Financial statement summary processing system, financial statement summary processing device, financial statement summary processing method, and computer program

Abstract

To cause a text generation AI to generate a financial statement document according to whether an announced financial statement summary corresponds to a quarterly financial statement or a full-year financial statement.SOLUTION: A financial statement summary processing device 20 prepares in advance a quarterly financial statement prompt for instructing generation of a financial statement document corresponding to a quarterly financial statement and a full-year financial statement prompt for instructing generation of a financial statement document corresponding to a full-year financial statement, and determines whether financial statement summary data distributed from a securities exchange system 2 corresponds to a quarterly financial statement or a full-year financial statement. The financial statement summary processing device 20 transmits, to a text generation AI system 5, a prompt corresponding to a result of the determination and text data converted from the distributed financial statement summary data, thereby causing the text generation AI system to generate a financial statement document corresponding to the quarterly financial statement or the full-year financial statement.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a financial statement processing system, financial statement processing device, financial statement processing method, and computer program that uses text generation AI (artificial intelligence) to generate text explaining the details of a company's financial results based on each company's financial statement summary distributed by a stock exchange, and in particular, that is capable of appropriately expressing content corresponding to quarterly or full-year financial results. [Background technology]

[0002] Stock exchanges have traditionally distributed data on each company's financial statements, and services have been offered that use this data to generate summaries of financial statements using AI (artificial intelligence).The data on financial statements distributed in this way comes in two formats: XBRL (eXtensible Business Reporting Language), which mainly contains numerical information related to the financial statements, and PDF (Portable Document Format), which includes explanatory text.These two types of data are sent as a set from stock exchanges.

[0003] The following non-patent documents 1 and 2 disclose the use of these two types of data to generate financial statements summaries using AI (artificial intelligence). Specifically, numerical information such as sales figures is read from XBRL-formatted data to generate statements about performance such as profits and sales figures. From PDF-formatted data, statements describing the factors behind the performance (performance factor statements) are extracted from statements explaining the performance overview contained in the data, and these statements are combined to generate a financial statement summary. A distinctive feature of financial statements generated in this way is that they can be completed in a much shorter time than when manually crafted.

[0004] The core technology used to extract performance factor statements from PDF data is the Language Understanding Laboratory's automatic text summarization and generation engine, which analyzes the case structure of the text, finds cause-and-effect sentence pairs, and classifies them into positive and negative sentences. Furthermore, the data is classified into performance factors and other sentences, and performance factor statements are selected based on whether the performance is good or bad and the positive or negative of the factor statements, an overview of overall performance, and prioritization of profitable business segments. [Prior art documents] [Patent documents]

[0005] [Non-Patent Document 1] Automatic Generation of Financial Statement Summaries, Masayuki Ishii, [online], [Retrieved April 24, 2024], Internet<URL:https: / / www.jstage.jst.go.jp / article / itej / 74 / 1 / 74_16 / _pdf / -char / ja> [Non-patent document 2] Fully automated financial statement summary by NIKKEI, [online], [searched on April 24, 2024], Internet<URL:https: / / www.nikkei.com / promotion / collaboration / qreports-ai / > Summary of the Invention [Problem to be solved by the invention]

[0006] The technical details disclosed in the above-mentioned Non-Patent Documents 1 and 2 generate financial statements summaries using XBRL format data, which primarily shows the numerical information in the financial statements, and PDF format data, which includes the general content of the financial statements. Since two types of data are processed, processing is required for each type of data, resulting in a problem of a large number of processes involved in generating the financial statements. Note that PDF format data corresponds to data in an electronic document file format internationally standardized by the International Organization for Standardization (ISO), and corresponds to financial statements data, which includes the general content of a company's quarterly or full-year financial statements.

[0007] Furthermore, there are two types of financial results summary: quarterly financial results summary, which are published every three months, and full-year financial results summary (final financial results summary), which cover the entire year. As these two types of financial results summary cover different periods, the circumstances and amount of information described in each financial results summary differ. However, the above-mentioned Non-Patent Documents 1 and 2 do not distinguish between these two types of financial results summary, and instead generate financial results summaries using the same process, which creates the problem of making it difficult to create expressions and content that are appropriate for both quarterly and full-year financial results summary.

[0008] Furthermore, PDF format data distributed by stock exchanges contains a lot of information related to corporate financial statements, making it generally difficult to summarize. In the above-mentioned non-patent documents 1 and 2, a financial statement summary is generated by extracting a statement describing the factors behind the performance (performance factor statement) from the explanatory statement of the performance overview contained in the PDF format data, which creates the problem that it is difficult to reflect information other than the explanatory statement of the performance overview contained in the PDF format data.

[0009] Furthermore, the technology used in the above-mentioned Non-Patent Documents 1 and 2 finds pairs of cause and effect sentences from the sentences contained in the financial results summary, classifies them into positive and negative sentences, and performs processing such as matching them with the positive and negative factor sentences.However, this processing is only performed with the aim of extracting sentences that describe the factors behind performance from the financial results summary, and there is a problem in that it does not reach a level where it can grasp the good and bad points of a company's performance, etc.

[0010] The present invention has been made in consideration of the above circumstances, and aims to provide a financial statement processing system, a financial statement processing device, a financial statement processing method, and a computer program that improve the efficiency of the text generation process by enabling a general-purpose AI (artificial intelligence) system that generates text to generate financial statement text from only data containing the general content of the financial statement (for example, data in PDF format) without using XBRL format data from the distributed financial statement data.

[0011] Another object of the present invention is to provide a financial statement processing system, financial statement processing device, financial statement processing method, and computer program that determine whether a distributed financial statement summary corresponds to a quarterly or full-year financial statement, and, based on the determination result, have AI generate text that corresponds to the type of financial statement summary distributed, thereby making it possible to generate text that is expressed in accordance with quarterly or full-year financial statements from data that includes the general content of a financial statement summary, which is generally difficult to summarize.

[0012] Furthermore, the present invention aims to provide a financial statement processing system, a financial statement processing device, a financial statement processing method, and a computer program that enable AI to generate financial statement documents that go beyond a simple summary and go into detail about the contents of the financial statement and the good and bad points of a company's performance, etc. [Means for solving the problem]

[0013] In order to solve the above problems, the present invention provides a financial statement processing system that generates corporate financial statements using an external text generation AI system based on financial statement data including the overall content of a company's quarterly financial statement or full-year financial statement distributed from a data distribution source, the system comprising: a financial statement database system that receives and stores the financial statement data distributed from the data distribution source; storage means that stores first instruction information that instructs the generation of corporate financial statements corresponding to quarterly financial statements and second instruction information that instructs the generation of corporate financial statements corresponding to full-year financial statements; means that detects that new financial statement data distributed from the data distribution source has been received by the financial statement database system; and means that detects the reception of new financial statement data by the financial statement database system. When the new financial statement summary data is received, the system is characterized by comprising: a discrimination means for discriminating whether the new financial statement summary data corresponds to a quarterly financial statement or a full-year financial statement; a means for reading from the storage means either first or second instruction information that instructs the generation of a corporate financial statement according to the result of the discrimination by the discrimination means; a conversion means for extracting the text portion from the new financial statement summary data and converting it into text data; a means for transmitting the instruction information read from the storage means and the text data converted by the conversion means to an external text generation AI system; and a means for receiving the corporate financial statement generated by the external document generation AI system from the external text generation AI system by transmitting the instruction information and text data.

[0014] The present invention is also characterized by the provision of a means for creating an explanation of a corporate financial statement summary by applying the contents of the received corporate financial statement document to format data that specifies a format appropriate to the financial statements when the document is received from an external document generation AI system. Furthermore, the present invention is characterized in that the first instruction information and the second instruction information include an instruction to generate a corporate financial statement document that lists the good and bad points from the contents of the financial statement summary.

[0015] The present invention relates to a financial statement processing device that processes an external text generation AI system to generate a company financial statement based on financial statement data including the general contents of a company's quarterly financial statement or full-year financial statement distributed from a data distribution source. The device comprises: a storage means for storing first instruction information instructing the generation of a company financial statement corresponding to a quarterly financial statement and second instruction information instructing the generation of a company financial statement corresponding to a full-year financial statement; a detection means for detecting the reception of new financial statement data in an external financial statement database system that receives and stores the financial statement data distributed from the data distribution source; and a processing means for detecting the reception of new financial statement data in the external financial statement database system when the detection means detects that new financial statement data has been received. The system is characterized by comprising a discrimination means for discriminating whether the financial statement data corresponds to a quarterly or full-year financial statement, a means for reading from the storage means either first or second instruction information that instructs the generation of a corporate financial statement according to the result of the discrimination by the discrimination means, a conversion means for extracting the text portion from the new financial statement data and converting it into text data, a means for transmitting the instruction information read from the storage means and the text data converted by the conversion means to an external text generation AI system, and a means for receiving the corporate financial statement generated by the external document generation AI system from the external text generation AI system by transmitting the instruction information and text data.

[0016] The present invention is also characterized by comprising a means for causing the detection means to start detection based on calendar information indicating the date and time of a company's financial results announcement. Furthermore, the present invention is characterized in that it comprises a means for storing volume information indicating the ranking of trading volume for each company's stock, and an order identification means for, when the detection means detects that new financial statement data for multiple companies has been received at an external financial statement database system, identifying the order of companies with the highest trading volume among the companies related to the new financial statement data whose reception has been detected, based on the ranking indicated by the volume information, and repeating the process of generating a corporate financial statement for each company in the order identified by the order identification means.

[0017] Furthermore, the present invention is characterized by comprising a means for storing access count information indicating the ranking of the number of accesses to information for each company's stock on a website that provides stock-related information, and an order identification means for, when the detection means detects that new financial statement data for multiple companies has been received at an external financial statement database system, identifying the order of companies with the most accesses among the companies related to the new financial statement data whose reception has been detected, based on the ranking indicated by the access count information, and repeating the process of generating a corporate financial statement for each company in the order identified by the order identification means.

[0018] The present invention is also characterized by comprising a format storage means for storing first format data specifying a format corresponding to quarterly financial statements and second format data specifying a format corresponding to full-year financial statements, and a means for reading out either the first or second format data from the format storage means according to the result of discrimination by the discrimination means, and by creating an explanation of the corporate financial statement summary by applying the contents of the corporate financial statement document to the format data read out from the format storage means.

