JP2021165967A - Accounting processor, accounting processing system, accounting processing method and program - Google Patents

Abstract

To perform automatic journaling from voucher data being an image of a voucher.SOLUTION: An accounting processor for performing automatic journaling from voucher data being an image of a voucher includes an image recognition part for extracting transaction information from the voucher data by image recognition processing, a character recognition part for generating text data including transaction detail information from the extracted transaction information by character recognition processing, and a journaling generation part for generating journaling data from the transaction detail information by selecting a combination of account titles corresponding to the transaction detail information.SELECTED DRAWING: Figure 3

Description

The present invention relates to accounting equipment, accounting systems, accounting methods and programs.

In conventional accounting processing, when entering a voucher such as a receipt or receipt as a journal, a tax accountant or accounting staff reads the information (transaction content) described on the voucher one by one and empirically examines the account item corresponding to the transaction content. I made a judgment and entered the journal. If the reading work is performed one by one, the work efficiency is poor, and the accuracy of the journal entry largely depends on the experience of the person in charge, and there is a problem that the accuracy of the journal entry varies.

Therefore, there is disclosed a technique for determining a store name and supporting entry in a household account book or the like by using a slip image obtained by scanning or photographing a receipt or the like (see, for example, Patent Document 1).
In addition, since the transaction content that occurs for each business partner is a continuous transaction, it is often the same transaction content, and journal data can be automatically generated from the transaction content for standard transactions (for standard transactions). For example, see Patent Document 2).

Japanese Unexamined Patent Publication No. 2017-174309 Japanese Patent No. 3929824

The system disclosed in Patent Document 1 uses the information of the business partner printed on the receipt or the like, and text data is obtained from the voucher image by optical character recognition processing using OCR (Optical Character Recognition). What you get.
Further, in the system disclosed in Patent Document 2, each transaction is referred to a correspondence table that holds a correspondence between a keyword and an account that can be included in the description of the transaction content based on the description of each transaction content. Then, journal entry is automatically performed for a specific account.

However, depending on the quality of the voucher image and the character size, it may not be recognized well by OCR. In addition, in transactions other than standard transactions, the system administrator or the like needs to create a correspondence table each time in order to automatically perform journal entry by referring to the correspondence table that holds the correspondence between keywords and accounts. There is.

Therefore, the present invention has been made in view of such circumstances, and an object of the present invention is to provide an accounting processing device (accounting processing system) that automatically performs journal entry from a voucher using artificial intelligence (AI).

In order to solve such a problem, the first invention is an accounting processing device that automatically journalizes from voucher data, which is an image of a voucher, and has an image recognition unit that extracts transaction information from voucher data by image recognition processing, and an extracted transaction. A journal that has learned to select a combination of account items corresponding to transaction detail information by performing machine learning based on teacher data in advance with a character recognition unit that generates text data including transaction detail information from information by character recognition processing. An accounting processing device that includes a journal generation unit that generates journal data from the text data using the generation AI, and displays the journal data created by the journal generation AI, the voucher data, and the transaction details information in parallel. It is characterized by.

The second invention includes a journal verification unit that compares and collates the reference journal data associated with the transaction detail information that is the same as or similar to the transaction detail information and the journal data.
The accounting processing device according to the first invention, wherein the journal verification unit determines a range to be compared and collated with a predetermined threshold value, and compares and collates the journal data generated by the journal generation AI with the reference journal data.

In the third invention, the former image recognition unit performs machine learning based on the teacher data in advance, and learns the position and contents of the transaction information in the voucher data from the transaction information corresponding to the area in which the voucher data is divided into a plurality of areas. The accounting apparatus according to the first invention or the second invention, which comprises the image recognition AI.

The fourth invention is characterized in that the character recognition unit has a character recognition AI that has learned in advance to perform machine learning based on teacher data and generate transaction detail information necessary for journal generation from the transaction information. It is characterized by the accounting device according to any one of the first to third inventions.

The fifth invention is based on the first invention, wherein the journal verification unit has a journal verification AI that has learned to perform machine learning based on teacher data in advance and determine the validity of the journal data. It features an accounting device according to any one of the fourth inventions.

A sixth aspect of the present invention is characterized in that the journal verification unit has a journal verification range AI that has learned to perform machine learning based on teacher data in advance and determine a journal verification range corresponding to transaction detail information. The accounting apparatus according to any one of the inventions 1 to 5 is characterized.

In the seventh invention, the journal generation unit has learned to generate classification information by classifying items of transaction detail information and select a combination of accounts corresponding to the transaction detail information including the classification information. The accounting device according to any one of the first to sixth inventions, characterized in that journal data is generated from the text data using the above.

In the eighth invention, the journal generation unit generates classification information in which items of transaction detail information are classified and amount classification information in which the amount of transaction detail information is classified in stages, and a combination of the classification information and the amount classification information. Any of the first to sixth inventions, characterized in that journal data is generated from the text data using the journal generation AI learned to select a combination of accounts corresponding to transaction detail information consisting of It features the accounting processing device.

In the ninth invention, the journal generation unit generates classification information by classifying the items of the transaction detail information, and selects a combination of accounts corresponding to the transaction detail information consisting of a combination of the classification information and the payment method. The first to sixth inventions characterized in that the journal data is generated from the text data by using the journal generation AI learned to generate the journal data from the text data by using the learned journal generation AI. It features one of the accounting devices of.

In the tenth invention, the journal generation unit generates amount classification information in which the amount of transaction detail information is divided in stages, and a combination of accounts corresponding to the transaction detail information composed of a combination of items and amount classification information. From the first invention characterized in that journal data is generated from the text data using the journal generation AI learned to generate journal data from the text data using the journal generation AI learned to select. It features an accounting device according to any one of the sixth inventions.

In the eleventh invention, the journal generation unit generates journal data from the text data using the journal generation AI learned to select a combination of accounts corresponding to transaction detail information consisting of a combination of items and payment methods. The accounting apparatus according to any one of the first to sixth inventions is characterized in that journal data is generated from the text data by using the journal generation AI learned to be performed.

In the twelfth invention, the journal generation unit causes the journal generation AI to perform machine learning using journal data classified by industry as teacher data in advance, and combines accounts corresponding to the combination of the business's industry and transaction detail information. A first invention characterized in that journal data is generated from the text data by using the journal generation AI learned to generate journal data from the text data by using the journal generation AI learned to select. It is characterized by the accounting processing apparatus according to any one of the sixth and sixth inventions.

A thirteenth invention is an accounting processing method for automatically journalizing voucher data, which is an image of a voucher, an image recognition step of extracting transaction information from voucher data by image recognition processing, and a character recognition process from the extracted transaction information. A character recognition process that generates text data including transaction detail information including at least items, and machine learning based on teacher data in advance, and at least an account item corresponding to transaction detail information including classification information that classifies items. A journal generation step of generating journal data from the text data using the journal generation AI learned to select a combination is provided, and the journal data created by the journal generation AI, the voucher data, and the transaction detail information are provided. It features an accounting method that associates and displays in parallel.

The fourteenth invention includes a journal verification step of comparing and collating the reference journal data associated with the transaction detail information that is the same as or similar to the transaction detail information and the journal data.
The journal verification step is characterized by the accounting method according to the thirteenth invention, wherein the range to be compared and collated is determined by a predetermined threshold value, and the journal data generated by the journal generation AI and the reference journal data are compared and collated. And.

In the fifteenth invention, the image recognition process performs machine learning based on the teacher data in advance, and learns the position and contents of the transaction information in the voucher data from the transaction information corresponding to the area in which the voucher data is divided into a plurality of areas. The accounting method according to the thirteenth invention or the fourteenth invention, which is characterized by having an image recognition AI.

The sixteenth invention is characterized in that the character recognition process has a character recognition AI that has learned to perform machine learning based on teacher data in advance and generate transaction detail information necessary for journal generation from the transaction information. It is characterized by the accounting method according to any one of the thirteenth to fifteenth inventions.

The seventeenth invention is based on the thirteenth invention, wherein the journal verification step has a journal verification AI that has learned to perform machine learning based on teacher data in advance and determine the validity of the journal data. It is characterized by the accounting method according to any one of the sixteenth inventions.

The eighteenth invention is characterized in that the journal verification step has a journal verification range AI learned to perform machine learning based on teacher data in advance and determine a journal verification range corresponding to transaction detail information. It is characterized by an accounting method according to any one of the thirteenth invention to the seventeenth invention.

In the nineteenth invention, the journal generation AI has learned that the journal generation process generates classification information by classifying items of transaction detail information and selects a combination of accounts corresponding to the transaction detail information including the classification information. The accounting method according to any one of the thirteenth to eighteenth inventions, characterized in that journal data is generated from the text data using the above.

The twentieth invention is characterized by an accounting program for causing a computer to execute the accounting method according to any one of the thirteenth invention to the nineteenth invention.

According to the present invention, the transaction information described in the voucher can be read without human intervention, and the journal can be automatically generated from the transaction detail information (transaction content) read from the transaction information, so that the work efficiency of journal input can be achieved. Can be improved. In addition, the journal data generated by the automatic journal can be compared and collated with the existing verified journal data, and it can be determined whether the result of the comparison and collation is the correct journal (normal data) or the largely incorrect journal (abnormal data). By highlighting the difference part of the incorrect journal, it is possible to improve the work efficiency of the journal verification by visually confirming the journal data generated by the automatic journal. In addition, since the scope of journal verification can be changed according to the transaction content when performing journal verification, the optimum journal verification target corresponding to the transaction content can be automatically set.

It is a figure which shows the whole structure of the accounting processing system 100 which concerns on this embodiment. It is a figure which shows the functional block of the accounting processing apparatus 1 which concerns on this embodiment. It is a flowchart which shows the processing in the accounting processing apparatus 1 which concerns on this embodiment. This is an example of voucher data displayed by the display unit 40. This is an example in which the image recognition unit 11 divides the voucher data into a plurality of areas.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In addition, since the components with the same reference numerals perform the same operation in the embodiment, the description may be omitted again.

<Overall configuration of accounting system 100>
FIG. 1 is a diagram showing an overall configuration of the accounting processing system 100 according to the present embodiment.

In the accounting processing system 100 shown in FIG. 1, a server 1a that mainly performs accounting data storage and calculation processing and a terminal 1b that performs data input and instruction for processing are connected by a network, and accounting processing is performed in cooperation with each other. It is a system configuration to be performed. In addition, the server 1a or terminal 1b of the accounting office that receives the request from the adviser executes various business application programs (hereinafter, the application program is also referred to as a "program") to perform accounting processing. It is a system.

The accounting processing system 100 is composed of, for example, a server 1a and a plurality of terminals 1b. The server 1a performs processing based on the operation data received from the terminal 1b. The plurality of terminals 1b are, for example, a person in charge of business processing (hereinafter, "person in charge" or "staff" in the accounting office) who performs business in an accounting office such as a tax accountant, a teacher who is a certified accountant, or a veteran employee. , The adviser (the company that has an adviser contract with the accounting firm) is also called an "employee"). As explained in the accounting office, it may be a department that performs accounting and accounting processing within the company.

Further, in FIG. 1, the accounting processing system 100 is described as being composed of a server 1a and a terminal 1b, which is an example. For example, a system in which a server and a terminal in an accounting office are connected via a network, an accounting processing system between an accounting office and an advisor, an accounting processing system between an accounting office and a person in charge (at home), and accounting processing between a head office sales office. It may be an accounting system between a system, a head office and a branch office, or an accounting system between a head office and a branch office.

FIG. 2 is a diagram showing a functional block of the accounting processing device 1 according to the present embodiment.
The accounting processing device 1 is, for example, a device provided in an accounting office to perform various accounting processing, and the server 1a of the accounting processing system 100 in FIG. 1 and each device (single machine) such as a terminal 1b are accounting processing devices. Corresponds to 1. The accounting processing device 1 performs a plurality of business processing corresponding to various accounting processing.

Further, the accounting processing device 1 may be composed of, for example, one stand-alone (single machine) computer (PC). Further, the accounting processing device 1 may be a workstation (WS) composed of a plurality of PCs. As described above, the device configuration may be processed as a single machine, or may be realized as a system configuration in which the server 1a and the terminal 1b are connected by a network.

Further, the accounting processing device 1 may be not in the accounting office but in the consulting company, in the company providing services such as business consignment, or in the cloud service. For example, the terminal 1b does not necessarily have to be provided in the accounting office, and includes a terminal of a staff member who works on the go or at home, and a terminal of a staff member who is outsourced. Further, the accounting processing device 1 may be a mobile terminal such as a mobile phone or a tablet, a digital camera, a scanner, or the like.

In addition, the accounting processing device 1 may be composed of only a server or a plurality of servers, and may be a server of an accounting office, or may store accounting files of a plurality of companies (employees). It may be an IDC (Internet Data Center) server that provides management services. Further, the accounting processing device 1 may be a virtual machine or a cloud whose substance is virtualized.

Further, when the head office or the like of a company performs financial management for each department or each branch (employee) or the like, the accounting processing device 1 may be a server in the company.
The accounting processing device 1 may be a device that exclusively performs the processing of the present invention, or the device that performs accounting processing has the function of the accounting processing device 1 as one of various functions. May be good.

Further, the accounting processing device 1 may be used as a device provided at the adviser, for example, reading a slip (primitive voucher) such as a receipt or a receipt of the adviser and generating a slip image. The voucher (primitive voucher) is delivered to the accounting office, for example, by mail from the adviser or by the person in charge of the adviser. When performing between the head office and the branch office, the accounting processing device 1 may be located in the head office, the accounting department of the head office, the branch office, the sales office of each base, or the branch office. Further, the accounting processing device 1 may be provided at the adviser's office, or may be provided at the person in charge's home or the like so that the person in charge of the accounting office can perform the work at home.

The embodiments described above may be implemented in computer hardware and computer programs running on it. In the above program, in the transmission of transmitting information and the reception of receiving information, processing performed by hardware, for example, processing performed by a modem or interface card in transmission (processing performed only by hardware). ) Is not included.

Further, the number of computers that execute the above program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

Further, in each of the above embodiments, it goes without saying that the two or more communication means (terminal information transmitting unit, terminal information receiving unit, etc.) existing in one device may be physically realized by one medium. .. The accounting processing device 1 may be processed as a single machine, or the server 1a using the accounting processing device 1 and the terminal 1b using the accounting processing device 1 may be realized as the accounting processing system 100 connected by a network. ..

Further, in each of the above embodiments, each process (each function) may be realized by centralized processing by a single device (system), or may be realized by distributed processing by a plurality of devices. You may.

It goes without saying that the present invention is not limited to the above embodiments, and various modifications can be made, and these are also included in the scope of the present invention. Further, this is only an example, and the technical scope of the present invention is not limited to this.

