KR102416998B1 - Appatus for automatically collecting and classification tax related documents and method thereof - Google Patents

Abstract

According to the present invention, provided are an apparatus and a method for automating collection and classification of tax-related documents, wherein the method includes: a document collection step of collecting at least one tax-related document related to a client terminal from at least one web site through web crawling; a document type classification step of classifying a type of the taxed-related document; a document information classification step of recognizing meaning for each of one or more character strings included in the tax-related document; and an account title classification step of classifying account titles for the tax-related document. The document type classification step classifies a type of the tax-related document into one specific type based on an appearance frequency of characteristic information in a plurality of character strings included in the collected tax-related document by using an artificial neural network. The present invention has an effect of greatly improving work efficiency.

Description

APPATUS FOR AUTOMATICALLY COLLECTING AND CLASSIFICATION TAX RELATED DOCUMENTS AND METHOD THEREOF

The present invention relates to a tax document collection and classification automation device and method, and in particular, a tax document collection and classification automation device that enables a tax firm to easily collect and classify customer tax documents requiring tax returns and electronically document them. and methods.

A person who independently supplies goods or services for business is obliged to pay various taxes to the state regardless of whether he is a sole proprietor or a corporation. For example, it is necessary to organize the details of regular sales and purchases to report value-added tax to the National Tax Service, and to report corporate tax, local tax, year-end settlement, and global income tax to the National Tax Service. When the process for paying tax is called tax work, tax work is largely composed of document collection such as supporting documents, accounting, and tax return. However, since these procedures are very complicated and cumbersome, most business operators pay a fee to an external expert such as a tax agent and commission a tax service agent.

Of these, supporting documents, such as receipts such as tax invoices, credit card sales slips, cash receipts, and transaction statements, are often in the possession of business entities, so it is difficult for a tax agent to clearly collect all supporting documents from customers. It is important. However, this is simply repetitive and very cumbersome. For example, the tax agent requests the customer's accountant to send the supporting documents by e-mail and receives the supporting data by e-mail. Inefficient. In addition, the amount of evidence delivered to the tax agent for each client company is huge. In particular, the tax agent needs to collect the supporting documents within a short period of time because business requests from clients come in during the tax filing period.

However, since there are currently no work tools such as a program for automatically collecting evidence data or managing the collected evidence data, it is dependent only on the skill or competency of the person in charge.

In addition, after collecting the supporting documents, it is necessary to perform a classification task for accounting, which is also very labor-intensive and cumbersome. In fact, a lot of time and manpower must be invested in the work that needs to be done for tax filing, and most tax corporations are currently suffering from inefficient business processing.
Prior art literature: Publication No. 10-2020-0043603 (April 28, 2020)

The present invention was created to solve the above problems, and an object of the present invention is to enable a tax corporation to automatically collect various tax documents for tax returns from customers, to classify the types of collected tax documents ( For example, using artificial intelligence to classify information included in the tax document (eg, merchant name, address, purchase amount), and to classify the account subject of the tax document (eg, welfare benefits) It is to provide a method and device for automatically collecting and classifying tax documents.

A tax document collection and classification automation method according to an embodiment of the present invention for solving the above-described problems includes: a document collecting step of collecting at least one tax document related to a client terminal from at least one web site by web crawling; a document type classification step of classifying the type of the tax document; a document information classification step of identifying a meaning of each of the at least one character string included in the tax document; and an account subject classification step of classifying an account subject with respect to the tax document, wherein the document type classification step is based on an appearance frequency of characteristic information among a plurality of character strings included in the collected tax document using an artificial neural network. This classifies the type of tax document into one specific type.

In addition, the step of collecting the document further includes the steps of requesting the client terminal to send at least one necessary tax document and receiving the tax document from the client terminal, wherein the requesting the tax document includes: It may be performed irregularly by the command of the corporate terminal.