[0019] The present invention relates to a financial statement processing method in which a financial statement processor causes an external text generation AI system to generate a corporate financial statement based on financial statement data including the general contents of a company's quarterly financial statement or full-year financial statement distributed from a data distribution source, and the financial statement processor stores first instruction information instructing the generation of a corporate financial statement corresponding to a quarterly financial statement and second instruction information instructing the generation of a corporate financial statement corresponding to a full-year financial statement, and the method includes the steps of detecting that new financial statement data has been received in an external financial statement database system that receives and stores the financial statement data distributed from the data distribution source, and detecting that new financial statement data has been received in the external financial statement database system. When detected, the system executes the following steps: determining whether the new financial statement data corresponds to a quarterly or full-year financial statement; reading from memory either first or second instruction information that instructs the generation of a corporate financial statement corresponding to the determination result; extracting the text portion from the new financial statement data and converting it into text data; transmitting the read instruction information and converted text data to an external text generation AI system; and receiving the corporate financial statement generated by the external document generation AI system from the external text generation AI system by transmitting the instruction information and text data.

[0020] The present invention provides a computer program for causing a computer that stores first instruction information instructing the generation of a company's financial statements in accordance with quarterly financial statements and second instruction information instructing the generation of a company's financial statements in accordance with full-year financial statements to perform processing for generating a company's financial statements using an external text generation AI system based on financial statement summary data including the general contents of a company's quarterly financial statements or full-year financial statements distributed from a data distribution source, the computer including a financial statement processing device that detects that new financial statement summary data has been received in an external financial statement database system that receives and stores the financial statement summary data distributed from the data distribution source, and a computer program for causing the computer to perform processing for generating a company's financial statements using an external text generation AI system based on financial statement summary data including the general contents of a company's quarterly financial statements or full-year financial statements distributed from a data distribution source, the computer including a financial statement processing device that detects that new financial statement summary data has been received in an external financial statement database system that receives and stores the financial statement summary data distributed from the data distribution source, the computer program including a step of detecting that new financial statement summary data has been received in the external financial statement database system, the ... When it is detected that the new financial statement summary data corresponds to a quarterly or full-year financial statement summary, the system executes the following steps: determining whether the new financial statement summary data corresponds to a quarterly or full-year financial statement summary; reading from memory either first or second instruction information that instructs the generation of a corporate financial statement corresponding to the result of the determination; extracting a text portion from the new financial statement summary data and converting it into text data; transmitting the read instruction information and the converted text data to an external text generation AI system; and receiving the corporate financial statement generated by the external document generation AI system from the external text generation AI system by transmitting the instruction information and text data.

[0021] In this invention, first instruction information instructing the generation of corporate financial statements corresponding to quarterly financial statements and second instruction information instructing the generation of corporate financial statements corresponding to full-year financial statements (final financial statements) are stored, the type of new financial statements distributed is determined, and instruction information corresponding to the determined type is sent to an AI (artificial intelligence) system to generate a statement, so that a statement with appropriate content corresponding to the type of newly distributed financial statements is generated. Also, in this invention, the distributed financial statement data in electronic document file format (equivalent to PDF format data, for example) is converted to text data and sent to the AI ​​system to be read, so that financial statements can be generated by the AI ​​system without using XBRL format data.

[0022] Furthermore, in the present invention, the AI ​​system uses format data that specifies the format according to the financial results, and applies the text generated by the AI ​​system to create the final explanation of the financial results summary, so that an explanation of the financial results summary organized in an easy-to-read layout can be provided to users such as investors.

[0023] Furthermore, in the present invention, the instruction information (first instruction information and second instruction information) sent to the AI ​​system includes instructions to generate a corporate financial statement that includes content listing the good and bad points from the contents of the financial statement summary (quarterly or full-year financial statement summary), making it possible to generate a statement that allows the good and bad points of the company's performance to be understood.

[0024] Furthermore, in the present invention, detection of new financial statements (financial statement data) is initiated based on calendar information indicating the financial statement announcement dates and times of multiple companies, thereby enabling the detection process of new financial statements to be carried out in a timely manner.

[0025] Furthermore, in the present invention, when many companies announce their financial statements at the same time, the order of the companies with the highest trading volume is identified for each company's stock, and the process for generating the company financial statements is carried out in accordance with the order of the companies identified, so that even during a period when financial statements are being announced one after another, the company financial statements can be generated in order of companies that are likely to attract the most investor interest.

[0026] Furthermore, in the present invention, the order of companies with the most frequent accesses is identified based on access count information indicating the number of accesses to information for each stock on a website that provides information on each company's stock, and processing related to generating corporate financial statements is performed according to the identified order of companies, so that even during times when many financial statements are being announced, corporate financial statements can be provided in order starting with companies that are expected to be of greatest interest to investors.

[0027] Furthermore, in the present invention, first format data specifying a format corresponding to quarterly financial statements and second format data specifying a format corresponding to full-year financial statements are prepared, and either the first or second format data is used depending on the result of determining the type of financial statement to create an explanation of the corporate financial statement summary. This makes it possible to obtain an explanation of the corporate financial statement summary in a format corresponding to the type of financial statement, thereby making it possible to provide a more understandable explanation to users such as general investors. [Effects of the Invention]

[0028] In this invention, the type of new financial results summary that has been distributed is identified, and instruction information corresponding to the identified type is sent to an AI (artificial intelligence) system to generate text, so that text with appropriate content can be generated according to the type of new financial results summary.In addition, the financial results summary data, including the general content of the financial results summary, is converted into text data and sent to a general-purpose text generation AI system for reading, so that appropriate corporate financial results text can be generated by a general-purpose AI system without using data in XBRL format.

[0029] Furthermore, in the present invention, the text generated by the text generation AI system is applied to format data including a format appropriate to the financial results, and an explanation of the company's financial results summary is finally created. This allows investors and other users to be provided with a well-laid, easy-to-read explanation of the financial results summary, which helps to expand the usefulness of the company's financial results text generated by the text generation AI system.

[0030] Furthermore, in the present invention, the instruction information (first instruction information and second instruction information) sent to the text generation AI system instructs the system to generate a corporate financial statement that includes content listing the good and bad points from the contents of the financial statement (quarterly or full-year financial statement), making it possible to automatically generate content that goes one step further than before, allowing users to understand the good and bad points of a company's performance, etc.

[0031] Furthermore, in the present invention, the detection process for new financial statement summary data is initiated based on calendar information showing the financial statement announcement dates and times of multiple companies, thereby enabling efficient system operation, such as flexibly changing the detection time to match periods with high or low numbers of financial statement announcements.

[0032] Furthermore, the present invention identifies the order of companies with the highest trading volume for each company's stock, and performs the process of generating corporate financial statements in accordance with that order. Therefore, even if many companies announce their financial statements at the same time, corporate financial statements can be obtained in order from companies that are likely to be of greatest interest to users such as investors.

[0033] Furthermore, the present invention identifies the order of companies related to stocks with the most accesses to information about each stock on a website that provides information about each company's stock, and performs processing related to generating corporate financial statements according to the order of the companies identified.Therefore, even during times when many companies are announcing their financial statements, corporate financial statements can be provided from companies that are likely to be of great interest to users such as investors.

[0034] Furthermore, in the present invention, an explanation of a corporate financial statement summary is created using either the first or second format data according to the result of determining the type of financial statement, so that an explanation of a corporate financial statement summary in an easy-to-understand format according to the type of financial statement can be provided. [Brief explanation of the drawings]

[0035] [Figure 1] 1 is a schematic diagram illustrating the overall configuration of a financial statement providing system including a financial statement processing system according to an embodiment of the present invention. [Figure 2] 1 is a block diagram showing an outline of the overall internal configuration of a settlement DB system. [Figure 3] 10A is a schematic diagram showing a part of an example of the contents of the PDF data of a quarterly financial statement summary, and FIG. 10B is a schematic diagram showing a part of an example of the contents of the PDF data of a full-year financial statement summary. [Figure 4] 2 is a block diagram showing the main internal configuration of a financial statement summary processing device. FIG. [Figure 5]FIG. 10 is a schematic diagram partially illustrating an example of the contents of a quarterly closing prompt. [Figure 6] FIG. 10 is a schematic diagram partially illustrating an example of the contents of a full-year settlement prompt. [Figure 7] FIG. 10 is a schematic diagram showing a part of an example of the contents of format data. [Figure 8] 1 is a first flowchart showing a series of processing steps related to a financial statement processing method. [Figure 9] FIG. 10 is a schematic diagram showing a portion of an example of the contents of a generated quarterly financial statement document. [Figure 10] 10 is a schematic diagram showing a portion of an example of the contents of a generated annual financial statement. FIG. [Figure 11] This is a schematic diagram showing an example of the contents of the range including the first half of the created quarterly financial results summary commentary. [Figure 12] This is a schematic diagram showing an example of the contents of the range including the latter half of the created quarterly financial results summary commentary. [Figure 13] This is a schematic diagram showing an example of the contents of the range including the first half of the created full-year financial results summary commentary. [Figure 14] This is a schematic diagram showing an example of the contents of the range including the latter half of the created full-year financial results summary commentary. [Figure 15] 10A is a schematic diagram showing a modified example in which a detection execution program and calendar information are stored in a storage unit, and FIG. 10B is a schematic diagram showing an example of the contents of the calendar information. [Figure 16] 10 is a second flowchart showing a series of processing steps relating to the execution of the detection process in the financial statement summary processing method. [Figure 17] 10A is a schematic diagram showing a modified example in which production volume information is stored in the storage unit, and FIG. 10B is a schematic diagram showing a modified example in which access count information is stored in the storage unit. DETAILED DESCRIPTION OF THE INVENTION [Example]

[0036] 1 is a schematic diagram showing the overall configuration of a financial statement provision system 1 including a financial statement processing system 10 according to the present invention. The financial statement provision system 1 includes a stock exchange system 2, a text generation AI system 5, and a financial statement processing system 10, and the financial statement processing system 10 causes the text generation AI system 5 to generate a corporate financial statement text based on data related to each company's financial statement distributed by the stock exchange system 2.

