Document Data X", and the attributes of the read OCR document data are "Department α" and "Product Range B". Furthermore, the example in Figure 8 shows that, based on the analysis results of the content of the page displayed in the OCR document data display area 811 of the read OCR document data, the technical field is "Chemistry" and the technical content is "Alumina Recovery".

[0107] The DB document data search button 820 is a button for generating reference document data. When user 141 presses the DB document data search button 820, the corresponding pages of the DB document data for each page of the OCR document data are searched, and reference document data is generated by combining, for example, the top three pages of the DB document data with the highest similarity. As a result, the reference document data is displayed in the reference document data display area 821.

[0108] As described above, the search for database document data corresponding to the OCR document data is performed on a page-by-page basis, and reference document data is generated for each page of the OCR document data. Therefore, the reference document data displayed in the reference document data display area 821 scrolls in accordance with the scrolling of the pages of the OCR document data displayed in the OCR document data display area 811.

[0109] The correction instruction button 830 is used to instruct the system to correct the misrecognition results of the read OCR document data on a page-by-page basis. When user 141 presses the correction instruction button 830, a prompt is generated on a page-by-page basis, and the system instructs the generation AI, such as a large-scale language model, to correct the misrecognition results of the OCR document data on a page-by-page basis. Once corrections have been instructed for all pages, the corrected OCR document data display area 831 displays the corrected OCR document data for all pages.

[0110] The registration button 840 is used to register corrected OCR document data for all pages. When user 141 presses the registration button 840, information based on the analysis results (technical field, technical content) is added to each of the corrected OCR document data pages displayed in the corrected OCR document data display area 831. As a result, database document data is generated for all the corrected OCR document data pages displayed in the corrected OCR document data display area 831. The generated database document data is then stored in the database area corresponding to the attributes and the attributes (product range) displayed in the attributes and analysis result display area 812.

[0111] (3) Specific examples of prompts Next, we will describe specific examples of prompts (system prompt, assistant prompt, user prompt) generated by the database generation unit 420.

[0112] Figure 9A shows an example of a system prompt. As described above, the modification unit 424 of the database generation unit 420 writes the rules for correcting typographical errors as a system prompt. The example in Figure 9A shows how the modification unit 424 of the database generation unit 420 specifies the role and work content of the generation AI, such as a large-scale language model, as a system prompt 910, and also writes six rules for correcting typographical errors, from condition 1 to condition 6.

[0113] In the example shown in Figure 9A, condition 5 of the system prompt 910 includes a rule stating that line breaks within each page of the corrected OCR document data should occur at the locations where "symbol" is added. This rule is included because, when a generative AI such as a large-scale language model corrects the misrecognition results of OCR document data on a page-by-page basis, the line correspondence between the original and corrected OCR document data may become ambiguous. In other words, by including the line break positions within the page in the system prompt 910, the line correspondence between the original and corrected OCR document data can be clarified.

[0114] Figure 9B shows an example of an assistant prompt and a user prompt. As described above, the modification unit 424 of the database generation unit 420 displays "document data to be referenced when making modifications" as an assistant prompt. The modification unit 424 of the database generation unit 420 also displays "document data to be modified" as a user prompt.

[0115] In the example in Figure 9B, Assistant Prompt 920 displays "Document data to refer to when making corrections," • Reference document data generated for the target pages of the OCR document data. • The page immediately preceding the target page of the OCR document data. • The page one page after the target page of the OCR document data. This shows how it was written.

[0116] In the example in Figure 9B, at user prompt 930, the "document data to be corrected" is: • Target pages of the OCR document data, This shows how it was written.

[0117] (4) Specific examples of various types of information used in processing by the database generation unit 420 Next, we will explain specific examples of the various types of information used in the database generation process by the database generation unit 420 (target pages of the OCR document data, summary of the target pages, list of search results, reference document data, and corrected OCR document data).

[0118] Figure 10A shows an example of a target page in OCR document data. In the example in Figure 10A, superimposed data 1010 is shown, in which the original handwritten document is overlaid on the OCR document data, so that the misrecognition results can be seen.