In addition, after receiving the tax document, the method may further include extracting a plurality of character strings included in the tax document through optical character recognition.

In addition, before the document type classification step, generating learning data by setting a plurality of character strings and a tax document type corresponding to the character string; and learning an artificial neural network (ANN) using the learning data; and generating the learning data includes characteristic information that can distinguish one tax document from another type of tax document for each tax document type. It may include the step of pre-defining the included feature information set.

In addition, the step of classifying the document information may include matching each of the plurality of character strings included in the tax document to one of predefined essential components for each tax document.

In addition, the step of classifying the document information may include matching the meaning of the text string to one of the essential constituent items based on the linguistic characteristics of the text string using natural language processing technology.

In addition, the step of classifying the document information may further include matching the meaning of the text string to one of the essential constituent items based on the positional characteristics of the text string using image processing technology.

In addition, after the document information classification step, the information omission determination step of determining whether there are essential components missing in the tax document; And when there is a missing item, the step of requesting the input of information to the terminal of the tax firm; further comprising, wherein the determining whether the information is missing, essential configuration items matched to each of the plurality of character strings in advance It is possible to check whether information is omitted by comparing it with the specified required configuration items.

In addition, after the step of receiving the tax document, the method further comprises a document omission determination step of comparing the received at least one tax document with the requested tax document to determine whether the receipt of the tax document is omitted, If there is a document, the request message for sending the missing tax document may be transmitted to the client terminal through a messenger program pre-installed in the tax office terminal by the instruction of the tax office terminal.

In addition, the step of classifying the account subject, the step of classifying the purpose of the received tax document into one of revenue or expense; And, based on the required component information of each tax document, the tax documents of the income account are classified into one of the accounts of sales or non-operating income, and the tax documents of the expense account are classified as one of the accounts of cost of goods sold, SG&A expenses, or corporate tax expenses. classifying; may include.

In addition, in the case of a tax invoice, the step of classifying the purpose includes dividing the income or expense accounts into revenue or expense accounts based on the client information and the supplier and recipient information among the required components of the tax invoice, and the step of classifying into the account subjects is, Account subjects may be classified based on at least one of supplier information or supplier information, transaction date, and transaction item name among the required components of the invoice.

In addition, in the case of the card transaction statement, the purpose is to classify the expense account, and the step of classifying the account item is based on at least one of the transaction date and time and the merchant name among the essential components of the card transaction statement. can do.

In addition, in the case of the cash receipt, the step of classifying it into the account subject may include classifying the account subject based on at least one information of a transaction date and time, a transaction item name, and a trade name among essential constituent items of the cash receipt.

On the other hand, the tax document collection and classification automation apparatus according to an embodiment of the present invention for solving the above-described problems is a web crawling that collects at least one tax document related to a client terminal from at least one web site by web crawling. a tax document collection unit including an execution unit; A document type classification unit for classifying the type of the tax document, a document information classification unit for understanding the meaning of each of the at least one character string included in the tax document, and an account subject classification unit for classifying account subjects with respect to the tax document and an electronic tax document conversion unit, wherein the document type classification unit uses an artificial neural network to specify the type of the tax document based on the frequency of appearance of characteristic information among a plurality of character strings included in the collected tax document. classified by type.

In addition, the tax document collection unit further comprises a document requesting/receiving unit for requesting the client terminal to send at least one required tax document and receiving the tax document from the client terminal, wherein the document requesting/receiving unit The tax document may be requested irregularly by the command of the corporate terminal.

In addition, the electronic tax document conversion unit pre-defines a characteristic information set in which characteristic information for distinguishing one tax document from another type of tax document is included for each tax document type, and sets a string included in various tax-related documents as an input value further comprising an artificial intelligence decision-making unit for learning a natural language processing model as It may further include a predefined characteristic information storage unit.

In addition, the document information classification unit may match each of the plurality of character strings included in the tax document to one of pre-defined essential components for each tax document.