[0037] The data related to each company's financial results summary distributed from the securities exchange system 2 is a set of data in two file formats: the first is financial results summary data in a predetermined standardized file format (e.g., XBRL: eXtensible Business Reporting Language), and the second is data in an electronic text file format (e.g., PDF: Portable Document Format). The first, XBRL data (data in standardized file format), is data with limited content that mainly shows the numerical information in the financial results summary released by each company, while the second, PDF data (data in electronic text file format), contains the overall content of the financial results summary released by each company. Note that PDF data has the advantage that it can be displayed in basically the same layout regardless of the device used.

[0038] In addition, the types of financial results summary data distributed from the stock exchange system 2 mainly include those corresponding to quarterly financial results summary that explain the contents of a company's quarterly financial results (quarterly financial results summary data) and those corresponding to full-year financial results summary related to the full-year financial results (final financial results) that explain the contents of a company's full-year financial results (full-year financial results summary data).Therefore, for each of these quarterly financial results summary data and full-year financial results summary data, the above-mentioned XBRL data and PDF data are distributed as a set.

[0039] The financial statement summary processing system 10 is composed of a financial statement DB (database) system 15 and a financial statement summary processing device 20. The financial statement DB system 15 receives and stores financial statement summary data (quarterly financial statement summary data, full-year financial statement summary data, etc.) distributed by the stock exchange. Note that a cloud system database can also be used as the financial statement DB system 15, and when a cloud system is used in this way, it will be connected to the financial statement summary processing device 20 via a network so that data can be sent and received. Furthermore, the stock exchange system 2 and the sentence generation AI system 5 are external systems to the financial statement summary processing system 10.

[0040] The sentence generation AI system 5 used in this embodiment is a general-purpose (open) artificial intelligence server system for sentence generation, which generates sentences based on instruction information (prompts) sent from an external source. In this embodiment, the sentence generation AI system 5 receives instruction information and text related to a company's financial results summary, and generates a company's financial results sentence according to the instruction information. The sentence generation AI system 5 is configured to connect to the outside via an API (Application Programming Interface), and in order to perform access control related to communication connections, an API key is issued in advance and sent to the external access side.

[0041] In the financial statement summary providing system 1 having the above-described general configuration, financial statement summary data distributed from the stock exchange system 2 as appropriate is received by the financial statement DB system 15, and the receipt of such financial statement summary data is monitored (detected) by the financial statement summary processing device 20. The financial statement summary processing device 20 is triggered by the receipt of financial statement summary data by the financial statement DB system 15 to start processing related to the financial statement and cause the external text generation AI system 5 to generate a corporate financial statement text. The present invention will be described in detail below, focusing on the financial statement DB system 15 and the financial statement summary processing device 20 that constitute the financial statement summary processing system 10.

[0042] 2 shows an outline of the internal configuration of the settlement DB system 15, which includes a DB management device 16 and a DB device 18. The DB management device 16 controls the process of accumulating data in the DB device 18, and includes a control unit 16a, a communication unit 16b, and a memory unit 16c.

[0043] The control unit 16a of the DB management device 16 controls the overall system, including the control of the sending and receiving processes of data and information in the communication unit 16b, the storage process of data and information in the memory unit 16c, and the storage process of data in the DB device 18.

[0044] In this embodiment, when financial statement summary data is distributed from the stock exchange system 2, which is the data distribution source, the control unit 16a performs control processing so that the communication unit 16b receives (acquires) the data. The financial statement summary data distributed from the stock exchange system 2 is accompanied by financial statement summary information, which includes information such as a company identification code and a summary tag. The company identification code included in the financial statement summary information is a code that identifies which company the distributed financial statement summary data belongs to, and the summary tag indicates whether the distributed financial statement summary data is a quarterly financial statement summary or a full-year financial statement summary.

[0045] When communication unit 16b receives the financial statement data, control unit 16a performs processing to store in memory unit 16c a message indicating that the financial statement data has been received and the financial statement information attached to the received financial statement data, linking the message to the date and time of reception, and also passes the received financial statement data itself to DB device 18, where it is stored. As described above, financial statement data consists of a set of XBRL format data (XBRL data) and PDF format data (PDF data), and therefore, financial statement DB system 15 receives this set of financial statement data and stores and accumulates the set of financial statement data in DB device 18. Furthermore, memory unit 16c is provided with update information table 17, and the message indicating that the financial statement data has been received, the financial statement information attached to the financial statement data, etc. are linked to the date and time of reception and stored in update information table 17.

[0046] The control unit 16a performs the above-mentioned processing each time financial statement summary data is received by the communication unit 16b, and therefore the update information table 17 in the memory unit 16c stores financial statement summary information etc. relating to the received financial statement summary data together with the date and time of reception, and the received financial statement summary data itself is stored in the DB device 18, so that the DB device 18 sequentially stores the financial statement summary data of each company that has been distributed up to now (for example, Company A's quarterly financial statement summary data and full-year financial statement summary data, Company B's quarterly financial statement summary data and full-year financial statement summary data, etc.).

[0047] Figure 3(a) shows part of the contents of quarterly financial results summary PDF data 3a as an example of financial results summary data (PDF data) that includes the general content of the financial results summary, among the sets of financial results summary data distributed from the securities exchange system 2 and accumulated in the financial results DB system 15 as described above. This quarterly financial results summary PDF data 3a contains X Company's third quarter financial results summary for the fiscal year ending March 2024, and as shown in Figure 3(a), there is a paragraph titled "Consolidated results for the third quarter of the fiscal year ending March 2024," and this paragraph contains tables and the like showing various numerical information under the heading "Consolidated operating results."

[0048] Additionally, outside the scope of Figure 3(a), this quarterly financial results PDF data 3a includes tables showing various numerical information for paragraphs such as "Dividend Status" and "Consolidated Earnings Forecast for the Fiscal Year Ending March 2024," as well as notes for items such as "Changes in Significant Subsidiaries" and "Changes in Accounting Policies and Accounting Estimates." Furthermore, this quarterly financial results PDF data 3a also includes attached documents such as a summary of the quarterly consolidated financial statements, resulting in a volume of information spanning several pages. Note that the content of the quarterly financial results PDF data 3 shown in Figure 3(a) is an example of Company X's third-quarter financial results for the fiscal year ending March 2024. Even within the same Company X's financial results, the content may differ at different times, and different companies' financial results may also have different paragraphs and items (the same applies to full-year financial results).

[0049] Figure 3(b) shows part of the contents of full-year financial results summary PDF data 3b, as another example of PDF data among the financial results summary data distributed from the stock exchange system 2 and accumulated in the financial results DB system 15. This full-year financial results summary PDF data 3b contains X Company's financial results summary for the fiscal year ending March 2023, and as shown in Figure 3(b), it contains a paragraph titled "Consolidated results for the fiscal year ending March 2024," in which a table showing various numerical information is placed under the heading "Consolidated operating results."

[0050] Furthermore, outside the scope shown in Figure 3(b), this Annual Financial Results Summary PDF Data 3b, like the above-mentioned Quarterly Financial Results Summary PDF Data 3a, also includes tables showing various numerical information corresponding to paragraphs such as "Dividend Status" and "Consolidated Earnings Forecast for the Fiscal Year Ending March 2024," as well as notes containing information corresponding to items such as "Changes in Significant Subsidiaries" and "Changes in Accounting Policies and Accounting Estimates." Furthermore, this Annual Financial Results Summary PDF Data 3b also includes consolidated financial statements and other documents as attachments, resulting in a volume of information spanning several pages.

[0051] 4 is a block diagram showing the main internal configuration of the financial statement summary processing device 20, which plays a central role in the present invention. The financial statement summary processing device 20 of this embodiment is constructed using a general server computer (server device), but it is of course also possible to construct a device system by combining multiple server devices, etc., through distributed processing, etc. When new financial statement summary data is received by the above-mentioned financial statement DB system 15, the financial statement summary processing device 20 begins executing various processes related to text generation.

[0052] The settlement summary processing device 20 is configured by connecting various devices etc. to an MPU 20a that performs various overall control processing via internal connection lines 20i, and the various devices etc. include a communication module 20b, RAM 20c, ROM 20d, input interface 20e, output interface 20f, DB memory unit 20g, and memory unit 20h.

[0053] The communication module 20b is a communication device equivalent to a connection module with a network, and conforms to a required communication standard (for example, a LAN module). The communication module 20b is connected to a network (including internal, external, or other communication networks) via a required communication device (not shown, such as a router), and enables communication with the settlement DB system 15 (for the settlement summary message processing device 20, the settlement DB system 15 corresponds to an external settlement database system), etc.

[0054] The RAM 20c temporarily stores the contents associated with the processing of the MPU 20a, various files, information, etc., and the ROM 20d stores programs, etc. that define the basic processing contents of the MPU 20a. The input interface 20e is connected to a keyboard 21, mouse, etc. that accepts operation instructions, etc. from an administrator, etc. of the financial statement summary processing device 20. The output interface 20f is connected to a display 22 (display output device), and outputs the contents associated with the processing of the MPU 20a to the display 22, allowing the administrator, etc. of the financial statement summary processing device 20 to check the current processing contents, etc.

[0055] The DB storage unit 20g stores the corporate financial statement commentary that is finally created through the processing described below. Such corporate financial statement commentary is created each time new financial statement data is distributed from the securities exchange system 2, so a large number of corporate financial statement commentaries will be accumulated in the DB storage unit 20g. Note that the DB storage unit 20g is provided with a folder for each company that issues a financial statement, and within each folder for each company, a folder for each accounting period is provided.

[0056] The memory unit 20h is a storage means and stores various programs, information, etc. The programs stored in the memory unit 20h include an OS program P1 and a processing program P2, and the information stored in the memory unit 20h includes an API key 24, a quarterly settlement prompt 25, a full-year settlement prompt 26, format data 27, and error criteria 28. Note that in order to install the processing program P2 in the memory unit 20h of the financial statement summary processing device 20, it is conceivable to store the processing program P2 in a storage medium such as an optical disk and install it in the memory unit 20h via the storage medium.