<Accounting device 1>
As shown in FIG. 2, the accounting processing device 1 includes a control unit 10, a storage unit 30, an input unit 20, a display unit 40, a communication unit 50, and a classifier 60. According to the configuration of the accounting processing system 100 of FIG. 1, the server 1a corresponds to the control unit 10, the storage unit 30, the communication unit 50, and the classifier 60, and the terminal 1b displays the input unit 20. It is an accounting processing apparatus 1 corresponding to the unit 40 and the communication unit 50.

The input unit 20 is an input device such as a keyboard or a mouse. Further, the input unit 20 may be a touch panel or pen input. Further, the input unit 20 may be an input device such as a microphone for voice input, a camera for image input (including a mobile terminal such as a mobile phone or a tablet, a digital camera, etc.), a scanner, or the like.

The display unit 40 is a display device composed of an LCD (liquid crystal display) or the like.

The communication unit 50 is, for example, an interface for communicating with an external device via a slip image reading device such as a scanner or a communication network.

The control unit 10 executes a program in a plurality of business processes corresponding to various accounting processes, and performs processing and management related to data such as data update.
Further, the control unit 10 can usually be realized from an MPU, a memory, or the like. The processing procedure of the control unit 10 is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

Further, the control unit 10 is a CPU (central processing unit) that controls the entire accounting processing device 1. The control unit 10 appropriately reads and executes the OS (operating system) and the application program stored in the storage unit 30 to cooperate with the above-mentioned hardware and execute various functions.

Further, the control unit 10 includes a management unit (not shown) that manages data such as data update, an image recognition unit 11, a character recognition unit 12, a journal generation unit 13, and a journal verification unit 14.

The image recognition unit 11 performs image recognition processing of the voucher data and extracts transaction information which is an image recognition result from the voucher data. A known method may be used for the image recognition process.

The character recognition unit 12 performs character recognition processing for converting voucher data into text using OCR (Optical Character Reader), and generates text data from the character recognition result. Specifically, the character recognition unit 12 converts a character string information list of transaction information (composed of a plurality of lines of character string information) into text to generate transaction detail information. Generate text data from the generated transaction details information of multiple lines and associate it with the voucher data. A known method may be used for the character recognition process.

The journal generation unit 13 generates journal data by processing transaction detail information. When performing journal processing, if the transaction details information is a standard transaction and a transaction pattern exists, refer to the transaction pattern dictionary (not shown) described later, and verify item values and verification confirmation of each item of the journal data. Set the value. Journal data is associated with transaction line information and voucher data.

Further, the journal generation unit 13 searches the registered journal data associated with the transaction detailed information that is the same as or similar to the transaction detailed information from the journal DB 31 described later using the transaction detailed information as a key, and uses the searched registered journal data as the reference journal data. get.

The journal verification unit 14 compares and collates the journal data generated by the journal generation unit 13 with the reference journal data, and whether the journal data generated from the result of the comparison and collation is the correct journal, etc. Perform journal verification (including journal verification).
In addition, the journal verification unit 14 can automatically set the journal data items (journal verification range) to be compared and collated according to the transaction pattern of the transaction detail information.

The image recognition unit 11, the character recognition unit 12, the journal generation unit 13, the journal verification unit 14, and the like can use AI using various learning models by using the classifier 60. Specifically, the image recognition unit 11, the character recognition unit 12, the journal generation unit 13, the journal verification unit 14, and the like input input data to the classifier 60, and the output data output by the classifier 60 is used. can.

The classifier 60 has a predetermined function Y = f (X, W) = X × W adjusted by the weighting W, and when the input data X is input, the output data is output by the predetermined function f (X, W). Y is output.

In addition, the classifier 60 learns the feature quantity that quantifies the regularity for classification by machine learning the teacher data in advance (finds the commonalities and differences of the things to be classified from a large amount of teacher data. ) Is a learning model.

Inference (judgment) in which AI using a learning model generates judgment conditions (features are equal to or more than a predetermined threshold value or less than a predetermined threshold value) from input data and weighting, and extracts output data (judgment results) that satisfy the judgment conditions. Perform processing. Machine learning using supervised data (supervised learning) may use known methods including linear regression analysis by the least squares method and the most probable estimation method, a support vector machine (SVM), and a neural network (NN).

The judgment result is visually confirmed, and if the judgment result is incorrect, the judgment result is corrected. When the judgment result is corrected, the correction content is fed back to the learning model, and the correction content fed back by the learning model is learned (the weighting is adjusted so that the correction content becomes the output target), so that the AI is inferred (judgment). ) The processing accuracy can be improved. The learning method is described in supervised machine learning, but not only machine learning with or without supervised learning, but also known AI learning methods such as deep learning may be used.

The image recognition unit 11 has an image recognition AI using a classifier (image recognition learning model) in which the registered image data in the image DB 33, which will be described later, is machine-learned as teacher data. Image recognition AI performs image recognition processing using image similarity as a determination condition, and extracts transaction information from voucher data. Image similarity divides voucher data into multiple areas, and features the frequency of items of transaction information (voucher type, date, business partner, item, amount, consumption tax amount, payment method, etc.) existing in each area. It is a value, and the image similarity is calculated from the total number of items and the types of transaction information items existing in each area. Further, as the feature amount, the variation of the item of the transaction information or the distribution of the type of the item of the transaction information may be used. The image similarity may be calculated using a known statistical process such as a normal distribution.

The character recognition unit 12 has a character recognition AI using a classifier (character recognition learning model) in which text data in the transaction DB 34, which will be described later, is machine-learned as teacher data. The character recognition AI performs character recognition processing using the character similarity as a determination condition, and extracts transaction detail information line by line from the transaction information. The character recognition unit 12 generates text data based on all the extracted transaction detail information. The character similarity is a value characterized by the frequency of items of transaction detail information (voucher type, date, business partner, item, amount, consumption tax amount, payment method, etc.) that exist for each line of transaction information. The character similarity is calculated from the total of the item types and the number of items of the transaction detail information existing in each line of. Further, as the feature amount, the variation of the item of the transaction information or the distribution of the type of the item of the transaction information may be used. The character similarity may be calculated using a known statistical process such as a normal distribution.

The journal generation unit 13 has a journal generation AI using a classifier (journal generation generation learning model) in which registered journal data in the journal DB 31 described later is machine-learned as teacher data. Journal generation AI performs journal generation processing using journal similarity as a judgment condition, and generates journal data from transaction detail information. The journal similarity is a value characterized by the frequency of the selected account among the accounts (combination of debit and credit accounts) corresponding to the transaction item information, and is based on the type and frequency of occurrence of the selected account. Calculate journal similarity. Further, as the feature amount, the variation of the account item or the distribution of the type of the account item may be used. The journal similarity may be calculated using a known statistical process such as a normal distribution.

The journal verification unit 14 has a journal verification AI using a classifier (journal verification learning model) in which the registered journal data in the journal DB 31 and the erroneous journal data in the erroneous journal DB 32, which will be described later, are machine-learned as teacher data. Journal verification The journal verification process is performed on the result of comparison and collation between the reference journal data and the journal data by the journal verification AI using the journal normal value or the journal abnormal value as a determination condition.

The journal normal value is calculated from the type and occurrence frequency of each item of the correct journal data, using the frequency of each item of the correct journal data such as transaction detail information and accounts as a feature quantity. In addition, the variation of each item of the correct journal data or the distribution of the type of each item of the correct journal data may be used as the feature amount. The normal journal entry value may be calculated using a known statistical process such as a normal distribution.

In addition, the journal error value is characterized by the frequency of each item of journal data with errors in transaction details information, account items, etc. (including those whose data is clearly incorrect), and the type of each item of journal data with errors and Calculated from the frequency of occurrence. Further, the feature amount may be the variation of each item of the journal data with an error or the distribution of the type of each item of the journal data with an error. The journal outlier may be calculated using a known statistical process such as a normal distribution.

Examples of journal data with errors (including those with clearly incorrect data) include cases where the consumption tax amount is calculated at the normal tax rate (10%) for products subject to the reduced tax rate (8%) and items. Usually when the amount of money (for example, 100,000 yen) is abnormally large compared to the consumption tax amount (for example, 60 yen) or when the consumption tax amount (for example, 1000 yen) is abnormally large compared to the amount of the item (for example, 300 yen). It is an error (journal) that is more characteristic than the error (journal) of.

Further, the journal verification unit 14 has a journal verification range AI using a classifier (journal verification range learning model) in which registered journal data in the journal DB 31 described later is machine-learned as teacher data. Journal verification range AI performs journal verification confirmation processing using the verification confirmation value as a judgment condition to determine the items of journal data (date, debit account and credit account, amount, consumption tax amount, description, etc.).

The verification final value is characterized by the distribution (variation) of the journal data items (journal verification range) that have undergone journal verification processing among the journal data items that correspond to transaction patterns such as standard transactions in transaction detail information. It is a value, and the verification confirmation value is calculated from the frequency of occurrence of items in the journal data for which the journal verification process has been performed. Further, the verification final value may be calculated by using a known statistical process such as a normal distribution to calculate the variation of the items of the journal data for which the journal verification process has been performed. When registering the verification item value and verification confirmation value of each item of the journal data corresponding to the transaction pattern such as the standard transaction of the transaction detail information in the transaction pattern dictionary (not shown) described later and confirming it in the journal verification range. You may refer to.

The storage unit 30 can usually be realized from an MPU, a memory, or the like. The processing procedure of the storage unit 30 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

The storage unit 30 is a storage area for a hard disk, a semiconductor memory element, or the like for storing programs, data, and the like necessary for the control unit 10 to execute various processes.

Further, the storage unit 30 includes a program storage unit (not shown), a journal entry DB 31, an image DB 33, a transaction DB 34, an erroneous journal entry DB 32, and a learning model storage unit 35.

The program storage unit is a storage area for storing various programs. The program storage unit stores application programs for executing various functions such as accounting processing.

The journal DB 31 is a database that stores journal data (including corrected journal data) that is determined to be correct as a result of visual confirmation among the journal data that has been subjected to journal verification processing by the journal verification unit 14 as registered journal data. The journal data is generated by the journal generation unit 13 and is associated with the transaction details information and the voucher data.

The erroneous journal DB 32 is a database that stores the journal data determined to be a significantly erroneous journal among the journal data that has been subjected to the journal verification process by the journal verification unit 14 as erroneous journal data.

Journal data (registered journal data, incorrect journal data) includes journal code, date, account (debit account and credit account), amount, consumption tax amount, description, journal similarity, blank flag, double accounting flag, and transaction details. It consists of information and verification item values (date, debit account, credit account, amount, consumption tax amount, description), verification final value item, and image code item.

The journal code is a unique code that identifies the journal data. The date is the date of occurrence of the journal data and is the same as the date contained in the voucher data. Accounts are a general term for classification items necessary to clearly record the changes in assets, liabilities, and capital, and the nature of expenses and income generated by company transactions. The amount is the amount of consideration paid in the transaction, which is the debit or credit amount of a single journal. The amount of consumption tax is the amount of consumption tax.

The description is the transaction content such as the name, name, asset or service provision content of the other party of the transaction, and is a supplementary content to the account item. Originally, the description of the journal does not have to be described at all, but the taxable business operator of the consumption tax needs to describe the requirements for the description of the consumption tax in the description of the journal. The journal similarity is the frequency of occurrence of the selected account (including the combination of the debit account and the credit account) among the accounts (including the combination of the debit account and the credit account) corresponding to the transaction item information.

The blank flag is a flag to be set when there is a blank in the journal data item described later to check whether there is a blank. The double counting flag is a flag to be set when there is double counting of journals by checking whether there is double counting of journals described later (whether multiple journals with the same content are generated). The transaction details information is the transaction content that is the basis of the journal data.

The verification item values (date, debit account, credit account, amount, consumption tax amount, description) indicate each item (journal verification range) of the journal data for which journal verification processing is performed and the status (normal or abnormal, etc.) of each item. It is shown numerically. The verification confirmation value is the range of the journal data for which the journal verification process is performed (which item of the journal data is to be verified) for each item of the journal data (date, debit account, credit account, amount, consumption tax amount, description). Is determined and set as the optimum journal verification range according to the transaction pattern of the transaction detail information.

The image DB 33 is a database in which the image recognition unit 11 performs image recognition processing on the voucher data generated by the input unit 20 and stores the voucher data reflecting the transaction information which is the result of the image recognition processing as registered image data. ..

The voucher data (registered image data) consists of an image code, a voucher image, transaction information, and image similarity items. The image code is a unique code that identifies the voucher data. The voucher image is image data of the voucher converted into electronic data (for example, "receipt 20200403.JPG" or the like). The image similarity is the total number of transaction information items existing in each area in the area where the voucher data is divided into a plurality of areas.

Transaction information consists of transaction information items, the number of transaction information items, labels, and items in the character string information list. The transaction information item is the transaction information item of the voucher data such as voucher type, date, business partner, item, amount, consumption tax amount, payment method, etc., and the number of transaction information items is the number of transaction information items of the voucher data. The label indicates the image content of the voucher data. The character string information list is a multi-line character string information included in the voucher data, and character string information for the number of lines such as character string information 1, character string information 2, ..., Character string information line number is set. Has been done.

Items of transaction information include voucher type, date, business partner, item, amount, consumption tax amount, payment method, and the like.

"Voucher type" of transaction information indicates information on the type of voucher data such as slips, receipts, receipts and invoices.

The "date" of the transaction information indicates information related to the date (transaction date such as purchase date and issue date) written on the receipt or the like (voucher).

The "customer" of the transaction information is the issuer of the receipt (voucher) written on the receipt (voucher), the company name of the purchaser of the product, the store name, the contact information, the person in charge (clerk), etc. Shows information about business partners, including.

The "item" in the transaction information indicates information on an item such as a product (including a purchase content including a product or product code and a symbol (mark) indicating a reduced tax rate) described on a receipt or the like (voucher).

The "amount" of the transaction information indicates the amount of the item written on the receipt (voucher), characters such as the total amount (included in the total or subtotal), and information on the amount.

"Consumption tax amount" in transaction information includes characters such as consumption tax amount (including reduced tax rate) and total consumption tax amount of items listed on receipts (vouchers) (including 8% and 10% and products subject to reduced tax rate). And show information about the amount.

The "payment method" of the transaction information indicates information on the payment method described on the receipt (voucher) of cash, credit card, electronic money, etc. If the payment method is not described on the receipt, etc. (voucher), the payment method is determined from the information (information on change, card company, etc.) described on the receipt, etc. (voucher).

The transaction DB 34 is a database that stores text data as registered text data. The text data is data generated from transaction detail information obtained by character-recognizing voucher data (list of character string information of transaction information). Text data is created in association with voucher data.