[0119] In the example in Figure 10A, "Akadoro" in the first and sixth lines is mistakenly identified as "Akazawa". Also in the example in Figure 10A, "Previously Reported" in the seventh line is mistakenly identified as "False Report". Furthermore, in the example in Figure 10A, "First Report" in the eighth line is mistakenly identified as "Forecast".

[0120] Figure 10B shows an example of a summary of the target page, a list of search results, and reference document data. In Figure 10B, the summary of the target page 1020 shows the summary obtained by analyzing the target page of the OCR document data shown in Figure 10A.

[0121] In Figure 10B, the list of search results 1030 shows the list of search results when a search is performed based on the summary of the target page 1020. As shown in Figure 10B, the list of search results 1030 is arranged in descending order of similarity, along with the similarity score, for DB document data pages that are similar to the summary of the target page of the OCR document data among the DB document data included in the search range.

[0122] In Figure 10B, the reference document data 1040 shows the reference document data generated by combining the top three pages with the highest similarity among the pages of the DB document data included in the search results list 1030.

[0123] Figure 10C shows an example of corrected OCR document data. In Figure 10C, in the corrected OCR document data 1050, "Akazawa" on the 1st and 6th lines has been corrected to "Akadoro", "false report" on the 7th line has been corrected to "previously reported", and "forecast" on the 8th line has been corrected to "first report".

[0124] Thus, according to the information processing device 140 of the first embodiment, even words that were previously difficult to correct by the generating AI, such as those registered in a dictionary, can be corrected by providing appropriate reference document data, allowing the generating AI to determine that they are misrecognized and make corrections.

[0125] <Summary> As is clear from the above description, the information processing apparatus 140 according to the first embodiment is Based on the analysis results obtained by analyzing the attributes and content of the OCR document data to be processed, the search range of the database is determined. • Search for database document data corresponding to the OCR document data to be processed within the specified search range. The system generates prompts designating the OCR document data to be processed as "document data to be corrected" and the reference document data generated based on the retrieved DB document data as "document data to be referenced when correcting." Using these prompts, the system instructs the generation AI, such as a large-scale language model, to correct the misrecognition results of the OCR document data to be processed. The database is updated by adding information based on the analysis results to the corrected OCR document data, which has had its misrecognition results corrected by a generative AI such as a large-scale language model, and storing it in an area within the database corresponding to the attributes of the OCR document data being processed.

[0126] Thus, in this embodiment, considering that in the case of generative AI such as large-scale language models, correcting misrecognition results may require information on specific technical fields, advanced expertise, etc., • Document data for the database containing information on specific technical fields, advanced expertise, etc., is prepared in advance. When correcting misrecognition results in OCR document data, the search range is narrowed based on attributes and analysis results. • Search for database document data that corresponds to the OCR document data within the narrowed search range. Based on the searched DB document data, reference document data is generated. • Provide the relevant reference document data to the generating AI.

[0127] This makes it possible to provide the generating AI with information suitable for correcting misrecognition results in OCR document data. As a result, the generating AI can properly correct the misrecognition results.

[0128] In other words, according to the first embodiment, it is possible to provide a technology for appropriately correcting document data using a generation AI and generating a database.

[0129] [Second Embodiment] In the first embodiment described above, when generating reference document data, the configuration was to combine the top three pages of DB document data with the highest similarity among the retrieved DB document data. However, the number of DB document data pages used for combining is not limited to three. Alternatively, the reference document data may be generated without combining, using the top one page.

[0130] In the first embodiment described above, it was explained that the assistant prompt should include one page before and one page after the target page of the OCR document data. However, the pages to be included in the assistant prompt are not limited to one page before and one page after the target page of the OCR document data. For example, one page from either the before or after page may be included, or multiple pages from both the before and after pages may be included.

[0131] In the first embodiment described above, the search range was determined based on the attributes of the OCR document data and the analysis results, but the method for determining the search range is not limited to this. The search range may be determined based on either the attributes of the OCR document data or the analysis results.

[0132] In the first embodiment described above, the OCR document data was processed on a page-by-page basis. However, the processing is not limited to page-by-page units, and may be performed within a range that results in an appropriate number of tokens. In this case, the assistant prompt may include appropriate ranges before and after the target range of the OCR document data.