[0057] Among the programs stored in the memory unit 20h, the OS program P1 specifies various processes according to the operating system for the server computer, and the MPU 20a performs processes based on these specifications, allowing the settlement summary processing device 20 to perform various functions as a server computer (server device).

[0058] Furthermore, the processing program P2 stored in the storage unit 20h defines various processes required for the MPU 20a to cause the external text generation AI system 5 to generate a corporate financial statement text based on the financial statement summary data, and ultimately to create a financial statement commentary. In accordance with the provisions of this processing program P2, the MPU 20a functions as a means for detecting receipt of financial statement summary data in the financial statement DB system 15 and determining whether the detected financial statement summary data is a quarterly or full-year financial statement summary. Details of this processing program P2 will be explained based on the first flowchart shown in Figure 8 below, but before that, the API key 24, quarterly financial statement prompt 25, full-year financial statement prompt 26, format data 27, etc. will be explained.

[0059] The API key 24 is used when the financial statement summary processing device 20 accesses the external text generation AI system 5, and is issued in advance for the financial statement summary processing device 20 by the external text generation AI system 5. This issued API key 24 is sent from the text generation AI system 5 to the financial statement summary processing device 20, and the sent API key 24 is stored in the memory unit 20h.

[0060] 5 shows a portion of the contents of an example of the quarterly financial statement prompt 25 stored in the storage unit 20h. The quarterly financial statement prompt 25 corresponds to instruction information (first instruction information) that instructs the text generation AI system 5 to generate a corporate financial statement corresponding to the quarterly financial statement, and includes instructions to generate a corporate financial statement that describes the good and bad points from the contents of the quarterly financial statement summary.

[0061] In this quarterly financial results prompt 25, the specific instruction information to the text generation AI system 5 first includes, as its contents, content that conveys the purpose of text generation, such as "summarize the attached quarterly financial results summary in PDF format so that general investors can understand it." By conveying such a purpose, the quarterly financial results prompt 25 of this embodiment requests the text generation AI system 5 to generate text at the level of a general investor, rather than at the level of an expert, as a prerequisite for text generation.

[0062] In addition, the quarterly financial results prompt 25 includes instruction information regarding a command to the sentence generation AI system 5, such as "make evaluation comments from a 'buying perspective' and a 'selling perspective' according to the format based on the financial information in the quarterly financial results summary."

[0063] This command instructs the text generation AI system 5 to generate comments evaluating the stock of the company that is the subject of the quarterly financial results summary from a buying and selling standpoint. This command causes the text generation AI system 5 to generate text that provides information that is directly relevant to general investors in determining whether they should buy or sell the stock of the company that is the subject of the quarterly financial results summary.

[0064] The command content described above is merely an example, and it is of course possible to include other command content, and furthermore, the command content may include not only one but multiple command content.

[0065] Other examples of instructions include, "Extract the specified information, notes, etc. from the attached file, and evaluate and summarize them for each paragraph," and "If the required explanation is for each business, evaluate and summarize them separately for each business."

[0066] Furthermore, the quarterly financial statement prompt 25 of this embodiment also includes content that specifies the output format, etc., and the output format instructs that content such as "Overview of current period's business results, etc." and "Future outlook (not shown in Figure 5)" be divided into paragraphs, and that items in those paragraphs be listed and content corresponding to each paragraph and item be generated.

[0067] Specifically, the paragraph "Overview of business results for the current fiscal year" includes instructions to summarize the positive and negative points of the "Business results (not shown in Figure 5)" and "Financial position" items in the quarterly financial results report in a way that is easy for general investors to understand. Quarterly Financial Results Prompt 25 also includes instructions to summarize the positive and negative points of the "Cash flow (not shown in Figure 5)" item included in the paragraph "Overview of business results for the current fiscal year" in a way that is easy for general investors to understand.

[0068] Furthermore, the quarterly financial results prompt 25 also includes sections such as "Earnings forecast," "Notes forecast," and "Buy / Sell evaluation" in the "Future Outlook" paragraph, which is not shown in Figure 5, and similarly provides instructions to write down the pros and cons. By including instructions to write down the pros and cons from the content of the quarterly financial results summary, the quarterly financial results prompt 25 can generate text with more in-depth content than conventional techniques, providing users such as retail investors with specific information for making investment decisions. When writing down the pros and cons, it is also possible to include instructions such as focusing on non-numeric characters, imposing a character limit on the generated content (e.g., 500-2000 characters), and indicating the perspective of the evaluation (e.g., the perspective of an economic commentator, a market participant, an analyst, etc.).

[0069] In addition, the quarterly financial statement prompt 25 may have a paragraph called "Dividends," which may include items such as "Dividend Policy," "Current Dividend," and "Next Dividend," and may provide instructions to write explanations and evaluations for these items.

[0070] Furthermore, the quarterly financial statement prompt 25 may have a paragraph titled "Consolidated financial statements and major notes," and within that paragraph, a section titled "Notes to consolidated financial statements," with instructions to write down the good and bad points for that section as well, as described above. Furthermore, within that paragraph, the prompt may have sections related to notes, sections related to various changes and fluctuations, sections related to various information, etc. By including such instruction information as described above, the quarterly financial statement prompt 25 ensures that the text generated by the text generation AI system 5 contains easy-to-understand information that will be helpful in determining whether to buy or sell the stock of the company (Company X) in this quarterly financial statement summary.

[0071] In addition, the quarterly financial statement prompt 25 of this embodiment preferably instructs the "tone (way of writing)" of the text to be generated, and also includes information to instruct the text generation AI system 5 as "rules" for the text to be generated, such as "output the text in the specified format, and if the quarterly financial statement to be processed does not contain any content that matches the format, write 'There are no relevant comments.'"

[0072] In this way, the Quarterly Financial Statement Prompt 25 instructs the creation of a corporate financial statement corresponding to the company's quarterly financial statement summary, and in particular instructs the user to evaluate and explain the good and bad points of various items included in paragraphs such as the current period's operating results for the quarterly period and future prospects. Therefore, the contents of the table containing the figures published in the original quarterly financial statement that was distributed are explained in digestible terms, explaining the good and bad points, and indicating the direction for creating something that is easy for users such as general investors to understand.

[0073] 6 shows a portion of the contents of an example of the annual settlement prompt 26 stored in the storage unit 20h. The annual settlement prompt 26 corresponds to instruction information (second instruction information) that instructs the text generation AI system 5 to generate a corporate settlement document corresponding to the annual settlement (final settlement) for one year, and includes instructions to generate a corporate settlement document that describes the good and bad points from the contents of the annual settlement summary.

[0074] The specific instruction information for the text generation AI system 5 includes, as with the quarterly financial results prompt 25 described above, firstly a statement conveying the purpose of text generation, such as "summarize the attached full-year financial results summary in PDF format so that it can be understood by general investors."

[0075] In addition, the annual financial statement prompt 26 includes instruction information regarding commands to the text generation AI system 5, such as "based on the financial information in the annual financial statement summary, provide evaluation and comment on both a "buying perspective" and a "selling perspective" according to the format."

[0076] The above command instructs the text generation AI system 5 to generate comments evaluating the stocks of the company that are the subject of the annual financial results summary from a buying and selling standpoint, thereby enabling the text generation AI system 5 to generate text that will provide general investors with information that will directly help them decide whether they should buy or sell the stocks of the company that is the subject of the annual financial results summary.

[0077] Furthermore, the full-year settlement prompt 26 of this embodiment also includes content for specifying a format, etc., similar to the above-mentioned quarterly settlement prompt 25. Specifically, the full-year settlement prompt 26 includes content in accordance with the full-year settlement summary, instructing that the good and bad points of items such as "Financial condition" in the paragraph "Overview of business results for the current fiscal year" be summarized in an easy-to-understand manner for general investors, and similarly includes content instructing that the good and bad points of each item in the paragraph "Future outlook" be summarized in an easy-to-understand manner for general investors.

[0078] In this way, the full-year financial results prompt 26 includes instructions to write and explain the good and bad points for each item in the "Overview of business results for the current fiscal year" and "Future outlook." This allows the text generation AI system 5 to obtain a concise explanation of the information other than the numbers contained in the full-year financial results summary in the original PDF data, dividing it into good and bad points.

[0079] In addition, the full-year financial statement prompt 26 may have a paragraph called "Dividends" in the area not shown in Figure 6, and may include items such as "Dividend Policy," "Current Dividend," and "Next Dividend" in that paragraph, and may provide instructions to write explanations and evaluations for those items.

[0080] Furthermore, to the extent not shown in Figure 6, the full-year financial statement prompt 26 may include a paragraph titled "Consolidated financial statements and major notes," and within that paragraph, an item titled "Notes regarding consolidated financial statements," with instructions to note the pros and cons of that item as well, similar to those above.

[0081] Furthermore, the Annual Financial Statement Prompt 26 may include items related to notes, items related to various changes and fluctuations, items related to various information, etc. In this way, by including instructions to write down the good and bad points of the Annual Financial Statement Prompt 26, investors and other users can obtain text that clearly expresses the good and bad points of the Annual Financial Statement Prompt as information that is useful for making investments.

[0082] The annual financial statement prompt 26 includes the instruction information described above, and thereby instructs the generation of a corporate financial statement corresponding to the annual financial statement. As a result, the text generated by the text generation AI system 5 includes content (information at a level that is easy for average investors to understand) that will be useful in determining whether to buy or sell the stock of the company (Company X) in the annual financial statement summary.

[0083] In addition, in the full-year financial statement prompt 26 of this embodiment, at the end, instructions are given regarding rules regarding the "tone of voice (how to write)" and writing, etc., similar to the above-mentioned quarterly financial statement prompt 25, so as to create a sense of unity in the sentences generated by the sentence generation AI system 5.

[0084] 7 shows part of the contents of an example of format data 27 stored in storage unit 20h. Format data 27 specifies a format according to the financial statements used when financial statement summary processing device 20 creates an explanation of a corporate financial statement summary, and lists the names of the paragraphs and items in the explanation, but the contents according to the explanation are left blank (blank areas 27a, 27b, 27c, etc., indicated by dashed rectangles in FIG. 7).