The text data consists of text code, multi-line transaction detail information, character similarity and image code items. The text code is a unique code that identifies the registered text data. The character similarity is the total quantity of existing transaction detail information items for each line of transaction information.
The transaction detail information is composed of the transaction detail character string information, the transaction detail information item, and the number of transaction detail information items.

The transaction detail character string information is the character recognition result of the character string information of each line of the character string information list of the transaction information. The transaction details information item is the transaction information item of the voucher data such as voucher type, date, business partner, item, amount, consumption tax amount, payment method, etc., and the number of transaction details information items is the number of transaction details information items of the voucher data. be.

Items of transaction information details include voucher type, date, business partner, item, amount, consumption tax amount, payment method, and the like.

The "voucher type" of the transaction information statement is the result of character recognition processing of the "voucher type" of the transaction information, and indicates information on the type of voucher data such as slips, receipts, receipts and invoices.

The "date" of the transaction information statement is the result of character recognition processing of the "date" of the transaction information, and indicates information on the date (transaction date such as purchase date and issue date) described on the receipt or the like (voucher).

The "customer" in the transaction information statement is the result of character recognition processing for the "customer" in the transaction information, and the issuer of the receipt, etc. (voucher) described on the receipt, etc. (voucher) or the company name of the purchaser of the product, etc. Information about the business partner, including the name of the store, its contact information, and the name of the person in charge (clerk).

The "item" in the transaction information statement is the result of character recognition processing for the "item" in the transaction information, and indicates the item such as the product (purchase content including the product or product code and the reduced tax rate) described on the receipt etc. (voucher). Shows information about (including symbols).

The "amount" of the transaction information statement is the result of character recognition processing of the "amount" of the transaction information, and the characters (including in the total and subtotal) and the amount of the item such as the amount and the total amount described on the receipt etc. (voucher). Provides information about.

The "consumption tax amount" of the transaction information statement is the result of character recognition processing of the "consumption tax amount" of the transaction information, such as the consumption tax amount (including the reduced tax rate) and the total consumption tax amount of the items described on the receipt etc. (voucher). Shows information about characters (including 8% and 10%, products subject to reduced tax rate, etc.) and amount.

The "payment method" of the transaction information statement is the result of character recognition processing of the "payment method" of the transaction information, and indicates the information on the payment method described on the receipt (voucher) of cash, credit card, electronic money, etc. If the payment method is not described on the receipt, etc. (voucher), the payment method is determined from the information (information on change, card company, etc.) described on the receipt, etc. (voucher).

A database in which a learning model storage unit 35 is prepared and a predetermined learning model such as an image recognition learning model, a character recognition learning model, a journal generation learning model, a journal verification model, and a journal verification range learning model is stored.

The company DB (not shown) stores (company) information of the business operator subject to accounting processing using the company code as a key. The company DB is a company code, an organization code, a department code, a company name, an industry, an organization name, a department name, a type, a year, an address, an organization manager, a division, and a person in charge. Stores company (business) data consisting of items. For company (business) data, in addition to the case where the accounting office uses this accounting processing device 1 for multiple advisors, the company can use it between the head office and branch offices (not shown) or between the head office and branch offices (not shown). It is used when it is used between the head office and the sales offices of each base (not shown).

The company code is the identification information that identifies the adviser, the organization store code is the head office or the head office branch office of the company, and the department code is the identification information that identifies the department to which the company belongs. The type of business is the type of business of the company, such as manufacturing or retail. The company name is the name of the company (example: ABC industry, etc.), the organization name is the name of the company's base (example: ○ island branch office, etc.) such as the head office, branch office, sales office, etc., and the department name is the accounting department. It is the name of a department such as sales or sales. The type is the type of work such as accounting, taxation, and accounting. The address is the address of the organization of the company. The head of the organization is the name of the head of the organization of the company, such as the representative (eg, ○ mountain x man for ABC industry) at the head office. The classification is the classification of corporations and individuals. The person in charge is the person in charge of operations such as accounting.

In addition to each item described above, the company DB may store, for example, a representative name, a company logo, a DB registration date, a business form, a commissioned business, and the like. The representative name may be used, for example, to link the parent-child relationship of a company. The DB registration date corresponds to the advisory contract date and the branch establishment date. The business form is whether it is a corporate business, an individual business, or a mere individual. The entrusted business is accounting business (monthly audit, settlement of accounts, etc.) and tax affairs (corporate tax filing, blue filing, etc.).

The business partner DB (not shown) is a database that stores business partner data including information (customer information) about the business partner of the business operator. The business partner data is composed of items such as a business partner code, an industry, a business partner name, an address, a store name, a telephone number, a company logo, a person in charge, and a payment method. The product code, the item (product), etc. handled may be added to the customer data.

The business partner code is identification information that identifies the business partner. Information indicating the business partner of the company (business operator) is the type of business, the business partner name, the address, the store name, the telephone number, the company logo, the person in charge of the transaction, and the payment method to the business partner. These business partner information may be used as a partner for sales, purchases, accounts receivable, accounts payable, and the like.

Although the various functions of the control unit 10 described above are executed and described as an application program by one program, the various functions of the control unit 10 described above may be executed as a plurality of programs subdivided for each function.

The computer referred to in the present invention refers to an information processing device including a control unit, a storage device, and the like, and the accounting processing device 1 is an information processing device including a control unit 10, a storage unit 30, and the like. Included in the concept of computers.

<Processing of accounting processing device 1>
Next, the processing by the accounting processing apparatus 1 will be described.
FIG. 3 is a flowchart showing processing in the accounting processing apparatus 1 according to the present embodiment.
FIG. 4 is an example of voucher data.
FIG. 5 is an example in which the image recognition unit 11 divides the voucher data into a plurality of areas.
The process of the first embodiment will be described with a flowchart. Regarding another description example such as a modification of Example 1, only the difference from Example 1 is described, and the description of the common part is omitted.

A voucher (hereinafter referred to as a receipt) including a slip, a receipt, a receipt, an invoice, etc. is read by a mobile terminal such as a mobile phone or a digital camera, and an input unit 20 such as a scanner, and the input unit 20 is a receipt or the like (voucher). Is read to generate digitized voucher data (step S1, hereinafter simply referred to as "S1"). The control unit 10 receives the voucher data generated for each receipt or the like (voucher). For example, when there is one receipt or the like (voucher), there is one voucher data, and when there are a plurality of receipts or the like (voucher), there are a plurality of voucher data.

The image recognition unit 11 performs image recognition processing of the voucher data received by the control unit 10 (S2).
Specifically, as image recognition processing, voucher data is divided into a plurality of areas, and information on each item of transaction information (including voucher type, date, business partner, item, amount) and the number of transaction information items in each divided area. To get. "Transaction information items (voucher type, date, business partner, item, amount, consumption tax amount, payment method, etc.)" and "number of transaction information items" of each divided area are totaled. Calculate and acquire transaction information and image similarity from "transaction information items (voucher type, date, business partner, item, amount, consumption tax amount, payment method, etc.)" and "number of transaction information items" in each divided area. ..

For example, when the receipt of FIG. 4 is divided into nine areas, the areas A1, the area A2, the area A3, the area A4, the area A5, and the area A6 are arranged in the order from the upper left to the lower right as shown in FIG. It is divided into nine regions such as region A7, region A8, and region A9.

Further, the image recognition unit 11 performs image recognition processing on a region in which voucher data such as a receipt (voucher) is divided into nine as shown in FIG.

The transaction information item of area A1 (multiple settings if there are multiple items) is the area because the phone number (customer) of store A, the name of the cashier (customer), and the issue date (date) are described in area A1. "Business partner, business partner, date" is set for the transaction information item of A1, and 3 is set for the number of transaction information items in area A1.

In addition, the transaction information items in area A1 are set by excluding duplicate items (for example, since the items in "customer" are duplicated in area A1, the transaction information item in area A1 is "customer, date". The number of transaction information items in area A1 may be set to 2), or the item with the largest number of transaction information items may be set (since the item with the largest number is "customer", the transaction in area A1 is set. The information item may be set to "customer", and the number of transaction information items in area A1 may be set to 2).

Area A2 contains the receipt (voucher type), store A (customer name), Saiyocho store (customer store name), telephone number (customer), cashier code (customer), and issue date (date). Since it is described, the transaction information item in the area A2 is set to "voucher type, business partner, business partner, business partner, business partner, date", and the number of transaction information items in the area A2 is set to 6.

Since the store code (customer) of the store A is described in the area A3, the transaction information item of the area A3 is set to "customer", and the number of transaction information items of the area A3 is set to 1.

Since 30001 * (product code) etc. and subtotal (amount) etc. are described in area A4, "item, item, item, amount, item, item, item, amount" is set as the transaction information item in area A4. , The number of transaction information items in the area A4 is set to 8.

Since beef (product name) and the like are described in the area A5, "item, item, item, item, item, item" is set as the transaction information item in the area A5, and the number of transaction information items in the area A5 is 6. Set.

Since ¥ 5,400 (amount) etc. is described in the area A6, the transaction information item of the area A6 is set to "amount, amount, amount, amount, amount, amount, amount, amount", and the transaction in the area A6. The number of information items is set to 8.

Area A7 contains the total (amount), consumption tax (consumption tax amount), change (payment method), and "Products marked with * are subject to the reduced tax rate" (consumption tax amount). "Consumption tax amount, consumption tax amount, consumption tax amount, consumption tax amount, amount, consumption tax amount, payment method, payment method, consumption tax amount" are set as transaction information items, and 9 is set as the number of transaction information items in area A7.

Since "Products marked with * are products subject to the reduced tax rate" (consumption tax amount) are described in area A8, "consumption tax amount" is set for the transaction information item in area A8, and the number of transaction information items in area A8 is 1 is set.

Area A9 contains ¥ 16,800 (amount), ¥ 1,522 (consumption tax), ¥ 3,200 (payment method), and "Products marked with * are subject to the reduced tax rate." (Consumption tax). Therefore, "consumption tax amount, consumption tax amount, consumption tax amount, consumption tax amount, amount, consumption tax amount, payment method, payment method, consumption tax amount" are set as the transaction information items in area A9, and the number of transaction information items in area A9 is 9. Is set.

The image similarity, which is the total number of transaction information items for each area in which the voucher data is divided into a plurality of areas, is totaled, and the one with the highest image similarity is preferentially extracted.

The image similarity, which is the total number of transaction information items in areas A1 to A9, is 51 (quantity of voucher type 1 + quantity of business partner 7 + quantity of date 2 + quantity of item 12 + quantity of amount 12 + quantity of consumption tax 13 + quantity of payment method 4 = 51).

Further, the image similarity may be set after weighting the important items in the transaction information items. For example, the weighting set by the user may be reflected in the item of transaction information.

Using the calculated image similarity as a search key, matching (comparison / collation) processing is performed on the image DB, and the registered image data having the closest image similarity is extracted as the similar registered image data from the matching result (S3). Transaction information (label) is acquired from the extracted similar registered image data, and the image recognition unit 11 determines voucher data (image recognition) from the acquired transaction information (label).
The image recognition process may be performed by using a known method.

When the judgment of the voucher data is completed, the character string information of the voucher data is read line by line and saved in the character string information list part of the transaction information described later. When the character string information is read, the character recognition (text conversion) process described later may be performed at the same time. Therefore, the image recognition unit 11 can acquire the arrangement of the transaction information on the voucher data and the contents of the transaction information by the image recognition process.

Registered image data 1 (label is a slip) having an image similarity of 78, registered image data 2 (label is a receipt) having an image similarity of 52, and registered image data 3 (label is a slip) having an image similarity of 45 in the image DB. If there is an invoice (not shown), the registered image data 2 (image similarity is 52, label is receipt) and the registered image data 1 are in order of closeness to the image similarity 51 of the voucher data. The registered image data 2 (image similarity is 52, label is receipt) having the closest image similarity among (image similarity is 45 and label is invoice) is extracted from the image DB as similar registered image data. The images are extracted in the order of close image similarity (similarity), but when the difference between the two similarities is the same, for example, in the case of image similarity 54 and character similarity 48, the order of image similarity is high (characters). The similarity 54 may be prioritized).

Transaction information (label "receipt") is acquired from the registered image data 2 (image similarity is 52, label is receipt) having the closest image similarity among the extracted registered image data, and the image recognition unit 11 acquires the acquired transaction. From the information, it is determined that the voucher data is a receipt image (image recognition).

Further, the image recognition unit 11 may have an image recognition AI using an image recognition learning model, and the image recognition AI may perform an image recognition process. When the receipt of FIG. 4 is divided into nine regions as shown in FIG. 5, the image similarity is 51. Therefore, the image recognition AI is a registered image having a threshold close to the image similarity 51 with the image similarity as a determination condition. From the data 2 (image similarity is 52, label is receipt), the image recognition AI infers (determines) that the voucher data is the receipt image.

The display unit 40 displays the voucher data and the image recognition processing result (transaction information) side by side, and makes a comparison by visual confirmation (S4). If there is a difference between the image recognition processing result and the voucher data, the transaction information is corrected.

After confirming the image recognition processing result, set "receipt" in the item of the transaction information label of the voucher data, and set the character string information of a plurality of lines described later in the character string information list of the transaction information, and update the voucher data. The updated voucher data is registered in the image DB 33 as registered image data.

For example, the image code is P001 and the voucher image is receipt 20200403. If the label is a receipt and character string information 1 "receipt", character string information 2 "store A", ... character string information 23 "Products marked with * are products subject to the reduced tax rate", the registered image data will be Image code "P001" and voucher image "Receipt 20200403.JPG" and transaction information (combination part of transaction information item and number of transaction information items) "Voucher type 1, date 2, customer 7, item 12, amount 12, consumption tax amount 13. Payment method 4 "and transaction information (label part)" Receipt "and transaction information (character string information list part)" Receipt, store A, ..., products marked with * are eligible for reduced tax rate. " It is composed of degrees "51".

Next, the character recognition unit 12 performs character recognition processing on the transaction information (character string information portion of a plurality of lines) of the voucher data (S5), and generates text data (composed of transaction details information) (S6). ). The transaction detail information is a character recognition result for the character string information portion of a plurality of lines of the transaction information.

Specifically, the character string information part of multiple lines of transaction information is subjected to character recognition (text conversion) processing for each line, and the character recognition result is converted into transaction detail character string information, and the transaction information item judged from the character recognition result. The content is set in the transaction detail information item, and the number of transaction information items determined from the character recognition result is set in the number of transaction detail information items. The character similarity is calculated from the number of transaction details information items in each line. When character recognition (text conversion) processing is performed together with the image recognition processing, the character string information of multiple lines from the character string information list part of the transaction information is directly converted into the transaction detail character string information of the transaction detail information of each line. Set.