[0133] In the first embodiment described above, the document data for the database is classified based on attributes when stored, but classification may also be based on other classification criteria besides attributes.

[0134] In the first embodiment described above, the attributes of the document data were exemplified as department (storage location for handwritten documents or OCR document data) and product range (product group handled by the department), but other attributes (for example, the creator of the handwritten document, the creator's affiliation, etc.) may also be included.

[0135] In the first embodiment described above, when analyzing the contents of the document data, the technical field and technical content were analyzed, but other analysis items (for example, technical name, method, background, etc.) may also be analyzed.

[0136] In the first embodiment described above, OCR document data was used as the processing target, but the processing target is not limited to OCR document data and may be other types of document data.

[0137] It should be noted that the present invention is not limited to the configurations shown in the above embodiments, including combinations with other elements. These aspects can be modified without departing from the spirit of the present invention and can be appropriately determined according to their application.

[0138] This application claims priority based on Japanese Patent Application No. 2024-192007, filed on 31 October 2024, which is incorporated herein by reference to the entire contents of the said Japanese Patent Application. [Explanation of symbols]

[0139] 100: Database generation system 110, 120: Information Processing Devices 130: Server device 131: Data storage unit for product range A 132: Data storage unit for product range B 133: Data storage unit for product range C 134: Data storage unit for product range D 140: Information Processing Device 410: Initial database generation unit 420: Database generation unit 421: OCR Document Data Acquisition Unit 422: Search range specification unit 423: Search section 424: Correction section 425: Storage Unit 600: Initial database generation screen 800: Database generation texts

Claims

1. A specific unit identifies the search range of the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed, A search unit searches for document data corresponding to the document data to be processed from the specified search range, A modification unit instructs the generating AI to modify the document data to be processed, using prompts that designate the document data to be processed as the document data to be modified, and the reference document data generated based on the retrieved document data as the document data to be referenced during modification. A storage unit updates the database by adding information based on the analysis results to the document data to be processed, which has been corrected by the generating AI, and storing it in an area within the database corresponding to the attributes of the document data to be processed. An information processing device having

2. The specified unit analyzes at least one of the following: technical field, technical content, technical name, method, and background of the document data to be processed. The information processing apparatus according to claim 1.

3. The attributes of the document data to be processed include, at a minimum, the department where the document data to be processed is stored, and the range of products handled by that department. The information processing apparatus according to claim 1.

4. The correction unit instructs the document data to be processed to be corrected on a page-by-page basis. The information processing apparatus according to claim 1.

5. When the modification unit instructs the generation AI to modify the page to be processed of the document data to be processed, it transmits the pages before and after the page to be processed to the generation AI. The information processing apparatus according to claim 4.

6. The system further includes an initial database generation unit that generates an initial database before the aforementioned database is updated, The initial database generation unit analyzes the contents of the document data to be stored in the initial database, adds information based on the analysis results, and stores it in an area corresponding to the attributes of the document data. The information processing apparatus according to claim 1.

7. A step of identifying the search range of the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed, The process of searching for document data corresponding to the document data to be processed from the specified search range, The process involves instructing the generating AI to modify the document data to be processed by using prompts that designate the document data to be processed as the document data to be modified, and the reference document data generated based on the retrieved document data as the document data to be used as reference during modification. The process involves adding information based on the analysis results to the document data to be processed, which has been modified by the generating AI, and storing it in an area within the database corresponding to the attributes of the document data to be processed, thereby updating the database. A method for generating a database that a computer executes.

8. A step of identifying the search range of the database based on the attributes of the document data to be processed or the analysis results obtained by analyzing the content of the document data to be processed, The process of searching for document data corresponding to the document data to be processed from the specified search range, The process involves instructing the generating AI to modify the document data to be processed by using prompts that designate the document data to be processed as the document data to be modified, and the reference document data generated based on the retrieved document data as the document data to be used as reference during modification. The process involves adding information based on the analysis results to the document data to be processed, which has been modified by the generating AI, and storing it in an area within the database corresponding to the attributes of the document data to be processed, thereby updating the database. A database generation program to run on a computer.

Images