[0085] Examples of paragraph titles in format data 27 include "Overview of Current Fiscal Year Business Results," "Future Outlook," "Dividends," and "Consolidated Financial Statements and Key Notes." Examples of item titles include "Business Results," "Financial Condition," "Cash Flow," "Earnings Forecast," "Mid-Term Plan Forecast," "AI Buy / Sell Evaluation," "Dividend Policy," "Current Fiscal Year Dividends," "Next Fiscal Year Dividends," and "Notes to Consolidated Financial Statements." All of these items, except for the item related to dividends, include "Good Points" and "Bad Points." Furthermore, format data 27 is designed to allow text (text data) to be pasted into blank spaces 27a, etc. The paragraph and item names described above are merely examples, and other paragraph and item division methods are of course also applicable.

[0086] In addition, the error criteria 28 stored in the memory unit 20h shown in Figure 4 specify the criteria (conditions) that the MPU 20a of the financial statement summary processing device 20 uses to determine whether there is a problem (error) in the format of the content of the text (corporate financial statement text) generated by the text generation AI system 5.

[0087] The error criteria 28 stores information on a plurality of criteria conditions for determining an error. Examples of the criteria conditions include when the AI ​​does not generate a sentence for a paragraph, item, etc. (no sentence is generated, leaving a blank space), when a sentence is generated for a paragraph, item, etc. but the number of characters is extremely small (e.g., 8 characters or less), when more than half of the characters are a mixture of English letters and numbers (when the sentence is not valid as Japanese), when the number of characters is below the minimum limit or exceeds the maximum limit specified in the prompt, etc. Next, the processing program P2 stored in the memory unit 20h will be described in detail.

[0088] This processing program P2 is triggered by the receipt of new financial statement summary data by the financial statement DB system 15, and starts various processes related to text generation, and prescribes various processes such as reading the API key 24 from the storage unit 20h, reading a prompt (quarterly financial statement prompt 25 or full-year financial statement prompt 26) to issue instructions necessary for text generation to the text generation API server 5, and reading (obtaining) the financial statement summary data received by the financial statement DB system 15 and converting it into text. The MPU 20a executes various processes as various means in accordance with the provisions of this processing program P2.

[0089] Figure 8 is a first flowchart illustrating the control processes performed by the MPU 20a of the financial statement summary processing device 20 as defined by the processing program P2, and the processing procedures shown in this first flowchart correspond to the financial statement summary processing method. Below, the processing contents of the MPU 20a as defined by the processing program P2 (the contents of the financial statement summary processing method) will be explained in accordance with the first flowchart in Figure 8.

[0090] First, in accordance with the provisions of processing program P2, MPU 20a of the financial statement summary processing device 20 monitors the status of the financial statement DB system 15 as detection means and detects that new financial statement summary data distributed from the securities exchange system 2 has been received by the financial statement DB system 15 (DB management device 16) (S1). This detection is performed by the financial statement summary processing device 20 (MPU 20a) checking the update information table 17 stored in the memory unit 16c of the DB management device 16, and if the contents of the update information table 17 are updated and the reception of new financial statement summary data is recorded, MPU 20a of the financial statement summary processing device 20 detects whether financial statement summary data has been received based on whether such update content has been recorded.

[0091] If the MPU 20a has not detected that the settlement DB system 15 has received the settlement summary data (S1: NO), the MPU 20a waits for reception detection, but if the MPU 20a detects that new settlement summary data has been received by the settlement DB system 15 (S1: YES), the MPU 20a reads the API key 24 from the storage unit 20h (S2). The API key 24 thus read is temporarily stored (memorized) in the RAM 20c (memory).

[0092] In addition, when the reception of new financial statement data is detected (S1: YES), it is also detected whether the number of new financial statement data received is one or multiple, and the detected number is also stored in RAM 20c (memory) under the control of MPU 20a.

[0093] Next, the MPU 20a, as a determination means, determines whether the received new financial statement data corresponds to quarterly settlement or full-year settlement (S3). This determination process is also performed by the MPU 20a by checking the update information table 17 in the storage unit 16c of the DB management device 16. In the update information table 17, information related to a flag (a flag indicating whether the data corresponds to quarterly settlement or full-year settlement) attached to the received new financial statement data is also stored as update content, so the MPU 20a makes the determination based on the information related to the flag.

[0094] If it is determined as a result of the determination that the new financial statement data corresponds to quarterly settlement (S3: quarterly settlement), the MPU 20a reads out from the storage unit 20h a quarterly settlement prompt 25 that corresponds to first instruction information that corresponds to the determination result (S4). On the other hand, if it is determined as a result of the determination that the new financial statement data corresponds to full-year settlement (S3: full-year settlement), the MPU 20a reads out from the storage unit 20h a full-year settlement prompt 26 that corresponds to second instruction information that corresponds to the determination result (S5). The read prompt (quarterly settlement prompt 25 or full-year settlement prompt 26) is temporarily stored (memorized) in the RAM 20c (memory).

[0095] Then, the MPU 20a reads (acquires) the new financial statement data whose reception was detected at the S1 stage from the data stored in the financial statement DB system 15, and performs a process of converting the new financial statement data (PDF data) into text (text data) as a conversion means (S6).

[0096] As mentioned above, PDF data has the characteristic of being able to be displayed in basically the same layout regardless of the device used, but to ensure this characteristic, it also contains tabs (control characters) and other characters that define the layout, etc. Such tabs (control characters) and other characters can become noise when the text generation AI system 5 reads the data and generates text, which can reduce the accuracy of the generated text. Therefore, in step S6, in order to remove such unnecessary tabs (control characters), the text portion of the PDF data is extracted from the new financial summary data (PDF data) and converted into text data. The converted text data is also temporarily stored (memorized) in RAM 20c (memory).

[0097] Then, as a transmission means, the MPU 20a uses the API key 24 read at step S2 to access the sentence generation AI system 5 via the API (Application Programming Interface), and once a connection is established, it transmits the instruction information stored in the RAM 20c, that is, the prompt (quarterly financial statement prompt 25 or annual financial statement prompt 26) and text data to the sentence generation AI system 5 (S7).

[0098] When the sentence generation AI system 5 receives a prompt and text data via the API, it reads the received text data and generates a sentence (corporate financial statement sentence) based on the read text data and in accordance with the instructions written in the received prompt. Furthermore, when the sentence generation AI system 5 generates the corporate financial statement sentence using the received prompt and text data, it transmits the generated corporate financial statement sentence to the access source (in this case, the financial statement summary processing device 20).

[0099] Figure 9 shows the beginning of Company X's quarterly financial statement document 30 in response to Company X's quarterly financial results summary, as an example of a corporate financial statement document generated by the text generation AI system 5. This Company X's quarterly financial statement document 30 is based on a format in accordance with the instructions of a prompt (quarterly financial statement prompt 25). Specifically, Company X's quarterly financial statement document 30 shown in Figure 9 includes explanations of good and bad points in each paragraph in accordance with the instructions of the prompt, and has appropriate content for a financial statement document in Japanese.

[0100] Furthermore, Figure 10 shows the beginning of Company X's full-year financial statement document 31, which corresponds to Company X's full-year financial statement summary, as another example of a corporate financial statement generated by the sentence generation AI system 5. Like Company X's quarterly financial statement document 30 shown in Figure 9 above, Company X's full-year financial statement document 31 is based on a format in accordance with the instructions of a prompt (however, in the case of Figure 10, full-year financial statement prompt 26). The format of Company X's full-year financial statement document 31 shown in Figure 10 is also basically based on the instructions of a prompt, but there are occasional English words mixed in with the names of paragraphs and items, such as "Overview of current fiscal year's business results, etc." being "management_performance," "good points" being "positive," and "bad points" being "negative."

[0101] As shown in Figure 10 above, even if a prompt (e.g., annual financial statement prompt 26) specifies that the language used in the sentence should be Japanese, the sentence may not be generated as specified because the general-purpose sentence generation AI system 5 generates the sentence. As shown in Figure 10, the names of paragraphs and items contain English characters, but the content of each paragraph and item is clearly written in Japanese, so the sentence is usable to the extent shown in Figure 10. On the other hand, if the content of each paragraph or item contains English characters or if the content of each paragraph or item is missing, it may be determined to be an error according to the condition information specified in the error criteria 28 described above. Note that, because the general-purpose sentence generation AI system 5 generates sentences, even when the same prompt and the same text data are sent, different expressions or phrases may be generated, or the amount of generated sentences may differ.

[0102] Returning to the first flowchart in Figure 8, the financial statement summary processing device 20 (MPU 20a) transmits the prompt and text data to the sentence generation AI system 5 at step S7, and then determines whether or not a corporate financial statement sentence is to be transmitted from the sentence generation AI system 5 (S8). If the corporate financial statement sentence is not received (S8: NO), the device waits for the corporate financial statement sentence to be transmitted.

[0103] On the other hand, if the MPU 20a receives the corporate financial statement generated by the sentence generation AI system 5 as a receiving means by sending a prompt and text data (S8: YES), it determines whether there are any problems with the format (style) of the corporate financial statement generated by the sentence generation AI system 5 based on the error criteria 28 stored in the memory unit 20h (S9).

[0104] If it is determined that there is no problem with the format of the generated corporate financial statement (S9: YES), the MPU 20a reads the format data 27 from the storage unit 20h, and extracts and applies the contents corresponding to the paragraphs and items related to the blanks 27a of the format data 27 from the generated corporate financial statement to the blanks 27a, 27b, 27c, etc. of the paragraphs and items of the read format data 27, thereby creating an explanation of the corporate financial statement summary (S10).

[0105] Specifically, in the blank 27a for "Good points" in the "Business performance" item in the paragraph "Overview of business performance, etc. for the current period" in format data 27, the content written in the "Good points" section of the "Overview of business performance, etc. for the current period" in the corporate financial statements generated by text generation AI system 5 (for example, X Company quarterly financial statements 30 and X Company full-year financial statements in Figures 9 and 10) that corresponds to the "Good points" in the "Overview of business performance, etc. for the current period" in format data 27 (for example, in X Company quarterly financial statements 30 in Figure 9, "Growth is progressing across all fields, which is expected to have a positive impact on operating profits.") is extracted and applied (pasted).