The transaction information of the voucher data includes a multi-line character string information part in the character string information. For example, the character string information list of the transaction information in the case of the receipt of FIG. 4 is a multi-line character string information in order from the top. Column information 1 "receipt", character string information 2 "store A", character string information 3 "Saiyocho store S0048", character string information 4 "TEL 03-111-222", character string information 5 "registration charge U0011" , Character string information 6 "Thursday, July 25, 2019 12:15", Character string information 7 "30001 * Beef ¥ 5,400", Character string information 8 "30002 * Wheat flour ¥ 2,160", Character string Information 9 "30003 * Shoyu ¥ 340", character string information 10 "subtotal ¥ 7,900", character string information 11 "40001 stationery ¥ 200", character string information 12 "40002 beer ¥ 3,300", character string information 13 "4003 Japanese sake ¥ 5,400", character string information 14 "subtotal ¥ 8,900", character string information 15 "(8% target ¥ 7,900)", character string information 16 "(consumption tax 8% ¥ 632)" , Character string information 17 "(10% target ¥ 8,900)", character string information 18 "(consumption tax 10% ¥ 890)", character string information 19 "total ¥ 16,800", character string information 20 " (Internal consumption tax, etc. ¥ 1,522) ”, character string information 21“ custody ¥ 20,000 ”, character string information 22“ change ¥ 3,200 ”, character string information 23“ * -marked products are subject to the reduced tax rate It is a product. "

The character recognition result "receipt" of the character string information on the first line of the transaction information is a voucher. Therefore, the transaction detail character string information of line number 1 (first line) is "receipt", the transaction detail information item of line number 1 is "voucher type", and the number of transaction detail information items of line number 1 is 1.

The character recognition result "store A" of the character string information in the second line of the transaction information is the business partner name. Therefore, the transaction detail character string information of line number 2 is "store A", the transaction detail information item of line number 2 is "customer", and the number of transaction detail information items of line number 2 is 1.

The character recognition result of the character string information on the third line of the transaction information "Saiyocho store S0048" is a combination of "Saiyocho store" (store name) and "S0048" (store code). Becomes a business partner. Therefore, the transaction detail character string information of line number 3 is "Saiyocho store, S0048", the transaction detail information item of line number 3 is "customer, business partner", and the number of transaction detail information items of line number 3 is 2. ..

The character recognition result "TEL 03-111-222" of the character string information on the fourth line of the transaction information is a combination of "TEL" (telephone mark) and "03-111-222" (telephone number). Both phone numbers are business partners. Therefore, the transaction detail character string information of line number 4 is "TEL, 03-111-222", the transaction detail information item of line number 4 is "customer, business partner", and the number of transaction detail information items of line number 4 is. It becomes 2.

The character recognition result of the character string information on the fifth line of the transaction information "Registration charge U0011" is a combination of "Registration charge" (business name) and "U0011" (Registration staff name). Both become business partners. Therefore, the transaction detail character string information of line number 5 is "registrant, U0011", the transaction detail information item of line number 5 is "business partner, business partner", and the number of transaction detail information items of line number 5 is 2. ..

The character recognition result of the character string information on the sixth line of the transaction information "July 25, 2019 (Thursday) 12:15" is "July 25, 2019 (Thursday) 12:15" is the (transaction) date. .. Therefore, the transaction detail character string information of line number 6 is "July 25, 2019 (Thursday) 12:15", the transaction detail information item of line number 6 is "date", and the number of transaction detail information items of line number 6 is Is 1.

Character recognition result of the character string information on the 7th line of transaction information "30001 * beef ¥ 5,400" is "30001 *" (product code), "beef" (product name) and "¥ 5,400" (amount) The product code and product name are both items. Therefore, the transaction detail character string information of line number 7 is "30001 *, beef, ¥ 5,400", the transaction detail information item of line number 7 is "item, item, amount", and the transaction detail information item of line number 7 is. The number will be three.

In the embodiment, "30001 *" is used, but different items such as "30000" (product code) and "*" (subject to reduced tax rate) may be used.

The character recognition result "30002 * flour ¥ 2,160" of the character string information on the eighth line of the transaction information is a combination of "30002 *", "flour" and "¥ 2,160". Therefore, the transaction detail character string information of line number 8 is "30002 *, flour, ¥ 2,160", the transaction detail information item of line number 8 is "item, item, amount", and the transaction detail information item of line number 8 is. The number will be three.

The character recognition result "30003 * salty soy sauce ¥ 340" of the character string information on the ninth line of the transaction information is a combination of "30003 *", "salty soy sauce", and "¥ 340". Therefore, the transaction detail character string information of line number 9 is "30003 *, soy sauce, ¥ 340", the transaction detail information item of line number 9 is "item, item, amount", and the number of transaction detail information items of line number 9 is. It becomes 3.

The character recognition result "subtotal ¥ 7,900" of the character string information on the 10th line of the transaction information is a combination of "subtotal" (amount) and "¥ 7,900" (amount), both of which are amounts. Therefore, the transaction detail character string information of line number 10 is "subtotal, ¥ 7,900", the transaction detail information item of line number 10 is "amount, amount", and the number of transaction detail information items of line number 10 is 2. ..

The character recognition result of the character string information on the 11th line of the transaction information "40001 stationery ¥ 200" is a combination of "40001" (product code), "stationery" (product name) and "¥ 200" (amount), and is a product. Both the code and the product name are items. Therefore, the transaction detail character string information of line number 11 is "40001, stationery, ¥ 200", the transaction detail information item of line number 11 is "item, item, amount", and the number of transaction detail information items of line number 11 is 3. It becomes.

The character recognition result "40002 beer ¥ 3,300" of the character string information on the 12th line of the transaction information is a combination of "40002", "beer" and "¥ 3,300". Therefore, the transaction detail character string information of line number 12 is "40002, beer, ¥ 3,300", the transaction detail information item of line number 12 is "item, item, amount", and the number of transaction detail information items of line number 12 Is 3.

The character recognition result "4003 sake ¥ 5,400" of the character string information on the 13th line of the transaction information is a combination of "4003", "sake" and "¥ 5,400". Therefore, the transaction detail character string information of line number 13 is "4003, sake, ¥ 5,400", the transaction detail information item of line number 13 is "item, item, amount", and the number of transaction detail information items of line number 13 Is 3.

The character recognition result "subtotal ¥ 8,900" of the character string information on the 14th line of the transaction information is a combination of "subtotal" and "¥ 8,900". Therefore, the transaction detail character string information of line number 14 is "subtotal, ¥ 8,900", the transaction detail information item of line number 14 is "amount, amount", and the number of transaction detail information items of line number 14 is 2.

The character recognition result "(8% target ¥ 7,900)" of the character string information on the 15th line of the transaction information is "(" and "8% target" (consumption tax rate) and "¥ 7,900" (consumption tax amount). ) And ")", and "(" and ")" are information irrelevant to the transaction content, so both the consumption tax rate and the consumption tax amount are consumption tax amounts. Therefore, the transaction detail character string information of line number 15 is "8% target, ¥ 7,900", the transaction detail information item of line number 15 is "consumption tax amount, consumption tax amount", and the number of transaction detail information items of line number 15 Is 2.

The character recognition result "(consumption tax 8% ¥ 632)" of the character string information on the 16th line of the transaction information is "(" and "consumption tax 8%" (consumption tax rate) and "¥ 632" (consumption tax amount). Since it is a combination of ")" and "(" and ")" are information irrelevant to the transaction content, both the consumption tax rate and the consumption tax amount are the consumption tax amount. Therefore, the transaction detail character string information of line number 16 is "consumption tax 8%, ¥ 632", the transaction detail information item of line number 16 is "consumption tax amount, consumption tax amount", and the number of transaction detail information items of line number 16 is. It becomes 2.

The character recognition result "(10% target ¥ 8,900)" of the character string information on the 17th line of the transaction information is "(", "10% target", "¥ 8,900" (consumption tax amount) and "). Is a combination of. Therefore, the transaction detail character string information of line number 17 is "10% target, ¥ 8,900", the transaction detail information item of line number 17 is "consumption tax amount, consumption tax amount", and the number of transaction detail information items of line number 17 Is 2.

The character recognition result "(consumption tax 10% ¥ 890)" of the character string information on the 18th line of the transaction information is a combination of "(", "consumption tax 10%", "¥ 890", and ")". Therefore, the transaction detail character string information of line number 18 is "consumption tax 10%, ¥ 890", the transaction detail information item of line number 18 is "consumption tax amount, consumption tax amount", and the number of transaction detail information items of line number 18 is. It becomes 2.

The character recognition result "total ¥ 16,800" of the character string information on the 19th line of the transaction information is a combination of "total" (total) and "¥ 16,800" (total amount). Is also an amount. Therefore, the transaction detail character string information of line number 19 is "total, ¥ 16,800", the transaction detail information item of line number 19 is "amount, amount", and the number of transaction detail information items of line number 19 is 2. ..

The character recognition result of the character string information on the 20th line of the transaction information "(consumption tax, etc. ¥ 1,522)" is "(" and "consumption tax, etc." (consumption tax) and "¥ 1,522" (consumption). Since it is a combination of tax amount) and ")" and "(" and ")" are information irrelevant to the transaction content, both consumption tax and consumption tax amount are consumption tax amounts. Therefore, the transaction detail character string information of line number 20 is "consumption tax, etc., ¥ 1,522", the transaction detail information item of line number 20 is "consumption tax amount, consumption tax amount", and the transaction detail information item of line number 20. The number is 2.

The character recognition result of the character string information on the 21st line of the transaction information "Custody ¥ 20,000" is a combination of "Custody" (payment method) and "¥ 20,000" (payment amount). Both payment amounts are payment methods. Therefore, the transaction detail character string information of line number 21 is "deposit, ¥ 20,000", the transaction detail information item of line number 21 is "payment method, payment method", and the number of transaction detail information items of line number 21 is. It becomes 2.

The character recognition result of the character string information on the 22nd line of the transaction information "change ¥ 3,200" is a combination of "change" (payment method) and "¥ 3,200" (payment amount), and the payment method and payment amount Are both payment methods. Therefore, the transaction detail character string information of line number 22 is "change, ¥ 3,200", the transaction detail information item of line number 22 is "payment method, payment method", and the number of transaction detail information items of line number 22 is 2. It becomes.

The character recognition result of the character string information on the 23rd line of the transaction information "Products marked with * are products subject to the reduced tax rate." (Reduced tax rate) is information related to the reduced tax rate, and the reduced tax rate is the consumption tax amount. Therefore, the transaction detail character string information of line number 23 is "Products marked with * are products subject to the reduced tax rate.", The transaction detail information item of line number 23 is "consumption tax amount", and the number of items of transaction detail information 23 is 1. It becomes.

Character similarity (quantity of voucher type 1 + quantity of business partner 7 + quantity of date 1 + quantity of item 12 + quantity of amount 12 + quantity of consumption tax 11 + quantity of payment method 4) = 48), so it becomes 48.

Further, for important items of transaction detail information, the number of transaction detail information items in each line may be weighted, and then the character similarity may be set. For example, the weighting set by the user is reflected in the item of transaction detail information.

For example, if the weighting count is set to 1.1 for the business partner, 2.0 for the item, 1.5 for the amount, and 0.3 for the date (other items are 1.0), the text data. The character similarity, which is the total number of transaction details information items in each line, is 66 (quantity of voucher type 1 x weighting 1.0 + quantity of customer 7 x weighting 1.1 + date quantity 1 x weighting 0.3 + Item quantity 12 x weighting 2.0 + amount quantity 12 x weighting 1.5 + consumption tax quantity 11 x weighting 1.0 + payment method quantity 4 x weighting 1.0 = 66).

The character recognition unit 12 generates text data based on the transaction detail information of each line, which is the result of the character recognition process. The text data is composed of multiple lines of transaction detail information and is generated in association with the voucher data. Calculate the character similarity, which is the total number of transaction details information items in each line of text data.

Using the calculated character similarity as a search key, matching (comparison / collation) processing is performed on the transaction DB, and the registered text data having the closest character similarity is extracted as the similarity registered text data from the matching result.

When there are registered text data 1 having a character similarity of 51, registered text data data 2 having a character similarity of 72, and registered text data data 3 having a character similarity of 43 in the transaction DB (not shown), the text. Since the registered text data 1 (character similarity 51) and the registered text data 3 (character similarity 43) are hit in the order of the character similarity to the character similarity 48 of the data, the characters of the two registered text data are hit. The registered text data 1 (character similarity 51) having the closest similarity is extracted from the transaction DB as the similar registered text data.
The characters are extracted in the order of closeness (similarity), but when the difference between the two similarities is the same, for example, in the case of character similarity 51 and character similarity 45, the order of character similarity is high (characters). The similarity 51 may be prioritized).

Further, the character recognition unit 12 has a character recognition AI, and the character recognition AI may perform character recognition processing. If the registration text data is "T001," 2019/6/25, 30001 *, beef, ¥ 5,600, ¥ 432 "," date, item, item, amount, consumption tax amount ", 5,50", the character string Assuming that the information is "2019/7/25 30001 * beef ¥ 5,600 ¥ 432", the character recognition AI is registered text data having a threshold close to the character similarity 48 with the character similarity as a judgment condition (character similarity is 50). Character recognition AI is text data "T001, (2019/7/25, 30001 *, beef, ¥ 5,600, ¥ 432), (date, item, item, amount, consumption tax amount)" , 5, 48 "is inferred (determined).

The display unit 40 displays the voucher data, the text data, and the similar registered text data side by side, and makes a comparison by visual confirmation (S7).

In this embodiment, the voucher data, the text data, and the registered text data are displayed side by side, but the text data and the registered text data may be displayed side by side for comparison by visual confirmation.

Further, before the visual confirmation, a comparison and collation (matching) process of the similar registered text data and the text data may be performed. Matching processing is performed between the extracted similar registered text data and the text data, and when there is a mismatched portion (difference) from the matching result, the display unit 40 highlights the mismatched portion of the text data. The highlighting may be a display different from the normal display, and the display format is not particularly limited. For example, underlining the inconsistent part, making the character color different, thickening the character, or near the inconsistent part. Express by displaying a flag or the like. Further, a plurality of highlighting types may be set according to the consistency, or only one type may be set.

If there is an error as a result of visual confirmation, correct the text data. Register the text data as registered text data in the transaction DB 34.

For example, the value of each item of the transaction detail information item is the value of the "date" item of the transaction detail information item. The value of the item "Account" of the information item is the customer information, and the value of the item of "Item" of the transaction item information item is the item of the transaction item information item. The character string to be judged is the item information, and the value of the "amount" item of the transaction detail information item is the amount information of the character string to be judged by the transaction detail information item as the amount. For the value, the character string that the transaction detail information item determines as the consumption tax amount is the consumption tax amount information, and for the value of the "payment method" item of the transaction detail information item, the character string that the transaction detail information item determines as the payment method is the payment method information. And.