In addition, the document information classification unit may match the meaning of the character string to one of the essential constituent items based on the linguistic characteristics of the character string using natural language processing technology.

In addition, the document information classification unit may match the meaning of the character string to one of the essential constituent items based on the positional characteristic of the character string using an image processing technology.

In addition, the tax document further includes an information omission determination unit for determining whether essential components are missing in the tax document, wherein the information omission determination unit is configured to pre-specify essential components matched to each of the plurality of character strings. It is checked whether information is omitted by comparing it with the constituent items, and when the missing items exist, it is possible to request input of information from the terminal of the tax firm.

In addition, the method further includes a document omission determining unit that compares the received at least one tax document with the requested tax document to determine whether a tax document is received or not, wherein the document omission determining unit is configured to: According to the instruction of the corporate terminal, a message requesting transmission of the missing tax document may be transmitted to the client terminal through a messenger program pre-installed in the tax corporation terminal.

In addition, the account subject classification unit divides the received tax document into one of revenue or expense, and based on the required component information of each tax document, the tax document of the income account is classified as one of the sales and non-operating income The tax documents of the expense account can be classified as one of the account items of cost of goods sold, SG&A expense, or corporate tax expense.

In addition, in the case of a tax invoice, the purpose is divided into revenue or expense accounts based on the supplier and recipient information among the client information and the required components of the tax invoice, and among the required components of the tax invoice, supplier information or supplier information, transaction Account subjects may be classified based on at least one of a date and a transaction item name.

In addition, the purpose of the card transaction statement may be divided into expense accounts, and the account subject may be classified based on at least one information of a transaction date and merchant name among essential constituent items of the card transaction statement.

In addition, in the case of the cash receipt, the account subject may be classified based on at least one information of a transaction date and time, a transaction item name, and a trade name among essential constituent items of the cash receipt.

According to an embodiment of the present invention, the process of collecting tax documents related to the client terminal basically performs a regular collection method through web crawling, the tax firm terminal requests the transmission of tax documents, and accordingly, from the client terminal In parallel with the irregular collection method of receiving and collecting tax documents. By adopting this method, it is possible to automate the collection of tax documents through regular collection of web crawling, and furthermore, the problem of omission of tax documents accompanying the regular collection method can be supplemented with irregular collection. Accordingly, it provides the advantage of enabling the terminal of a tax firm to easily collect documents necessary for tax filing.

On the other hand, by using artificial intelligence to automatically perform a series of processes of classifying document types, classifying information contained in documents, and assigning account subjects to the collected tax documents using artificial intelligence, conventional human labor Compared to the method of classifying tax documents with this method, it has the effect of significantly improving work efficiency.

On the other hand, there is an advantage of facilitating the identification of the document collection status by the tax firm terminal by comparing the tax documents requested from the tax office terminal with the tax documents received from the client and displaying the missing documents on the tax firm terminal. In addition, when there are missing documents, a message requesting transmission of the missing tax documents under the command of the tax office terminal can be easily transmitted to the client terminal through a messenger program pre-installed in the tax office terminal. Accordingly, the situation in which documents required for tax filing are received from customers in the prior art, and if documents are found missing, contact the customer individually by e-mail, etc. There is an advantage in that it can greatly reduce the stress and work load caused by the work method in which the above series of tasks are repeated again.