[0106] Furthermore, the contents of the "negative points" section of the "Business performance" section in the "Overview of business performance, etc. for the current period" section of the X Company quarterly financial statement document 30 in Figure 9 ("Sales were revised downward due to exchange rate fluctuations, etc., and also...") are extracted and applied (pasted) to the blank 27b for "negative points" in the "Business performance" section in the "Overview of business performance, etc. for the current period" section of the format data 27. By having the MPU 20a perform such processing, the present invention ultimately creates an explanation of the corporate financial statement summary.

[0107] 11 and 12 show an example of a corporate financial statement commentary created by the financial statement processor 20 (MPU 20a) at step S10 described above, which is an X Company Quarterly Financial Statement Commentary 40 corresponding to the quarterly financial statement of X Company. This X Company Quarterly Financial Statement Commentary 40 is arranged in such a way that the content written in the paragraphs, items, etc. in the X Company Quarterly Financial Statement Text 30 (see FIG. 9) that correspond to each paragraph, item, etc. in the format data 27 of FIG. 7 is applied to those paragraphs, items, etc., and is therefore much easier to read than the X Company Quarterly Financial Statement Text 30 generated by the text generation AI system 5 shown in FIG. 9.

[0108] Furthermore, compared to the contents of the original quarterly financial results PDF data 3a shown in Figure 3(a), the contents of the original, which were presented in a table including numerical values ​​as shown in Figure 3(a), are expressed in text in Company X's quarterly financial results commentary 40 in Figures 11 and 12, making the content easier to understand for general investors who are not accustomed to interpreting tables. Moreover, Company X's quarterly financial results commentary 40 explains the "good points" and "bad points" in text in paragraphs and items related to business performance, making it easier to understand the company's business performance from both its good and bad points.

[0109] 13 and 14 show Company X Full-Year Financial Results Commentary 41 corresponding to Company X's full-year financial results, as another example of a corporate financial results commentary created by the financial results commentary processing device 20 (MPU 20a) at the above-mentioned step S10. This Company X Full-Year Financial Results Commentary 41, like Company X Quarterly Financial Results Commentary 40 in Figures 11 and 12 described above, is arranged in such a way that the contents of the paragraphs, items, etc. in the Company X Full-Year Financial Results Text 31 (see Figure 10) corresponding to each paragraph, item, etc. in the format data 27 in Figure 7 are applied to those paragraphs, items, etc., and is therefore easier to read than Company X Full-Year Financial Results Text 30 generated by the text generation AI system 5 shown in Figure 10.

[0110] Furthermore, compared to the contents of the original PDF data 3b of the annual financial results summary shown in Figure 3(b), the content that was presented in a table including numerical values ​​in the original is presented in text in the commentary 41 of X Company's annual financial results summary shown in Figures 13 and 14, which has the advantage that text is easier to understand for general investors who are not accustomed to understanding information presented in tables. In particular, in the commentary 41 of X Company's annual financial results summary, the "good points" and "bad points" are explained in text in paragraphs and items related to business performance, etc., making it easier to understand the good and bad points of the company's business performance, etc., and therefore providing content that can be used to make investment decisions.

[0111] Returning again to the first flowchart of Figure 8, once the financial results summary processing device 20 (MPU 20a) creates a corporate financial results summary commentary in step S10, it stores the created corporate financial results summary commentary in the DB storage unit 20g (S11). During this storage, the created corporate financial results summary commentary is stored in a folder for each fiscal period within a folder for each company provided in the DB storage unit 20g, making it easy to read when using the created corporate financial results summary commentary. Therefore, for example, the corporate financial results summary commentary created from Company X's third quarter financial results summary for the fiscal year ending March 2024 shown in Figure 3(a) is stored in the 2024 fiscal year folder within Company X's folder in the DB storage unit 20g.

[0112] The financial statement processing device 20 (MPU 20a) then determines whether there is any financial statement data to be processed next (S12). This determination is made based on the number of pieces of financial statement data stored in RAM 20c whose reception has been detected (the number stored in RAM 20c when S1: YES). If there is only one piece of financial statement data stored in RAM 20c, that piece has been processed through the steps described above, and it is determined that there is no financial statement data to be processed next (S12: NO). As a result, the processing of the MPU 20a shown in the first flowchart in Figure 8 is temporarily terminated. Note that the termination of this processing closes the running state of the processing program P2.

[0113] On the other hand, if there are multiple items stored in RAM 20c, it is determined that there is financial statement data to be processed next (S12: YES), and the processing of MPU 20a returns to step S3, whereupon the same processing as above is carried out for the second item of financial statement data. Note that if it is determined that there is financial statement data to be processed next (S12: YES), the MPU 20a performs processing (countdown processing) to decrement by one the number of items stored in RAM 20c at that time (the number of financial statement data to be processed).

[0114] Furthermore, if it is determined at step S9 in the first flowchart that there is a problem (error) in the format of the generated corporate financial statement (S9: NO), the MPU 20a determines (S13) whether the error is with the first corporate financial statement generated by the document generation AI system 5. If it is determined that the error is the first time (first time) (S13: YES), the fact that the first error has occurred is recorded in the RAM 20c, and the MPU 20a returns its processing to step S7 in the first flowchart.

[0115] When processing returns to step S7, steps S7 and S8 are performed again, and the same prompt and text data (the data converted to text in step S6 and stored in RAM 20c) are passed to the sentence generation AI system 5 to generate a corporate financial statement sentence. The generated corporate financial statement sentence (the corporate financial statement sentence generated in the second processing) is then sent to the financial statement summary processing device 20, and in step S9, the format of the corporate financial statement sentence is checked by the financial statement summary processing device 20 (MPU 20a). As mentioned above, even under the same conditions, a general-purpose sentence generation AI system 5 may generate sentences that differ (e.g., different expressions or sentence length). Therefore, even if the same prompt and text data as in the first processing are passed as described above, the generated corporate financial statement sentence may differ from the corporate financial statement sentence generated in the first processing.

[0116] If it is determined that there are no problems (errors) in the format of the second generated corporate financial statement (S9: YES), the process from S10 onwards is carried out as described above. In this case, the number of errors stored in RAM 20c (indicating that the first error occurred) is deleted (reset) under the control of MPU 20a.

[0117] On the other hand, if it is determined that the second generated corporate financial statement document has a problem (error) in its format, etc. (S9: NO), it is determined whether the error is in the first document (S13), and since it is determined that this is an error in the second document (S13: NO), the financial statement summary processing device 20 (MPU 20a) outputs a notice of the occurrence of an error to the display 22 (S14). In this case, the process flow in the first flowchart proceeds to step S12, where it is determined whether there is financial statement summary data to be processed next.

[0118] In this embodiment, when an error is output, the administrator or the like takes manual action, viewing the error displayed on the display 22, manually correcting the error, and generating a company financial statement. Then, the MPU 20a performs the processing of steps S10 and S11 on the generated company financial statement, and manually creates an explanation of the company financial statement summary and stores it in the DB storage unit 20g.

[0119] As described above, the present invention determines whether the financial statement data, which corresponds to the entire content of the financial statement containing tables and other data including numerical values, is a quarterly or full-year financial statement, and issues instructions to the general-purpose text generation AI system 5 to generate a corporate financial statement corresponding to the result of this determination, thereby enabling the generation of a corporate financial statement with content appropriate to the fiscal period, and furthermore, the instructions include content that mentions both good and bad points regarding the company's performance, making it easy for general investors to understand. Moreover, the present invention extracts the content of the corporate financial statement generated by AI (artificial intelligence) and finally creates a corporate financial statement commentary with an easy-to-understand layout, which can be used in a variety of ways as user-friendly content.

[0120] A specific example of how the created corporate financial statement commentary can be used is to post the corporate financial statement commentary created by the financial statement processor 20 on a website that distributes information related to stocks, etc. As described above, the corporate financial statement commentary is automatically created based on the corporate financial statement text generated by the text generation AI system 5, which significantly reduces the creation time compared to when a writer or the like creates it manually. As a result, the commentary can be quickly posted on the website and quickly provided to users such as investors (the creation cost can also be significantly reduced compared to when it is created manually). When such a website provides information to both paid and free members, one possible use would be to make the created corporate financial statement commentary available to paid members in its entirety, while making only a portion of the created corporate financial statement commentary available to free members.

[0121] The present invention is not limited to the above-described embodiment, and various modifications are possible. For example, in the above description, the financial statement data of each company distributed from the stock exchange system 2 covers two patterns: quarterly financial statements and full-year financial statements (final financial statements). However, companies may announce revised financial statements that amend the contents of their announced quarterly or full-year financial statements. Therefore, the sentences corresponding to these revised financial statements may also be generated by the sentence generation AI system 5, and the financial statement processing device 20 (MPU 20a) may also create an explanation.

[0122] When applying a modified example that can also handle revised financial statements in this way, a prompt for revised financial statements is stored in advance in the memory unit 20h of the financial statement summary processing device 20, and the financial statement summary processing device 20 detects that revised financial statement data is received by the financial statement DB system 15.When this is detected, it determines that the data corresponds to a revised financial statement (the revised financial statement data that is distributed is accompanied by tag information indicating that the data is a revised financial statement), reads out the prompt for revised financial statements, and then performs processing from S6 onwards in the first flowchart of Figure 8, thereby causing the text generation AI system 5 to generate a corporate financial statement text corresponding to the revised financial statement, and finally the financial statement summary processing device 20 (MPU 20a) can create a corporate financial statement commentary corresponding to the revised financial statement.

[0123] In this variant, text and explanations corresponding to revised financial statements are also obtained, so that it can accommodate three types of corporate financial statement content distributed from the stock exchange system 2, and various corporate financial statement texts and explanations can be quickly and automatically provided for these three types, which has the advantage of enabling expansion into a wide range of business services.

[0124] Furthermore, in the above explanation, the format data 27 stored in the memory unit 20h of the financial statement summary processing device 20 is common regardless of whether the financial statement summary data received by the financial statement DB system 15 corresponds to quarterly financial statements or full-year financial statements, but it is also possible to store dedicated format data that specify formats corresponding to quarterly financial statements and full-year financial statements, such as format data for quarterly financial statements (first format data) and format data for full-year financial statements (second format data), in the memory unit 20h (corresponding to the format storage means), and to use each dedicated format data when creating the corporate financial statement summary commentary.