The transaction detail information includes transaction detail information (hereinafter referred to as journal generation transaction detail information) used to generate journal data such as transaction detail information 7 to 9 and transaction detail information 11 to 13, and other (journals). There are two types of transaction detail information (hereinafter referred to as auxiliary transaction detail information) that are not used to generate data. The text data is composed of journal entry generation transaction detail information and auxiliary transaction detail information.

Date information, business partner information, consumption tax amount information, and payment method information may not be set in the transaction details information for journal generation. In that case, the date information, the business partner information, the consumption tax amount information, the payment method information, etc. are acquired from the auxiliary transaction detail information including the date information, the business partner information, the consumption tax amount information, the payment method information, etc., and the journal is generated. Set in transaction details information.

The journal generation unit 13 acquires date information, business partner information, consumption tax amount information, payment method information, etc. from the auxiliary transaction detail information, and generates journals for the acquired date information, business partner information, consumption tax amount information, payment method information, etc. By setting in the transaction detail information for use (hereinafter, transaction detail information supplement processing), the transaction detail information necessary for generating the journal data is prepared. When a large number of journals are generated, the character recognition unit 12 may perform the transaction detail information replenishment process instead of the journal generation unit 13 for load balancing.

Consumption tax information requires consumption tax rate (two types of reduced tax rate and normal tax rate) and tax display information (two types of tax-excluded and tax-included). In addition, since the required consumption tax amount information (reduced tax rate or normal tax rate) differs depending on the transaction detail information for journal generation (item information, etc.), a plurality of consumption tax amount information is required.

The consumption tax rate (8% and 10%) and the tax display information (excluding tax) are acquired from a plurality of auxiliary transaction detail information (for example, from transaction detail information 15 to transaction detail information 18). Consumption tax amount information (for example, transaction detail character string information of line number 16 "consumption tax 8%, ¥ 632", transaction detail character string information of sentence line number 17 "10% target, ¥ 8,900", transaction of line number 18 Obtain the consumption tax rate (8% and 10%) and tax display information (excluding tax) from the detailed character string information "consumption tax 10%, ¥ 890").

From the acquired consumption tax amount information, the consumption tax rate (reduced tax rate or normal tax rate) and tax display information will be "tax excluded". Also, since the item information (beef) is food, the consumption tax rate is the reduced tax rate (for example, 8%), and by multiplying the amount information (¥ 5,400) by the reduced tax rate (8%), the consumption tax is ¥ 432 (for example). Consumption tax amount information) is calculated.

The necessary consumption tax amount information (consumption tax rate and tax display information) is determined from the item information "beef" of the transaction detail information for journal generation (for example, transaction detail information 7), and the transaction detail information for journal generation (for example, transaction detail information 7) is determined. The consumption tax amount information "¥ 432" is calculated from the amount information "¥ 5,400" and the consumption tax amount information (reduced tax rate and tax excluded) using the calculation formula of the consumption tax amount described later.

Set the tax display information (excluding tax), reduced tax rate (8%), and ¥ 432 (consumption tax amount) acquired as the consumption tax amount information of the transaction detail information for journal generation (for example, transaction detail information 7) (line number 7). "Excluding tax, 8%, ¥ 432" is additionally set in the transaction detail character string information, and "consumption tax amount, consumption tax amount, consumption tax amount" is additionally set in the transaction detail information item of line number 7).

The date information "July 25, 2019 (Thursday) 12:15" is acquired from the auxiliary transaction detail information (transaction detail information 6). Since the date information "July 25, 2019 (Thursday) 12:15" acquired by the journal generation unit 13 includes the day and time, the date information "July 25, 2019" converted only to the date. Is set as the date information of the transaction detail information for journal generation (for example, transaction detail information 7) (transaction detail information character string "July 25, 2019" of line number 7 and transaction detail information item "date" of line number 7. Is additionally set).

Acquire the customer information "Store A" from the auxiliary transaction detail information (transaction detail information 2). The journal generation unit 13 sets the acquired supplier information "Store A" as the supplier information of the transaction detail information for journal generation (for example, transaction detail information 7). If the customer name cannot be obtained directly from the auxiliary transaction detail information, refer to the customer DB (not shown) to obtain the customer name as the customer information, and then obtain the transaction detail information for journal generation (for example, transaction detail information 7). ) Is set as the customer information (the transaction detail information character string "store A" of line number 7 and the transaction detail information item "customer" of line number 7 are additionally set).

The payment method information may not be specified in the journal generation transaction detail information or the auxiliary transaction detail information, and the payment judgment dictionary (not shown) described later is used from the payment method information obtained from the plurality of auxiliary transaction detail information. The payment method information obtained by reference is set in the transaction detail information for journal generation (for example, transaction detail information 7).

The payment judgment dictionary (not shown) includes one or more keywords ("deposit", "other payment", "change", "change", etc.) and payment method (cash, credit card, electronic money, etc.) described on the receipt, etc. It is a dictionary associated with (or postpay, etc.).

Since there is no payment method in the transaction detail information item of line number 7, transaction detail information 21 (transaction detail character string information of line number 21 "Custody, ¥ 20" has transaction detail information with payment method information from all lines of text data. From the transaction detail information item "payment method, payment method" of line number 21 and ", 000", the payment method information of transaction detail information 21 (transaction detail character string information "keeping" of line number 21) and transaction detail information 22 ( From the transaction details character string information "change, ¥ 3,200" of line number 22 and the transaction details information item "payment method, payment method" of line number 22, the payment method information of transaction detail information 22 (transaction details of line number 22) Extract the character string information "change").

Payment method information "deposit and change" extracted from a plurality of auxiliary transaction detail information (for example, transaction detail information 21 and transaction detail information 22) (payment method information "deposit" of transaction detail information 21 and transaction detail information 22) Obtain payment method information "change"). The payment judgment dictionary is referred to based on the payment method information "deposit", and three payment methods "cash, credit card, electronic money" corresponding to the payment method information "deposit" are extracted. The payment method "cash" corresponding to the payment method information "change" is extracted by referring to the payment judgment dictionary based on the payment method information "change".

Since "cash" is the only payment method that satisfies the extracted payment method information "deposit and change", it can be confirmed that the payment method is "cash". Set "cash" in the information (additionally set the transaction detail information character string "cash" of line number 7 and the transaction detail information item "payment method" of line number 7).

When the payment method is specified by the business partner, the payment method is acquired by referring to the business partner DB (not shown) instead of the payment method acquired in the payment determination dictionary.

Further, an item dictionary (not shown) in which the item information and the classification are associated may be prepared by setting the classification in which the item information having the same characteristics is grouped. Further, the consumption tax rate and the classification may be associated and registered in the item dictionary (not shown).

When the acquired transaction detail information item is an item, the item information (character string for determining that the transaction detail information item is an item in the transaction detail character string information) described later refers to the item dictionary (not shown). Obtain the classification (information) that matches the item information.

For example, by registering food (classification information) and beef (item information) in the item dictionary (not shown) in association with each other, food (classification information) and consumption tax rate can be associated in the item dictionary (not shown). For example, for beef (item information), the consumption tax rate for each item can be found by referring to the item dictionary (not shown).

For example, in the case of the receipt of FIG. 4, the seventh line of the text data is the transaction detail information 7 (transaction detail character string information "30001 *, beef, ¥ 5,400" of line number 7 and the transaction detail information item of line number 7. "Item, item, amount").

When using the item dictionary, transaction detail information 7 (transaction detail character string information "30001 *, beef, ¥ 5,400, groceries in line number 7" and transaction detail information item "item, item, in line number 7" Amount, classification ").

Since the date information "July 25, 2019", the customer information "Store A", the payment method information "cash", and the consumption tax amount information "tax excluded, 8%, ¥ 432" are added, the transaction details information before the update. 7 (transaction detail character string information "30001 *, beef, ¥ 5,400" in line number 7 and transaction detail information item "item, item, amount" in line number 7) is the updated transaction detail information 7 (line number). Transaction detail information item of line number 7 with transaction detail character string information "30001 *, beef, ¥ 5,400, groceries, tax excluded, 8%, ¥ 432, 2019/7/25, store A, cash" Updated to "Item, Item, Amount, Classification, Consumption Tax Amount, Consumption Tax Amount, Consumption Tax Amount, Date, Business Partner, Payment Method").

The journal generation unit 13 performs journal processing from the transaction detail information included in the text data by referring to the account dictionary (not shown) described later or using the journal generation AI, and generates journal data (not shown). (S8).

The account dictionary (not shown) consists of at least account, lending, item, payment method, and journal similarity items. Register the item (including classification) and payment method of transaction detail information in the account dictionary in association with the account and balance. Journal similarity is set for each account (including combinations of accounts).

Among the accounts corresponding to the transaction detail information, the accounts with high journal similarity are preferentially extracted. Further, when there are a plurality of accounts corresponding to the transaction details information, the frequently used accounts may be weighted and then the journal similarity may be set.

Specifically, transaction detail information is acquired from text data, and the account (including the combination of accounts) corresponding to the transaction detail information is selected by referring to the account dictionary (not shown) to generate journal data. do.

The journal generation unit 13 uses transaction detail information 7 (transaction detail character string information “30001 *, beef, ¥ 5,400, groceries, tax excluded, 8%, ¥ 432, 2019/7/25, store A” of line number 7. , Cash ”and the transaction statement information item“ Item, Item, Amount, Classification, Consumption Tax Amount, Consumption Tax Amount, Consumption Tax Amount, Date, Business Partner, Payment Method ”) to the journal data“ 2019/7/25 (Date) ), NUML (debit account), NULL (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax), NULL (summary), NULL (journal similarity), NULL (blank flag), NUML (Double accounting flag), transaction detail information 7 (transaction detail information), NUML (verification item value date), NUML (verification item value debit account), NUML (verification item value credit account), NULL (verification item value) Amount), NUML (verification item value consumption tax amount), NULL (verification item value summary), NULL (verification confirmed value) ”is generated.

The journal generation unit 13 acquires the classification of the "product code" by referring to the item dictionary (not shown) for "30001 *" (item information). For "30001 *" (item information) and "product code" (classification), refer to the account dictionary (not shown), and since there is no corresponding account, no account is set.

For "beef" (item information), refer to the item dictionary (not shown) to obtain the classification of "food". For "beef" (item information), "food" (classification), and "cash" (payment method information), refer to the account dictionary (not shown), and for the corresponding accounts "purchase" and credit "debit". Obtain the combination and account item "cash" and debit "credit" combination and journal similarity "0.6".
If it is not possible to narrow down the account items corresponding to the transaction details information, the business type (for example, restaurant business, retail business, etc.) and journal similarity may be added as conditions.

The journal generation unit 13 debits the journal data from the combination of the acquired account "purchase" and the debit "debit", and the journal data from the combination of the account "cash" and the debit "credit". Set "Cash" for the credit account and "0.6" for the journal similarity.

The description is generated by referring to the description dictionary described later, or when creating a new one, according to a certain rule (summary setting rule) described later.

The journal generation unit 13 sets the description of the journal data by referring to the description dictionary (not shown) from the transaction detail information. The description dictionary is a dictionary in which the description corresponding to the transaction detail information is registered. For example, when the transaction statement character string information including "beef", "8%" and "store A" and the payment method of "store A" are specified as "cash", the description "store A / beef" is added to the description dictionary. -Cash / reduced tax rate compatible items "are associated.

Transaction detail information 7 (transaction detail character string information of line number 7 "30001 *, beef, ¥ 5,400, groceries, tax excluded, 8%, ¥ 432, 2019/7/25, store A, cash" Description from the transaction statement information item "Item, item, amount, classification, consumption tax amount, consumption tax amount, consumption tax amount, date, business partner, payment method", transaction statement information item number 10) of number 7, "Store A, beef, cash"・ Acquire "Items compatible with reduced tax rates".

Further, the journal generation unit 13 refers to the transaction pattern dictionary (not shown), and when the transaction detail information 7 corresponds to a transaction pattern such as a standard transaction, the verification item value of each item acquired from the transaction pattern dictionary ( Set the date, debit account, credit account, amount, description) and verification confirmation value as the verification item value (date, debit account, credit account, amount, description) and verification confirmation value of each item of the journal data. .. If there is no transaction pattern corresponding to the transaction pattern dictionary, set the verification item value (date, debit account, credit account, amount, description) and verification confirmation value of each item of the journal data to 1.0.

The transaction pattern dictionary (not shown) is the verification item value (date, debit account, credit account, date, debit account, credit account, etc.) of each item of the journal data, which is the journal verification range corresponding to the transaction pattern (transaction detail information) such as standard transactions. It is a dictionary that registers the amount of money, description) and the verification confirmation value.

For example, in the transaction pattern dictionary, the transaction pattern "newspaper transaction" and transaction detail information 30 (transaction statement character string information "newspaper fee, ¥ 3,000, direct debit", transaction detail information item "item, amount, payment method" , Transaction detail information item 30 (transaction detail information), 1.0 (verification item value date), 2.0 (verification item value debit account), 2.0 (verification item value credit account), 2.0 ( It is assumed that the verification item value amount), 2.0 (verification item value consumption tax amount), 2.0 (verification item value summary), and 1.5 (verification confirmed value) are associated.

Journal data created from transaction statement information 30 "NUML (date), newspaper book expenses (debit account), ordinary deposit (credit account), ¥ 3,000 (amount), ¥ 240 (consumption tax), ○○ Newspaper, tax excluded, 8%, direct debit (summary), 0.8 (journal similarity), NUML (blank flag), NUML (double account flag), transaction statement information 30 (transaction statement information), 1. 0 (verification item value date), 2.0 (verification item value debit account), 2.0 (verification item value credit account), 2.0 (verification item value amount), 2.0 (verification item value consumption) Tax amount), 2.0 (verification item value summary), 1.5 (verification confirmed value) ”, and the journal verification range is the account item, amount, and consumption tax amount that satisfy the judgment condition 1.5 (verification confirmed value) or more. It becomes the item of.

Further, since the transaction detail information 7 is not a transaction pattern corresponding to the transaction pattern dictionary, the journal generation unit 13 uses the transaction detail information 7 to enter the journal data "2019/7/25 (date), purchase (debit account), cash". (Credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax amount), store A (summary), 0.6 (journal similarity), NUML (blank flag), NULL (double accounting flag), Transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit account), 1.0 (verification item value credit account), 1.0 (verification item) Value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value summary), 1.0 (verification confirmed value) ”is generated.

Further, the journal generation unit 13 may have a journal generation AI, and the journal generation AI may perform the journal generation process. For example, transaction detail information 7 (transaction detail character string information "30001 *, beef, ¥ 5,400" in line number 7, transaction detail information item "item, item, amount" in line number 7, number of transaction detail information items 3 ), Registered journal data "2019/6/25 (date), purchase (debit account), cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax), store A / beef -Cash / reduced tax rate compatible items (summary), 0.7 (journal similarity), NUML (blank flag), NUML (double account flag), transaction detail information 27 (transaction detail information), 1.0 (verification item) Value date), 1.0 (verification item value debit account), 1.0 (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 1 .0 (summary of verification item value), 1.2 (verification confirmed value) ”.