[0125] In this modified example, the format data for quarterly settlement specifies names of paragraphs, items, etc. that make it clear that it is quarterly settlement, and the format data for full-year settlement specifies names of paragraphs, items, etc. that make it clear that it is full-year settlement. These format data are used in such a way that, at step S3 of the first flowchart in Figure 8, the result of determining whether the new financial statement data whose reception is detected corresponds to quarterly settlement or full-year settlement is stored in RAM 20c, and at step S10, the format data corresponding to the determination result stored in RAM 20c is read, and the content of the corporate financial statement document is applied to the read format data to create an explanation of the corporate financial statement.

[0126] In this way, by using format data for quarterly settlement and full-year settlement, the final corporate financial statement commentary will have a format that clarifies whether it is quarterly or full-year settlement, making it easier to understand the contents of each paragraph and item in the corporate financial statement commentary. Note that the modified version of this format data can also be applied to the case of revised settlement described in the previous modified version, and in this case, format data for revised settlement will also be stored in the storage unit 20h.

[0127] Furthermore, the contents of the instruction information of the quarterly settlement prompt 25 shown in Figure 5 or the full-year settlement prompt 26 shown in Figure 6 are merely examples, and it is of course possible to apply instruction information of other content corresponding to quarterly settlement to the quarterly settlement prompt 25, or to apply instruction information of other content corresponding to full-year settlement. Specifically, it is possible to envision different divisions such as paragraph divisions or item divisions, and the divisions may be made more roughly than in Figures 5 and 6, using a paragraph name such as "Major changes in corporate performance," or conversely, the divisions may be made more finely than in Figures 5 and 6.

[0128] Furthermore, when conveying information about stocks to buy or sell to investors, items such as a buying trend perspective and a selling trend perspective are provided, and instruction information such as generating sentences explaining the reasons for stocks that are in a buying trend and generating sentences explaining the reasons for stocks that are in a selling trend is included in each prompt 25, 26. When changing the content related to the paragraphs, items, etc. of each prompt 25, 26 as described above, the names of the paragraphs, items, etc. included in format data 27 (see FIG. 7) will also be changed accordingly.

[0129] Furthermore, in the above example, the time interval for detecting receipt of financial statement summary data at step S1 in Fig. 8 was not specified, so it can be assumed that it will be performed at a predetermined time interval (for example, every 15 minutes or 30 minutes, etc.), but in order to generate corporate financial statement documents and create corporate financial statement summary commentaries in a more timely manner, it would be preferable to perform the detection process in accordance with the timing of the financial statement announcements of each listed company. In other words, since the timing of when each company will announce its financial statements is often known in advance, and the periods when financial statement announcements of each company are concentrated are also known in advance, it would be preferable in terms of processing efficiency to perform the above detection process in accordance with such periods.

[0130] Figure 15(a) shows the contents stored in the memory unit 20h of the financial statement summary processing device 20 in the modified example described above in which detection processing is performed in accordance with the timing of each company's financial statement announcement. In this modified example, in addition to the contents stored in the memory unit 20h shown in Figure 4, a detection execution program P3 and calendar information 50 are stored in the memory unit 20h. The detection execution program P3 specifies that the MPU 20a will perform processing to determine the timing and time interval for performing detection at stage S1 in Figure 8, based on the timing function of the MPU 20a and the calendar information 50. The calendar information 50 indicates the date and time of each company's financial statement announcement.

[0131] 15(b) shows a portion of an example of the contents of calendar information 50. In this example, information (announcement information) is shown indicating that companies A, D, F, etc. will announce their financial results at 3:30 PM on February 14, 2024, and information is also shown indicating that companies B, E, T, etc. will announce their financial results at 4:00 PM on the same day.

[0132] Figure 16 is a second flowchart showing the processing steps (contents of the financial statement summary processing method) for the MPU 20a to determine the timing to start the first flowchart of Figure 8 and start detection as a detection means, in accordance with the provisions of the detection execution program P3. This second flowchart utilizes the timekeeping function (calendar / clock function) of the MPU 20a based on the basic provisions of the OS program P1, etc., and in the first step S20, the MPU 20a uses the timekeeping function to determine whether the current time has reached midnight. If the determination result shows that the current time has not reached midnight (S20: NO), the MPU 20a enters a state of waiting for midnight.

[0133] On the other hand, if the result of the determination is that the current time is midnight (S20: YES), the day will change accordingly, and the MPU 20a refers to the calendar information 50 to determine whether there will be a financial results announcement on the new day at midnight (S21). If it is determined from the calendar information 50 that there will be no financial results announcement on that day (S21: NO), the process returns to step S20, and the MPU 20a waits for midnight of the next day.

[0134] Furthermore, if it is determined from the calendar information 50 that a financial results announcement will be made on that day (S21: YES), the MPU 20a reads out the time of the financial results announcement indicated in the calendar information 50 and stores it in the RAM 20c (S22). Note that if it is indicated that financial results announcements will be made at multiple times on that day, all of those multiple times will be read out and stored in the RAM 20c.

[0135] The MPU 20a then determines whether the current time has reached the time stored in RAM 20c (S23), and if the current time has not reached the time stored in RAM 20c (S23: NO), it enters a state of waiting for the time stored in RAM 20c to arrive. If the current time has reached the time stored in RAM 20c (S23: YES), the MPU 20a issues an instruction to start the processing program P2, thereby starting the processing shown in the first flowchart in Fig. 8 (S24). This causes the settlement summary processing device 20 (MPU 20a) to perform the detection at step S1 of the first flowchart.

[0136] In the second flowchart of FIG. 16, the MPU 20a uses its timer function to determine whether five minutes have passed since the startup process in step S24 (S25). If five minutes have not passed (S25: NO), the MPU 20a waits for five minutes to pass. If five minutes have passed (S25: YES), the MPU 20a issues an instruction to start the processing program P2 again (S26). The reason for issuing the startup instruction again in step S26 is that, depending on the timing of the process, the startup instruction in step S24 may not be able to detect the receipt of newly distributed financial summary data, so the program is started again as a backup. Note that if the processing program P2 is running and the process in the first flowchart of FIG. 8 is being executed at step S26, the instruction to start the processing program P2 again is canceled.

[0137] The MPU 20a then determines whether there will be another financial announcement later that day, based on the contents stored in the RAM 20c (S27). If there will be another financial announcement later that day (S27: YES), the process returns to step S23, where it is determined whether the next financial announcement time has arrived, and if the next financial announcement time has arrived (S23: YES), the process from step S24 onwards is carried out. Also, if there is no financial announcement later that day at step S27, the process of the MPU 20a returns to the initial step S20.

[0138] 16, it is possible to efficiently detect the receipt of new financial statement summary data from the securities exchange system 2, thereby reducing the processing load on the financial statement summary processing device 20. As mentioned above, since financial statement summary data relating to revised financial statements may also be distributed from the securities exchange system 2, it is effective to also execute the process of starting the processing program P2 at predetermined time intervals (for example, every 30 minutes or every hour) in parallel with the process of the second flowchart in FIG.

[0139] Furthermore, multiple companies often announce their financial results at the same time, in which case the receipt of financial statement data from multiple companies is detected at step S1 in the first flowchart in Figure 8. In this first flowchart, each piece of financial statement data is processed individually, so when the receipt of financial statement data from multiple companies is detected, it is necessary to determine the order in which to process the multiple pieces of financial statement data. Possible ways to determine the order include the order in which the financial statements are received, or, since each company is assigned a four-digit code number, determining the order in descending or ascending order of the code number, or determining the order in alphabetical order of the company names.

[0140] Figure 17(a) shows the contents stored in the storage unit 20h when one example (one of the modified examples) of determining the order of the financial statement data is performed when receipt of financial statement data for multiple companies is detected. The example in Figure 17(a) shows a situation in which trading volume information 51 indicating the trading volume (amount of trades completed) of each company's stock for the previous month is stored in the storage unit 20h. This trading volume information 51 indicates the ranking of trading volume for each company's stock for the previous month, and in this example, the information content is such that each company is ranked in descending order of trading volume, starting with the company with the largest trading volume, and the financial statement processing device 20 obtains trading volume information 51 from an external system each month and stores it in the storage unit 20h.

[0141] 8, when the MPU 20a detects receipt of financial statement summary data of multiple companies as detection means, it references such trading volume information 51, and performs a process of identifying the order of companies with the highest trading volume among the companies related to the financial statement summary data whose reception was detected, based on the ranking indicated by the trading volume information 51, and storing the identified order in RAM 20c. Then, in accordance with the order stored in RAM 20c, it selects the financial statement summary data corresponding to the company that is the first target of processing, and performs subsequent processing (processing from S2 onwards).

[0142] If processing of the first financial statement data is completed at step S11 or S14, the financial statement data corresponding to the next company to be processed is determined at step S12 according to the processing order stored in RAM 20c, and processing of the next financial statement data to be processed begins at step S3. Thereafter, the above-described processing is repeated up to the financial statement data corresponding to the last company to be processed, thereby repeating the processing of generating a company financial statement document for each company and the processing of generating a company financial statement commentary for each company in the order stored in RAM 20c.

[0143] It is expected that the order of stocks with the highest trading volume will tend to be the same as the order of stocks that are of interest to each investor, so when the earnings release data of multiple companies is to be processed in this way, by performing the processing shown in the first flowchart in Figure 8 in the order of the earnings release data of companies with the highest trading volume, it becomes possible to generate corporate earnings statements starting from those that are of most interest to each investor and provide commentary on the corporate earnings releases.