Since the journal similarity is 0.6 from the combination of the transaction item information 7 and the corresponding account (debit account "purchase" and credit account "cash"), the journal generation AI uses the journal similarity as a judgment condition. Journal generation from registered journal data (journal similarity is 0.7) with a threshold close to journal similarity 0.6 The journal data generated from transaction detail information 7 is "2019/7/25 (date), journal (Debit account), Cash (Credit account), ¥ 5,400 (Amount), ¥ 432 (Consumption tax amount), Store A / Beef / Cash / Reduced tax rate compatible item (Summary), 0.6 (Journal similarity ), NULL (blank flag), NULL (double accounting flag), transaction detail information 7 (transaction detail information), NUML (verification item value date), NULL (verification item value debit account), NULL (verification item value credit) "Account), NUML (verification item value amount), NULL (verification item value consumption tax amount), NULL (verification item value summary), NULL (verification confirmed value)" is inferred (judged).

Whereas the conventional journal AI determines the account item corresponding to the transaction detail information, the journal AI of the present invention can determine the combination of the transaction detail information and the corresponding account, and therefore, from the transaction detail information. It is possible to generate highly accurate journal data.

When the conventional journal AI is used, the account corresponding to the transaction item information 7 is the debit account "purchase", and since the combination of accounts is not specified, the payment method cannot be determined, and the conventional journal AI is the journal. Data "2019/7/25 (date), purchase (debit account), NUML (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax), beef (summary), NULL (similar to journal entry) Degree), NUML (blank flag), NULL (double accounting flag), transaction detail information 7 (transaction detail information), NULL (verification confirmed value) ”. Even if a credit account is set, the credit account is not set because the payment method is fixed, so the credit account is only set by chance.

However, since the journal generation AI of the present invention specifies the combination of accounts as the account corresponding to the transaction detail information 7, the journal generation AI is the journal data "2019/7/25 (date), purchase (debit account). Subject), Cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax), Store A / Beef / Cash / Reduced tax rate compatible item (Summary), 0.6 (Journal similarity), NUML (Blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), NUML (verification item value date), NULL (verification item value debit account), NULL (verification item value credit account) , NULL (verification item value amount), NULL (verification item value consumption tax amount), NULL (verification item value summary), NULL (verification confirmed value) ”is generated.

In addition, even for the same item, the account may change depending on the classification of the amount. The amount classification is the range in which the account changes depending on the amount. For example, in the case of fixed assets, the amount classification is less than 100,000 yen (account is consumable office supplies), 100,000 yen or more and less than 200,000 yen (account). There are three items: lump-sum depreciable assets) and 200,000 yen or more (accounts are tools and fixtures). In addition, even for transactions of the same product, the account item may change depending on the type of business of the business operator and the transaction status (customer, payment method, amount classification).

The journal generation unit 13 generates reference journal data to be used in the journal verification determination described later (S9).
The reference journal data is registered journal data extracted from the journal DB using transaction detail information as a key, and is at least the date, account (debit account and credit account), amount, consumption tax amount, and description of items for which journal verification judgment is performed. It consists of the items of.

The journal verification unit 14 searches for similar transaction detail information (hereinafter referred to as similar transaction detail information) from the journal DB using the transaction detail information as a key, and extracts registered journal data associated with the similar transaction detail information hit in the search. Further, the registered journal data associated with the similar registered image data may be acquired from the similar registered image data extracted in S3.

Compare and collate transaction detail information with similar transaction detail information, and extract for each item (date, debit account and credit account, amount, consumption tax amount, description, etc.) of matching similar transaction detail information items and corresponding journal data. Set the registered journal data as the reference journal data. Of the reference journal data, items that have not been set are acquired from the transaction details information and set.

The journal verification unit 14 makes a journal verification determination to verify whether the journal data generated by the journal generation unit 13 is correctly journalized (S10).

As a verification of whether the journal data is correctly journalized, it is checked whether there is a blank in the journal data item and whether there is double counting of journals (whether multiple journals with the same content are generated). If there is a blank in the journal data item, check the blank flag (not shown) in the journal data and end the journal verification there.

If there is no blank in the journal data item, check the double counting of journals, and if there is double counting, check the double counting flag (not shown) of the duplicate journal data. Performs duplicate journal elimination processing to delete or invalidate journal data containing the reprint flag.

In addition, when the duplicate journal elimination process is executed or there is no double accounting, "tax display information and consumption tax rate" by comparing and collating the journal data and the reference journal data as verification of whether the journal data is correctly journalized. Verification of the amount of journal data and consumption tax amount by combination "(verification of amount including consumption tax) and" whether the journal data generated from transaction detail information matches the reference journal data (judgment based on normal journal value) "(verification of journal accuracy) ) Is verified.

In the case of amount verification including consumption tax, each item amount, total item amount, each item consumption tax amount, and total item consumption tax amount are obtained from the text data. Assuming that the acquired item amount, total item amount, each item consumption tax amount, and total item consumption tax amount are tax-included, each item amount (tax excluded) and item amount total (tax excluded) are calculated, and each item amount acquired. And the total item amount (tax included) and the total item amount (tax included) are calculated assuming that the total item amount, each item consumption tax amount, and the total item consumption tax amount do not include tax.

If there are two types of consumption tax rates, reduced tax rate and normal tax rate, each item amount (tax excluded) and item amount total (tax excluded), each item amount (tax included), and item amount total (tax included) of the reduced tax rate and normal tax rate ) Is calculated.

There are two types of tax display information, tax-excluded (tax included) and tax-included (tax excluded), and there are two types of consumption tax rate, reduced tax rate or normal tax rate, so the amount and consumption tax amount are a combination of tax display information and consumption tax rate. There are four types. If the price of each item is the tax-excluded price, the consumption tax amount is calculated by the formula "consumption tax amount = tax-excluded price x consumption tax rate". If the price of each item is tax-included, the tax-excluded price can be calculated by the formula "tax-excluded price = tax-included price ÷ (1 + consumption tax rate)".

The consumption tax amount is calculated by the formula "consumption tax amount = tax-excluded price x consumption tax rate". The consumption tax amount calculated by the formula "consumption tax amount = tax-excluded price x consumption tax rate" and the consumption tax amount calculated by "consumption tax amount = tax-included price-tax-excluded price" are also compared and collated with each item consumption tax amount (tax-excluded). Calculate the total item consumption tax amount (excluding tax), each item consumption tax amount (tax included), and the total item consumption tax amount (tax included).

Acquired item amount, item amount total, item consumption tax amount, item consumption tax amount total, each item amount (tax excluded), item amount total (tax excluded), each item amount (tax included), and item amount total (tax included) ), Each item consumption tax amount (tax excluded), item consumption tax total (tax excluded), each item consumption tax amount (tax included), and item consumption tax total (tax included) Tax-included or tax-excluded) can be judged (amount verification).

If the tax display information (tax included or tax excluded) is incorrect as a result of the amount verification, highlight the item of the transaction detail information of the inconsistent part by the amount verification, and obtain it from the transaction detail information in the reference journal data item. If there are items (item amount, item consumption tax amount, etc.), reset the item amount, item consumption tax amount, etc. calculated with the correct tax display information (tax included or tax excluded) in the reference journal data.

If the tax display information (tax included or tax excluded) is correct, the (item) consumption tax amount in the transaction details information is the formula "(item) consumption tax amount = (item) consumption tax total x (item) amount ÷ (item) amount total" Calculate from.
Reset the calculated item consumption tax amount, etc. in the reference journal data.

Journal accuracy verification is a two-step judgment of journal verification range judgment and journal verification judgment.
In the journal verification range determination, the journal verification unit 14 determines the range in which the reference journal data and the journal data are compared and collated.

Specifically, the efficient journal verification range for each transaction detail information (transaction pattern) is the verification item value (date, debit account, credit account, amount, consumption tax amount, description) and verification confirmation value of each journal data. Is set.

Assuming that the judgment condition is equal to or more than the verification confirmed value, the journal verification unit 14 determines the verification item value (date, debit account, credit account, amount, consumption) of each item of the journal data whose verification confirmed value satisfies the predetermined value or more. Only the tax amount and description) are judged to be the journal verification range.

Journal data "July 25, 2019 (date), purchase (debit account), cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax amount), store A (summary), 0. 6 (journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit account) Subject), 1.0 (Verification item value credit account), 1.0 (Verification item value amount), 1.0 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1.0 (Verification item value summary) Verification confirmed value) ”.

The journal verification unit 14 determines that the verification item value of each item of the journal data satisfying the determination condition that the threshold value is 1.0 (verification confirmed value) or more is defined as the journal verification range, using the verification confirmed value as the determination condition. In the case of journal data created from transaction detail information 7, the verification item value of all items is 1.0, so the journal verification range is determined by the journal verification range judgment for all items of the journal data (date, debit account and credit account). , Amount, consumption tax amount, description, etc.).

By changing the journal verification range, the comparison and collation target is also changed depending on the state of the reference journal data (from all items verified to only one item verified), so refer to it like an appropriate journal verification judgment according to the transaction pattern. It is possible to make a journal verification judgment only for a specific item of journal data.

In addition, since the journal verification range is only for accounts (debit account and credit account), if the amount cannot be verified, the date, amount, and consumption tax amount are different in the reference journal data. Even if the journal verification judgment used is performed, verification can be performed with high accuracy.

In addition, by performing the journal verification range judgment, the range of the reference journal data to be compared and collated with the journal data (from "journal data generated by making all transaction information the same" to "only some transaction information such as accounts are the same" Since you can control up to the journal data generated by Can be prepared for.

The journal verification unit 14 compares and collates the reference journal data and the journal data within the determined journal verification range, and whether there is a discrepancy (difference) with the result of the comparison and collation using the journal normal value or the journal abnormal value as the determination condition. Make a journal verification judgment to verify.

Judgment conditions are judgment based on normal journal values (whether the journal data matches the reference journal data) or judgment based on abnormal journal values (whether the data is clearly incorrect compared to the reference journal data). Alternatively, the judgment may be made based on both the normal journal entry value and the abnormal journal entry value.

Specifically, the journal verification unit 14 compares and collates the reference journal data and the journal data, and if the comparison and collation results match, each item of the matching journal data (date, debit account and credit account, amount, consumption tax amount). , Description, etc.) verification item values are not updated. If the comparison and collation results do not match, the difference value is calculated from the comparison and collation results, and the verification items for each item (date, debit account and credit account, amount, consumption tax amount, description, etc.) of the journal data that do not match. Update the verification item value by adding the difference value to the value.

If the difference value is a character string such as date, account, description, etc. (for example, if the debit account of the reference journal data is "cash" and the debit account of the journal data is "checking account"), different characters in the character string The number of characters (3 characters) (“current” and “current deposit”) may be used (the debit account verification item value in that case is 1 + 3 = 4), and the reference journal data for the number of characters (4 characters) in the character string of the journal data. The ratio of the number of characters (3 characters) of different characters (0.75 = 3/4) may be used (in that case, the debit account verification item value is 1 + 0.75 = 1.75).

Also, if the difference value is a number such as the amount of money or consumption tax (for example, the amount of reference journal data is "3,000 yen" and the amount of journal data is "300,000 yen"), the difference (297,000 yen) is also acceptable. It may be good (the amount verification item value in that case is 1 + 297,000 = 297001), and the ratio of the difference in the reference journal data to the amount of the journal data or the consumption tax amount (0.99 = 297,000 ÷ 300,000) may be good (the amount verification item in that case). The value is 1 + 0.99 = 1.99). The difference value is calculated as an absolute value.

For example, reference journal data "2019/6/25 (date), purchase (debit account), cash (credit account), ¥ 5,600 (amount), ¥ 448 (consumption tax), store A (summary) , 0.6 (journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item) Value debit account), 1.0 (Verification item value credit account), 1.0 (Verification item value amount), 1.0 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1 .0 (verification confirmed value) ”, journal data“ 2019/7/25 (date), purchase (debit account), cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax amount) ), Store A (summary), 0.6 (journal similarity), NUML (blank flag), NUML (double account flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date) , 1.0 (verification item value debit account), 1.0 (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value consumption tax amount) Verification item value summary), 1.0 (verification confirmed value) ”, and the difference value is the ratio of different parts to the whole.

Since the journal verification unit 14 targets the verification item values that satisfy the threshold value of the verification confirmation value of 1.0 or more from the journal verification range judgment, all the items of the journal data (date, debit account, credit account, amount, consumption). Tax amount, description) is applicable.

The journal verification unit 14 compares and collates the reference journal data with the journal data for all items (date, debit account, credit account, amount, consumption tax amount, description) of the journal data as the journal verification range.

As a result of comparison and collation, the debit account, the credit account, and the description match, so the verification item value is not updated and remains at 1.0. Also, since the date is the journal data "2019/7/25" for the reference journal data "2019/6/25", the difference is 1 for the entire character string 10 characters (month is calculated with 2 digits). It is a character. Therefore, since the difference value is 0.1 (1/10 = 0.1), the verification item value of the date is 1.1 (1.0 + 0.1 = 1.1).

Further, since the amount is the journal data "¥ 5,400" with respect to the reference journal data "¥ 5,600", the difference is "¥ 200". Therefore, the difference value is 0.04 (¥ 200 ÷ ¥ 5,400 = 0.037 rounded to 0.04), so the verification item value of the amount is 1.04 (1.0 + 0.04 = 1. 04).

Further, since the consumption tax amount is the journal data "¥ 432" with respect to the reference journal data "¥ 448", the difference is "¥ 16". Therefore, the difference value is 0.04 (¥ 16 ÷ ¥ 432 = 0.037 rounded to 0.04), so the verification item value of the consumption tax amount is 1.04 (1.0 + 0.04 = 1.04). ).

Therefore, the verification item value of the date of the journal data is changed from 1.0 to 1.1, the verification item value of the amount is changed from 1.0 to 1.04, and the verification item value of the consumption tax amount is changed from 1.0 to 1.04. Will be updated.

Therefore, journal data "July 25, 2019 (date), purchase (debit account), cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax amount), store A (summary), 0.6 (journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.1 (verification item value date), 1.0 (verification item value) Debit account), 1.0 (Verification item value credit account), 1.04 (Verification item value amount), 1.04 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1. 0 (verification confirmed value) ”.