[0144] In the above explanation, the trading volume information 51 is from the previous month, but the trading volume information 51 is not limited to that from the previous month, and any information showing the trading volume of each stock over a predetermined period in the past can be used. For example, trading volume information from any period from the most recent one to four weeks can be used, and in such a case, the financial statement summary processing device 20 will obtain the trading volume information from an external system and store it in the memory unit 20h each time it is updated. In this way, by using trading volume information from any period from the most recent one to four weeks, each investor can obtain corporate financial statements and corporate financial statement commentaries in the order of stocks that they have been interested in over the most recent period. Conversely, it is of course possible to use trading volume information from a longer period,

[0145] 17(b) shows the contents stored in the storage unit 20h in another example (another modified example) of determining the order of the financial statement data when receipt of financial statement data for multiple companies is detected. This Fig. 17(b) shows a situation in which access count information 52 indicating the ranking of the number of accesses to information for each stock of each company in the previous month (number of accesses per stock) is stored in the storage unit 20h on a website that provides information on stocks. This access count information 52 contains information in which companies are ranked in order from 1st place according to the stock with the most accesses, and the financial statement processing device 20 obtains access count information 52 from an external system that operates the website that provides information on stocks each month and stores it in the storage unit 20h.

[0146] 8, this access count information 52 is also referenced when the MPU 20a detects receipt of financial statement summary data of multiple companies as detection means at step S1 of the first flowchart in Fig. 8, and performs a process of identifying the order of companies with the most accesses among the companies related to the financial statement summary data whose reception was detected based on the ranking indicated by the access count information 52, and storing the identified order in RAM 20c. Then, the process of generating a corporate financial statement document for each company and the process of generating a corporate financial statement commentary are repeated in the order stored in RAM 20c.

[0147] The order of the most frequent accesses to information on each company's stock is likely to basically follow the same trend as the order of stocks of interest to general investors, but compared to trading volume information 51 based on actual trading, access count information 52 includes the interest of investors and other users in predicting future trading, and therefore more broadly represents stocks of interest to investors and other users. Therefore, there may be differences between the order of stocks of companies with the most frequent accesses indicated by access count information 52 and the order of stocks of companies with the most trading volume indicated by trading volume information 51 shown in Figure 17(a), and in this respect, there is a significance in using trading volume information 51 and access count information 52 separately.

[0148] The access count information 52 is used in the same way as the trading volume information 51 described above, and in addition to using the access count information 52 from the previous month, any information showing the number of accesses to each stock during a specified period in the past can be used, just like the trading volume information 51 described above.

[0149] Furthermore, in the above explanation, PDF data is used as the financial statement data including the general content of a company's quarterly or full-year financial statement distributed from the securities exchange system 2, but the data format of such financial statement data is not limited to PDF data, and other data formats can also be used. For example, HTML (Hyper Text Markup Language) data is actually used as the format of financial statement data including the general content of a company's quarterly or full-year financial statement in overseas countries such as the United States, and such HTML data can also be used as financial statement data including the general content of the financial statement in the present invention.

[0150] HTML format data includes tags and other elements that make up the format, but in the present invention, tags and other elements are removed by text conversion to create text data that represents the content of the financial results summary. Therefore, even if the financial results summary data to be distributed is in HTML format, the text generation AI system 5 can automatically generate a corporate financial results summary in accordance with the processing in the first flowchart shown in Figure 8, and ultimately automatically create an explanation of the corporate financial results summary. [Industrial Applicability]

[0151] The present invention determines whether a published corporate financial results summary is a quarterly or full-year financial statement, and issues instructions to a general-purpose text generation AI system to generate a corporate financial statement document with content corresponding to the determination result.Therefore, it can be suitably used for applications such as automatically and quickly generating corporate financial statement documents with content corresponding to quarterly or full-year financial statements. [Explanation of symbols]

[0152] 1. Financial Results Briefing System 2. Stock Exchange System 5. Text generation AI system 10. Financial statement processing system 15. Financial Statement DB System 16 DB management device 17 Update Information Table 18 DB device 20. Financial statement summary processing device 20a MPU 20g DB storage unit 20h storage section 24 API Key 25 Quarterly Earnings Prompts 26 Full-year Financial Statement Prompt 27 Format Data 28 Error Criteria

Claims

1. In a financial statement processing system, an external text generation AI system generates corporate financial statements based on financial statement data including the general content of a company's quarterly financial statement or full-year financial statement distributed from a data source. a financial statement database system that receives and stores financial statement summary data distributed from a data distribution source; a storage means for storing first instruction information instructing the generation of a company financial statement corresponding to a quarterly settlement and second instruction information instructing the generation of a company financial statement corresponding to a full-year settlement; means for detecting that new financial statement data distributed from a data distributor is received by the financial statement database system; a determination means for determining whether the new financial statement data corresponds to a quarterly financial statement or a full-year financial statement when the receipt of new financial statement data is detected in the financial statement database system; a means for reading from said storage means either first or second instruction information for instructing the generation of a company financial statement in accordance with the result of the determination by said determination means; conversion means for extracting a text portion from the new financial summary data and converting it into text data; means for transmitting the instruction information read from the storage means and the text data converted by the conversion means to an external sentence generation AI system; a means for receiving, from an external document generation AI system, a corporate financial statement document generated by the external document generation AI system by transmitting the instruction information and text data; A financial statement summary processing system comprising:

2. The financial statement processing system of claim 1 is provided with a means for creating a corporate financial statement commentary by applying the contents of the received corporate financial statement document to format data that specifies a format according to the financial statements when the financial statement document is received from an external document generation AI system.

3. 3. The financial statement processing system according to claim 1, wherein the first instruction information and the second instruction information include an instruction to generate a corporate financial statement document that lists the good and bad points of the financial statement from the contents of the financial statement.

4. In a financial statement summary processing device, an external text generation AI system generates corporate financial statement text based on financial statement summary data including the general content of a company's quarterly financial statement summary or full-year financial statement summary distributed from a data distribution source. a storage means for storing first instruction information instructing the generation of a company financial statement corresponding to a quarterly settlement and second instruction information instructing the generation of a company financial statement corresponding to a full-year settlement; a detection means for detecting that new financial statement data has been received in an external financial statement database system that receives and stores financial statement data distributed from a data distribution source; a determination means for determining whether the new financial statement data corresponds to a quarterly financial statement or a full-year financial statement when the detection means detects that new financial statement data has been received in the external financial statement database system; a means for reading from said storage means either first or second instruction information for instructing the generation of a company financial statement in accordance with the result of the determination by said determination means; conversion means for extracting a text portion from the new financial summary data and converting it into text data; means for transmitting the instruction information read from the storage means and the text data converted by the conversion means to an external sentence generation AI system; a means for receiving, from an external document generation AI system, a corporate financial statement document generated by the external document generation AI system by transmitting the instruction information and text data; A financial statement summary processing device comprising:

5. A financial statement summary processing device as described in claim 4, which is provided with a means for creating a corporate financial statement summary commentary by applying the contents of the received corporate financial statement document to format data that specifies a format corresponding to the financial statement when the financial statement document is received from an external document generation AI system.

6. 6. The financial statement summary processing device according to claim 4, wherein the first instruction information and the second instruction information include an instruction to note good and bad points of the financial statement summary content.

7. 6. The financial statement summary processing device according to claim 4, further comprising means for causing said detecting means to start detecting based on calendar information indicating the date and time of a company's financial statement announcement.

8. a means for storing trading volume information indicating the ranking of trading volume for each company's stock; an order specifying means for specifying, when the detection means detects that new financial statement summary data of a plurality of companies has been received in an external financial statement database system, the order of companies with high trading volume among the companies related to the new financial statement summary data whose reception has been detected, based on the order indicated by the trading volume information; Equipped with 6. A financial statement summary processing device according to claim 4, wherein the process of generating a company financial statement document for each company is repeated in the order specified by said order specifying means.

9. a means for storing access count information indicating a ranking of access counts to information on each company's stock on a website that provides stock-related information; an order specifying means for specifying, when the detection means detects that new financial statement data of a plurality of companies has been received in an external financial statement database system, the order of companies having the most number of accesses among the companies related to the new financial statement data whose reception has been detected, based on the order indicated by the access count information; Equipped with 6. A financial statement summary processing device according to claim 4, wherein the process of generating a company financial statement document for each company is repeated in the order specified by said order specifying means.

10. a format storage means for storing first format data defining a format corresponding to quarterly settlement and second format data defining a format corresponding to full-year settlement; means for reading out either first or second format data from said format storage means in accordance with the result of the determination by said determination means; Equipped with 6. The financial statement processing device according to claim 5, wherein an explanation of the financial statement is created by applying the content of the financial statement to the format data read from the format storage means.

11. In a financial statement processing method, a financial statement processing device processes financial statement data including the general content of a company's quarterly financial statement or full-year financial statement distributed from a data distribution source, and causes an external text generation AI system to generate a corporate financial statement. The settlement processing device is First instruction information instructing the creation of a company financial statement corresponding to a quarterly settlement and second instruction information instructing the creation of a company financial statement corresponding to a full-year settlement are stored; detecting that new financial statement data has been received in an external financial statement database system that receives and stores financial statement data distributed from a data distribution source; When it is detected that new financial statement data has been received by the external financial statement database system, determining whether the new financial statement data corresponds to a quarterly financial statement or a full-year financial statement; a step of reading from storage either first or second instruction information instructing the generation of a corporate financial statement document according to the determination result; A step of extracting a text portion from the new financial summary data and converting it into text data; Sending the read instruction information and the converted text data to an external sentence generation AI system; receiving, from the external document generation AI system, the enterprise financial statement document generated by the external document generation AI system by transmitting the instruction information and text data; A financial statement processing method characterized by executing the above.

12. A computer program for causing a computer that stores first instruction information instructing the generation of a company's financial statements in accordance with quarterly financial statements and second instruction information instructing the generation of a company's financial statements in accordance with full-year financial statements to perform a process of generating a company's financial statements using an external text generation AI system based on financial statement summary data including the general content of a company's quarterly financial statements or full-year financial statements distributed from a data distribution source, The computer, The settlement processing device is detecting that new financial statement data has been received in an external financial statement database system that receives and stores financial statement data distributed from a data distribution source; When it is detected that new financial statement data has been received by the external financial statement database system, determining whether the new financial statement data corresponds to a quarterly financial statement or a full-year financial statement; a step of reading from storage either first or second instruction information instructing the generation of a corporate financial statement document according to the determination result; A step of extracting a text portion from the new financial summary data and converting it into text data; Sending the read instruction information and the converted text data to an external sentence generation AI system; receiving, from the external document generation AI system, the enterprise financial statement document generated by the external document generation AI system by transmitting the instruction information and text data; A computer program characterized by causing the computer to execute the following:

Images