When the judgment condition is less than or equal to the normal journal value (threshold value 1.0), the journal verification unit 14 determines that the verification item value (debit account, credit account, description) of each item of the journal data has a threshold value of 1.0. The following items are judged to have correct values. Items whose verification item values (date, amount, consumption tax amount) of each item in the journal data exceed the threshold value of 1.0 may be highlighted as incorrect journals. The highlighting may be a display different from the normal display, and the display format is not particularly limited. For example, underlining the inconsistent part, making the character color different, thickening the character, or near the inconsistent part. Express by displaying a flag or the like. Further, a plurality of highlighting types may be set according to the consistency, or only one type may be set.

Examples of incorrect journal entries are when the consumption tax amount is calculated at the normal tax rate (10%) for products subject to the reduced tax rate (8%) (threshold value 2.5) and the amount of the item (for example, 100,000 yen). Is abnormally large compared to the consumption tax amount (for example, 60 yen) (threshold value 3.0) or the consumption tax amount (for example, 1000 yen) is abnormally large compared to the item amount (for example, 300 yen) (threshold value 3.5). ) Etc.

If the judgment condition is an abnormal journal value (threshold value 1.6) or more, the verification item value (date, debit account, credit account, amount, consumption tax amount, description) of each item of the journal data is used for the journal verification judgment. Items with a threshold value of 1.6 or more are judged to have incorrect values. Since the verification item values (date, debit account, credit account, amount, consumption tax amount, description) of all items of the journal data do not satisfy the judgment conditions, it can be judged that there is no abnormal data in the journal data.

Further, the journal verification unit 14 has a journal verification range AI, and the journal verification range AI may be subjected to the journal verification range processing. Journal data "July 25, 2019 (date), purchase (debit account), cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax), store A (summary), NULL ( Journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit account) , 1.0 (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value summary), 1.2 (verification confirmed If you set the weight to 2.0 for the account and description (the other items are weighted to 1,0), the journal verification range AI weights each item in the journal data and journals. The data is "2019/7/25 (date), purchase (debit account), cash (credit account), ¥ 5,400 (amount), ¥ 432 (consumption tax), store A (summary), NULL ( Journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 2.0 (verification item value debit account) , 2.0 (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 2.0 (verification item value summary), 1.2 (verification confirmed Value) ”.

Journal verification range AI is a journal data item "Purchase (debit account item)" that satisfies the judgment condition that the value of the verification item value of the journal data is 1.2 or more, with the verification confirmation value (1.2 or more) of the journal data as the judgment condition. ), Cash (credit account), Store A (summary) ”is the journal verification range and inferred (judgment).

Further, the journal verification unit 14 has a journal verification AI, and the journal verification generation AI may perform the journal verification process. Reference journal data "June 25, 2019 (date), purchase (debit account), accounts payable (credit account), ¥ 5,600 (amount), ¥ 448 (consumption tax), store A (summary), NUML (Journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit account) ), 1.0 (Verification item value credit account), 1.0 (Verification item value amount), 1.0 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1.3 (Verification item value summary) Final value) ”, journal data“ 2019/7/25 (date), purchase (debit account), cash (credit account), ¥ 540,000 (amount), ¥ 432 (consumption tax), store A (Summary), NUML (journal similarity), NUML (blank flag), NUML (double account flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification) Item value debit account), 1.0 (Verification item value credit account), 1.0 (Verification item value amount), 1.0 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1.2 (verification confirmed value) ”, and the difference value is the ratio of different parts to the whole. Further, it is assumed that the payment method of the business partner specifies credit card payment, and the weighting count of the debit account and the credit account is set to 2.0.

The journal verification AI compares and collates the reference journal data and the journal data, and as a result of the comparison and collation, calculates the difference value between the date, the credit account, the amount, and the consumption tax amount that do not match. Although the verification item value of the debit account is weighted, the verification item value (debit account) and the verification item value (verification item value) are not updated because the verification item values match as a result of comparison and collation. Applicable) remains at 1.0.

The difference value of the date item 0.1 (1/10 = 0.1) and the difference value of the credit account item 0.67 (rounded to 2/3 = 0.666) and the amount item The difference value of 0.99 (¥ 16 ÷ ¥ 540,000 = 0.989 rounded to 0.99) and the difference value of the consumption tax amount item (¥ 16 ÷ ¥ 432 = 0.037 rounded to the nearest 0. 04) is calculated.

Verification item value (date) from 1.0 to 1.1 (1.0 + 0.1 = 1.1), verification item value (credit account) from 1.0 to 3.34 (1.67 x weighting) 2.0 = 3.34), verification item value (amount) from 1.0 to 1.99 (1.0 + 0.99 = 1.99), verification item value (consumption tax amount) from 1.0 to 1. It is updated to .04 (1.0 + 0.04 = 1.04).

In the journal verification AI, when the judgment condition is less than or equal to the normal journal value (threshold value 1.0), does the journal verification judgment satisfy the threshold value of the normal journal value of 1.0 or less, which is the judgment condition, for the verification item value of each item of the journal data? Make a journal verification judgment. The verification item value (debit account, description) of the journal data that satisfies the threshold value of the journal normal value of 1.0 or less is inferred (judged) as correct (normal data), and the threshold value of the journal normal value of 1.0 or less is not satisfied (. It is inferred (determined) that the verification item values (date, credit account, amount, consumption tax amount) of the journal data (exceeding the threshold value 1.0) are incorrect.

When the judgment condition is the journal abnormality value (threshold value 1.6) or more, the journal verification judgment determines whether the verification item value of each item of the journal data satisfies the threshold value 1.6 or more of the journal abnormality value which is the judgment condition. Make a journal verification judgment. The verification item value (credit account, amount) of the journal data that satisfies the threshold value of the journal error value of 1.6 or more is inferred (judged) as incorrect journal data (including those whose data is clearly incorrect), and the journal is abnormal. The verification item values (date, debit account, consumption tax amount, description) of the journal data that do not satisfy the value threshold of 1.6 or more (threshold of less than 1.6) are inferred (determined) as normal journal data.

In addition, misjournalized data that is incorrect (including data that is clearly incorrect) may be highlighted. The highlighting may be a display different from the normal display, and the display format is not particularly limited. For example, underlining the inconsistent part, making the character color different, thickening the character, or near the inconsistent part. Express by displaying a flag or the like. Further, a plurality of highlighting types may be set according to the consistency, or only one type may be set. In addition, incorrect journal data and normal journal data, which are incorrect (including those whose data is clearly incorrect), may be highlighted separately so that the difference can be clearly seen.

Since the journal data before correction has a journal item that is largely incorrect, the journal data before correction is registered in the erroneous journal DB 32 as erroneous journal data.

Display voucher data, journal data, and reference journal data side by side for comparison.

As a result of visual confirmation, if there is an error in the comparison result of voucher data, journal data and reference journal data, correct the journal data. After that, the voucher data (registered image data) and the journal data (including the corrected journal data) are associated and registered in the journal DB as registered journal data.

According to this embodiment, the person in charge can automatically create the journal data just by converting the voucher into electronic data, which has the effect of reducing the trouble of inputting the journal data.

A description example of a modified example of Example 1 is shown below.

As a modification of the processing of the image recognition unit 11, an example of setting the image similarity after weighting important items in the transaction information will be described.

For example, when the voucher type is set to 1.5, the supplier is set to 1.5, the item is set to 2.0, the amount is set to 1.5, and the date is set to 0.5 as the weighting count (other items are 1.0). Since the transaction information item is "customer, business partner, date" in the area A1, the number of transaction information items is 3.5 (customer quantity 2 x weighting 1.5 + date quantity 1 x weight). Addition 0.5 = 3.5).

In area A2, since the transaction information item is "voucher type, business partner, date", the number of transaction information items is 8 (quantity of voucher type 1 x 1.5 + quantity of business partner 4 x 1.5 + quantity of date 1 x weight). Addition 0.5 = 8).

In the area A3, since the transaction information item is "customer", the number of transaction information items is 1.5 (quantity of business partner 1 x 1.5 = 1.5).

Since the transaction information item in the area A4 is "item, item, item, item, item, item", the number of transaction information items is 12 (quantity of item 6 × 2.0 = 2).

Since the transaction information item in the area A5 is "item, item, item, item, item, item", the number of transaction information items is 12 (quantity of item 6 × 2.0 = 2).

In the area A6, since the transaction information item is "amount, amount, amount, amount, amount, amount, amount, amount", the number of transaction information items is 12 (quantity of amount 8 × 1.5 = 12).

In area A7, the number of transaction information items is 9.5 (quantity of amount 1) because the transaction information items are "consumption tax amount, consumption tax amount, consumption tax amount, consumption tax amount, amount, consumption tax amount, payment method, payment method, consumption tax amount". × 1.5 + quantity of consumption tax 6 × 1.0 + quantity of payment method 2 × 1.0 = 9.5).

In area A8, since the transaction information item is the "consumption tax amount", the number of transaction information items is 1 (consumption tax amount quantity 1 x 1.0 = 1).

In area A9, the number of transaction information items is 9.5 (quantity of amount 1) because the transaction information items are "consumption tax amount, consumption tax amount, consumption tax amount, consumption tax amount, amount, consumption tax amount, payment method, payment method, consumption tax amount". × 1.5 + quantity of consumption tax 6 × 1.0 + quantity of payment method 2 × 1.0 = 9.5).

The image similarity, which is the total number of transaction information items in areas A1 to A9, is 72 (quantity of voucher type 1 x weighting 1.5 + supplier quantity 7 x weighting 1.5 + date quantity 2 x weighting 0. 5 + Item quantity 12 x Weight 2.0 + Amount quantity 12 x Weight 1.5 + Consumption tax quantity 13 x Weight 1.0 + Payment method quantity 4 x Weight 1.0 = 72).

As a modification of the processing of the journal generation unit 13, a case where the account items are not combined will be described below.
For "beef" (item information) and "food" (classification), refer to the account dictionary (not shown), and select the corresponding account "purchase", debit "debit", and journal similarity "0.6". get. The journal generation unit 13 sets "purchase" in the debit account of the journal data from the combination of the acquired account "purchase" and the debit "debit".

For "cash" (payment method information), refer to the account dictionary (not shown) and acquire the corresponding account "cash", the debit "credit", and the journal similarity "0.8". The journal generation unit 13 sets "cash" in the credit account of the journal data from the combination of the acquired account "cash" and the credit "credit".

There are two journal similarity, the journal similarity of the debit "debit" (hereinafter referred to as the debit journal similarity) and the journal similarity of the credit "credit" (hereinafter referred to as the credit journal similarity). And, among the credit accounts, set the journal similarity of the journal data based on the policy of whether the business operator has priority.

For example, if the policy of whether the business operator has priority is "priority is given to the account with many setting mistakes", if there are many mistakes in the setting of the debit account, the debit journal similarity is 0.6 and the journal data is similar to the journal. Set as degree. Registered journal data with similar journal similarity is extracted from the journal DB using the set journal similarity (debit journal similarity 0.6).

Conversely, if there are many mistakes in the setting of the credit account, set the credit journal similarity 0.8 as the journal similarity of the journal data. In addition, when the registered journal data is extracted by setting the debit journal similarity as the journal similarity, if there are multiple registered journal data with the same journal similarity, the credit journal similarity is used to further narrow down. May be extracted.

If the policy is "I want to improve the accuracy of the combination of debit and credit accounts", then "Total of debit journal similarity and credit journal similarity (0.6 + 0.8 = 1.4)" or "Debit". The difference between the journal similarity and the credit journal similarity (0.8-0.6 = 0.2) ”may be used as the journal similarity.

In addition, the journal similarity may be set by weighting the account items. For example, if the policy is "I want to improve the accuracy of the combination of debit and credit accounts" and "Sum of debit journal similarity and credit journal similarity" is set as the journal similarity, set the account "Cash". Assuming that the weight is 2.0, the account "accounts receivable" is 0.5, the account "purchase" is 1.5, and the account "purchase" is 0.8, the character string information " "30001 * Beef ¥ 5,400" becomes the debit account "Purchase" and the credit account "Cash", and the debit journal similarity is 0.9 (0.6 x 1.5 = 0.9), and the credit journal Since the similarity is 1.6 (0.8 × 2.0), the journal similarity is “total of debit journal similarity and credit journal similarity (0.9 + 1.6 = 2.5)” 2 .5 is set.

Further, as a modification of the processing of the journal generation unit 13, when a new description is generated without referring to the description dictionary, an example of generating a description based on the description setting rule will be described. From the transaction details information, "Set the description in the order of customer information, item information, amount information, consumption tax amount information, payment method information" and "Item information classification is product code, amount information is amount (number), consumption tax amount If the information is a monetary amount (number), do not set it in the description "and" If you want to set multiple description information such as supplier information, item information, etc. in the description, separate them with and then combine them "based on the description setting rule Set the generated description as the description of the journal data.

Since there is a description setting rule "If the item information classification is a product code, the amount information is an amount (number), and the consumption tax amount information is an amount (number), it is not set as a description", so "30001 *" (item information) is Do not set to description.

When setting multiple description information such as "customer information, item information, amount information, consumption tax amount information, payment method information in this order" and "customer information, item information, etc." in the description setting rule Since there is "to combine after separating with (separator)", the journal generation unit 13 puts "・" (separator), "beef" (item information), and "・" in the journal data description "Store A". (Separator) Set by adding (combining) in the order of "Reduced tax rate compatible item" (consumption tax amount information), "・" (separator), and "cash" (payment method information).

Since "¥ 5,400" (amount information) is an amount (number), the journal generation unit 13 sets "¥ 5,400" as the amount of the journal data. Since there is a description setting rule "If the item information classification is a product code, the amount information is an amount (number), and the consumption tax amount information is an amount (number), it is not set as a description", so "¥ 5,400" (amount information) ) Is not set in the description.

Further, assuming that the transaction detail information 7 is an item to be purchased regularly every month, the verification confirmation value is set to 1.2.

Therefore, the journal generation unit 13 includes transaction detail information 7 (transaction detail character string information “30001 *, beef, ¥ 5,400, groceries, tax excluded, 8%, ¥ 432, 2019/7/25, line number 7), Journal data "2019/7/25" from the transaction details information item "Item, Item, Amount, Classification, Consumption tax amount, Consumption tax amount, Consumption tax amount, Date, Business partner, Payment method" of "Store A, Cash" and line number 7. (Date), Purchase (Debit account), Cash (Credit account), ¥ 5,400 (Amount), ¥ 432 (Consumption tax amount), Store A / Beef / Cash / Reduced tax rate compatible item (Summary), 0 .6 (Journal similarity), NUML (blank flag), NUML (double accounting flag), transaction detail information 7 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit) Account), 1.0 (Verification item value credit account), 1.0 (Verification item value amount), 1.0 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1.0 (Verification confirmed value) ”is generated.

Further, as a modification of the processing of the journal generation unit 13 including the journal generation AI, a combination of the determination condition "business type of business operator" and transaction detail information will be described as an example.

For example, if the "business type" is a retail business and the payment method is specified by the business partner (the payment method "deferred payment" is specified for the business partner "store A"), the transaction statement information 11 (line number 11) For the transaction details character string information "40001, stationery, ¥ 200" and the transaction details information item "item, item, amount" of line number 11, the item information "stationery, office work" can be found by referring to the item dictionary (not shown). By acquiring "Supplies (classification)" and determining the payment method by referring to the payment determination dictionary (not shown) and the customer DB (not shown), the payment method information "deferred payment" is acquired.

The transaction detail information 11 before the update (transaction detail character string information "40001, stationery, ¥ 200" of line number 11 and transaction detail information item "item, item, amount" of line number 11) is the transaction detail information 11 after update. (The transaction detail character string information of line number 11 "40001, stationery, ¥ 200, office supplies, deferred payment" and the transaction detail information item of line number 11 "item, item, amount, classification, payment method") are updated.

When the business type is "retail business", the journal generation unit 13 including the journal generation AI has transaction detail information 11 (transaction detail character string information of line number 11 "40001, stationery, ¥ 200, office supplies, postpaid". And the transaction detail information item "item, item, amount, classification, payment method" of line number 11), the journal data "2019/7/25 (date), purchase (debit account), accounts payable" is determined by the journal generation judgment. Money (credit account), ¥ 200 (amount), ¥ 20 (consumption tax), Store A / stationery / deferred payment (summary), 0.4 (journal similarity), NUML (blank flag), NULL (double counting) Flag), transaction detail information 11 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit account), 1.0 (verification item value credit account), 1.0 (Verification item value amount), 1.0 (Verification item value consumption tax amount), 1.0 (Verification item value summary), 1.0 (Verification confirmed value) ”is generated.

When the business type is "manufacturing", the journal generation unit 13 including the journal generation AI has transaction detail information 11 (transaction detail character string information "40001, stationery, ¥ 200, office supplies, line number 11". If you enter "deferred payment" and the transaction statement information item "item, item, amount, classification, payment method" of line number 11, the journal data "2019/7/25 (date), consumables cost (debit account), accounts payable (debit account) Credit account), ¥ 200 (amount), ¥ 20 (consumption tax), store A / stationery / deferred payment (summary), 0.5 (journal similarity), NUML (blank flag), NULL (double counting flag) , Transaction detail information 11 (transaction detail information), 1.0 (verification item value date), 1.0 (verification item value debit account), 1.0 (verification item value credit account), 1.0 (verification Item value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value summary), 1.0 (verification confirmed value) ”is generated.

Further, as a modification of the processing of the journal generation unit 13 including the journal generation AI, a combination of the determination condition "classification of amount" and transaction detail information will be described as an example.

The amount classification is the range in which the account changes depending on the amount. For example, in the case of fixed assets, the amount classification is less than 100,000 yen (account is consumable office supplies), 100,000 yen or more and less than 200,000 yen (account). There are three items: lump-sum depreciable assets) and 200,000 yen or more (accounts are tools and fixtures). In addition, even for transactions of the same product, the account item may change depending on the type of business of the business operator and the transaction status (customer, payment method, amount classification).

The journal generation unit 13 including the journal generation AI has, for example, transaction detail information 50 (transaction detail character string information of line number 50 "personal computer, ¥ 90,000, office supplies, deferred payment" and transaction detail information item of line number 50 "item, When you enter "amount, classification, payment method"), there are three account candidates, "consumable office supplies cost, lump-sum depreciable assets, tool equipment cost", which correspond to "PC" (item) of transaction detail information 50. Is selected as. The account "Consumable office supplies costs" is the amount category "less than 100,000 yen", the account "Batch depreciable assets" is the amount category "100,000 yen or more and less than 200,000 yen", and the account "Tools and equipment" Is associated with the amount category "200,000 yen or more", so from the combination of "PC" (item) and "¥ 90,000" (amount) of transaction detail information 50, the account item is "consumable office supplies cost". Is specified in.

Therefore, journal data "Null (date), consumable office supplies expenses (debit account), accounts payable (credit account), ¥ 90,000 (amount), ¥ 20 (consumption tax amount), personal computer / office supplies / deferred payment (summary) ), NULL (journal similarity), NULL (blank flag), NUML (double accounting flag), transaction detail information 50 (transaction detail information), NULL (verification item value date), NULL (verification item value debit account) , NULL (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value summary), 1.0 (verification confirmed value) Is generated.

Further, the journal generation unit 13 including the journal generation AI has, for example, transaction detail information 51 (transaction detail character string information "personal computer, ¥ 150,000, office supplies, deferred payment" of line number 51) and transaction detail information item "line number 51". When you enter "item, amount, classification, payment method"), there are three accounts "consumable office supplies cost, lump-sum depreciable assets, consumable office supplies cost" as the accounts corresponding to "PC" (item) of transaction detail information 51. Is selected as a candidate for. The account item "Consumable office supplies costs" is the amount category "less than 100,000 yen", the account item "lump-sum depreciable assets" is the amount category "100,000 yen or more and less than 200,000 yen", and the account item "tool equipment" Is associated with the amount category "200,000 yen or more", so the account item is changed to "lump-sum depreciable assets" from the combination of "PC" (item) and "¥ 150,000" (amount) in transaction statement information 51. Be identified.

Therefore, journal data "Null (date), lump-sum depreciable assets (debit account), accounts payable (credit account), ¥ 150,000 (amount), ¥ 20 (consumption tax), personal computer / office supplies / deferred payment (summary) , NULL (journal similarity), NULL (blank flag), NULL (double accounting flag), transaction detail information 51 (transaction detail information), NULL (verification item value date), NULL (verification item value debit account), NULL (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value summary), 1.0 (verification confirmed value) " To generate.

Further, the journal generation unit 13 including the journal generation AI has, for example, transaction detail information 52 (transaction detail character string information "personal computer, ¥ 270,000, office supplies, deferred payment" of line number 52) and transaction detail information item "line number 52". When you enter "item, amount, classification, payment method"), there are three accounts "consumable office supplies cost, lump-sum depreciable assets, tools and fixtures" as the accounts corresponding to "computer" (item) in transaction detail information 52. Selected as a candidate. The account item "Consumable office supplies costs" is the amount category "less than 100,000 yen", the account item "lump-sum depreciable assets" is the amount category "100,000 yen or more and less than 200,000 yen", and the account item "tool equipment" Is associated with the amount category "200,000 yen or more", so the account item is changed to "tools and fixtures" from the combination of "PC" (item) and "¥ 270,000" (amount) of transaction statement information 50. Be identified.

Therefore, journal data "Null (date), tools and fixtures (debit account), accounts payable (credit account), ¥ 270,000 (amount), ¥ 20 (consumption tax), personal computer / office supplies / deferred payment (summary) , NULL (journal similarity), NULL (blank flag), NULL (double accounting flag), transaction detail information 52 (transaction detail information), NULL (verification item value date), NULL (verification item value debit account), NULL (verification item value credit account), 1.0 (verification item value amount), 1.0 (verification item value consumption tax amount), 1.0 (verification item value summary), 1.0 (verification confirmed value) " To generate.

Therefore, even if the same "personal computer" (item) is purchased, the correct account can be selected even if the accounts of "consumable office supplies, lump-sum depreciable assets, tools and fixtures" change.

Further, by adding weighting information to each item (verification item value) of the journal data as a modification of the processing of the journal verification unit 14 including the journal verification range AI, the journal verification range can be adjusted according to the journal accuracy of the journal data. Can be changed. For example, among standard transactions, transactions whose amount changes according to monthly usage, such as electricity charges, are subject to journal verification at least for accounts, whereas transactions with a fixed amount each month, such as newspapers, are at least. Accounts, amounts and consumption tax amounts are subject to journal verification.

For example, in the case of a transaction in which the amount changes according to the monthly usage such as electricity charges, the debit account is 2.0 and the credit account is 2.0 as a weighted count in the journal accuracy verification by referring to the transaction pattern dictionary. When 2.0 is set, 0.3 is set for the amount, 0.2 is set for the consumption tax amount, and 0.1 is set for the date (other items are set at 1.0), the threshold value of the verification confirmation value is set to 1.5 or more. Then, the debit account (threshold 1.0 x weighting 2.0 = 2.0) and the credit account (threshold 1.0 x weighting 2.0 = 2.0) have a threshold of 2.0. Amount (threshold 1.0 x weighting 0.3 = 0.3), consumption tax amount (threshold 1.0 x weighting 0.2 = 0.2) and date (threshold 1.0 x weighting 0.1 =) Since the threshold value of 0.1) is less than 1.0, the debit account and the credit account, which are the items to be verified by the journal that satisfy the judgment condition (verification confirmed value 1.5 or more), are the journal verification range.

Further, the processing in the present embodiment may be realized by software. Then, this software may be distributed by software download or the like. Further, this software may be recorded on a recording medium such as a CD-ROM and disseminated. This also applies to other embodiments herein.

The software that realizes the information processing device in this embodiment is the following program. That is, this program may be a program that executes Example 1 or a modification of those Examples.

As described above, the accounting processing apparatus, accounting processing method, and accounting processing program according to the present invention can automatically create journal data by imaging a voucher such as a receipt with a scanner or the like, and input journal data. It has the effect of reducing the time and effort required, and is useful as an accounting processing device, an accounting processing method, an accounting processing program, and the like.

1 Accounting processing device 1a Server 1b Terminal 10 Control unit 11 Image recognition unit 12 Character recognition unit 13 Journal generation unit 14 Journal verification unit 30 Storage unit 31 Journal DB
32 Image DB
33 Transaction DB
45 Input unit 46 Display unit 49 Communication unit 100 Accounting processing system

Claims (20)

An accounting device that automatically journals from voucher data, which is an image of a voucher.
An image recognition unit that extracts transaction information from voucher data by image recognition processing,
A character recognition unit that generates text data including transaction details information from the extracted transaction information by character recognition processing.
It is provided with a journal generation unit that generates journal data from the text data using the journal generation AI that has learned to perform machine learning based on teacher data in advance and select a combination of accounts corresponding to transaction detail information. Accounting device. It is provided with a journal verification unit that compares and collates the reference journal data associated with the transaction detail information that is the same as or similar to the transaction detail information and the journal data.
The accounting processing device according to claim 1, wherein the journal verification unit determines a range to be compared and collated with a predetermined threshold value, and compares and collates the journal data generated by the journal generation AI with the reference journal data. The image recognition unit has an image recognition AI that performs machine learning based on the teacher data in advance and learns the position and contents of the transaction information in the voucher data from the transaction information corresponding to the area in which the voucher data is divided into a plurality of areas. The accounting processing apparatus according to claim 1 or 2. Claims 1 to 3 characterized in that the character recognition unit has a character recognition AI that has learned to generate transaction detail information necessary for journal generation from the transaction information by performing machine learning based on teacher data in advance. The accounting processing device according to any one of the above. According to any one of claims 1 to 4, the journal verification unit has a journal verification AI that has learned to perform machine learning based on teacher data in advance and determine the validity of the journal data. The accounting device described. Any of claims 1 to 5, wherein the journal verification unit has a journal verification range AI that has learned to perform machine learning based on teacher data in advance and determine a journal verification range corresponding to transaction detail information. The accounting device described in paragraph 1. The journal generation unit generates classification information by classifying items of transaction detail information, and uses the journal generation AI learned to select a combination of accounts corresponding to the transaction detail information including the classification information. The accounting apparatus according to any one of claims 1 to 6, wherein the journal data is generated from. The journal generation unit generates classification information in which items of transaction detail information are classified and amount classification information in which the amount of transaction detail information is classified in stages, and is converted into transaction detail information composed of a combination of classification information and amount classification information. The accounting apparatus according to any one of claims 1 to 6, wherein journal data is generated from the text data using the journal generation AI learned to select a corresponding account combination. The journal generation unit generates classification information by classifying items of transaction detail information, and learns to select a combination of accounts corresponding to transaction detail information consisting of a combination of classification information and payment method. The accounting processing apparatus according to any one of claims 1 to 6, wherein the journal data is generated from the text data by using the method. The journal generation unit has learned to generate amount classification information in which the amount of transaction detail information is divided in stages, and to select a combination of accounts corresponding to the transaction detail information consisting of a combination of item and amount classification information. The accounting processing apparatus according to any one of claims 1 to 6, wherein journal data is generated from the text data using the journal generation AI. The journal generation unit is characterized in that journal data is generated from the text data by using the journal generation AI that has learned to select a combination of accounts corresponding to transaction detail information consisting of a combination of items and payment methods. The accounting processing apparatus according to any one of claims 1 to 6. The journal generation unit makes the journal generation AI perform machine learning using journal data classified by industry as teacher data in advance, and learns to select a combination of accounts corresponding to the combination of the business's industry and transaction detail information. The accounting processing apparatus according to any one of claims 1 to 6, wherein journal data is generated from the text data using the journal generation AI. It is an accounting method that automatically journals from voucher data, which is an image of vouchers.
An image recognition process that extracts transaction information from voucher data by image recognition processing,
A character recognition process that generates text data including transaction details information including at least items by character recognition processing from the extracted transaction information, and a character recognition process.
Journal data from the text data using the journal generation AI that learned to perform machine learning based on teacher data in advance and select a combination of accounts corresponding to transaction detail information including at least classification information that classified items. With a journal generation process to generate
An accounting method in which the journal data created by the journal generation AI, the voucher data, and the transaction details information are associated and displayed in parallel. It is provided with a journal verification step of comparing and collating the reference journal data associated with the transaction detail information that is the same as or similar to the transaction detail information and the journal data.
The accounting method according to claim 13, wherein the journal verification step determines a range to be compared and collated with a predetermined threshold value, and compares and collates the journal data generated by the journal generation AI with the reference journal data. The image recognition step has an image recognition AI that performs machine learning based on the teacher data in advance and learns the position and contents of the transaction information in the voucher data from the transaction information corresponding to the area in which the voucher data is divided into a plurality of areas. The accounting method according to claim 13 or 14. Claims 13 to 15 characterized in that the character recognition step has a character recognition AI that has learned to generate transaction detail information necessary for journal generation from the transaction information by performing machine learning based on teacher data in advance. The accounting method described in any one of the above. According to any one of claims 13 to 16, the journal verification step has a journal verification AI that has learned to perform machine learning based on teacher data in advance and determine the validity of the journal data. The accounting method described. Any of claims 13 to 17, wherein the journal verification step has a journal verification range AI learned to perform machine learning based on teacher data in advance and determine a journal verification range corresponding to transaction detail information. The accounting method described in paragraph 1. The journal generation process generates classification information by classifying items of transaction detail information, and uses the journal generation AI learned to select a combination of accounts corresponding to the transaction detail information including the classification information. The accounting method according to any one of claims 13 to 18, characterized in that journal data is generated from. An accounting program for causing a computer to execute the accounting method according to any one of claims 13 to 19.

Images