1 is a network configuration diagram including a tax document collection and classification automation apparatus according to an embodiment of the present invention.
2 is a block diagram schematically illustrating the configuration of an apparatus for collecting and classifying tax documents for automation according to an embodiment of the present invention.
Figure 3a is a flow chart sequentially illustrating a tax document collection automation method according to an embodiment of the present invention.
Figure 3b is a flowchart illustrating a tax document collection automation method according to an embodiment of the present invention.
4A is a flowchart illustrating a method for irregularly collecting tax documents in response to a request for sending tax documents to a client terminal according to an embodiment of the present invention.
4B is a flowchart illustrating a method for irregularly collecting tax documents in response to a request for sending tax documents to a client terminal according to another embodiment of the present invention.
5A is a photograph illustrating an image of a receipt as one type of tax document.
5B exemplarily shows a result of extracting text from the receipt image of FIG. 5A through optical character recognition.
6A is a diagram exemplarily showing a document form of a receipt as one type of tax document.
FIG. 6B is an exemplary diagram illustrating an electronic receipt document generated by filling in the text of FIG. 5B into a field in the document form of FIG. 6A .
6C is a diagram exemplarily illustrating a document form of a tax invoice as one type of tax document.
7 shows an example in which the omission determination result in the document omission determination unit is visualized on the monitor of the tax firm terminal.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Examples of the present invention are provided to more completely explain the present invention to those of ordinary skill in the art, and the following examples may be modified in various other forms, and the scope of the present invention is as follows It is not limited to an Example. Rather, these examples are provided so that this disclosure will be more thorough and complete, and will fully convey the spirit of the invention to those skilled in the art.

In addition, in the following drawings, the thickness or size of each layer is exaggerated for convenience and clarity of description, and the same reference numerals refer to the same elements in the drawings. As used herein, the term “and/or” includes any one and all combinations of one or more of those listed items. In addition, in the present specification, "connected" means not only when member A and member B are directly connected, but also when member A and member B are indirectly connected by interposing member C between member A and member B. do. The terminology used herein is used to describe specific embodiments, not to limit the present invention. As used herein, the singular form may include the plural form unless the context clearly dictates otherwise. Also, as used herein, “comprise, include” and/or “comprising, including” refer to the referenced shapes, numbers, steps, actions, members, elements, and/or groups thereof. It specifies the presence and does not preclude the presence or addition of one or more other shapes, numbers, movements, members, elements and/or groups.

Although the terms first, second, etc. are used herein to describe various members, parts, regions, layers and/or parts, these members, parts, regions, layers, and/or parts are limited by these terms so that they It is self-evident that These terms are used only to distinguish one member, component, region, layer or portion from another region, layer or portion. Accordingly, a first member, component, region, layer or portion discussed below may refer to a second member, component, region, layer or portion without departing from the teachings of the present invention.

Space-related terms such as “beneath”, “below”, “lower”, “above”, and “upper” refer to an element or feature shown in the drawing It may be used to facilitate understanding of other elements or features. These space-related terms are for easy understanding of the present invention according to various process conditions or usage conditions of the present invention, and are not intended to limit the present invention. For example, if an element or feature in a figure is turned over, an element or feature described as "below" or "below" becomes "above" or "above". Accordingly, "below" is a concept encompassing "above" or "below".

1 is a network configuration diagram including a tax document collection automation device according to an embodiment of the present invention.

The terminal for a tax firm is a terminal for a tax agent acting on behalf of a tax return, and a client terminal refers to a terminal of a sole proprietor or corporation that is obligated to file a tax return. The user of the tax document collection and classification automation apparatus and method according to the present invention is a terminal for a tax firm, and the tax firm may install a program including the method for automating the collection and classification of tax documents in the tax firm terminal. The apparatus and method of the present invention provide a work tool that can collect, classify, store, analyze and manage the tax documents of the client on-line with respect to the terminal of the tax firm. The device 100 of the present invention communicates with a server through a wired/wireless communication network.

Meanwhile, the client terminal and the tax office terminal are connected through a wired/wireless communication network. Such a wired/wireless communication network may be implemented as a wireless mobile communication network or a wired network communication network. For example, the client terminal and the terminal of the tax firm are mutually composed of a base transceiver station (BTS) and a mobile switching center (MSC). It is possible to transmit and receive related information such as tax documents using a mobile communication network. Alternatively, it may be implemented as a wired network communication network, and related information may be transmitted/received through the Internet communication network according to an Internet protocol such as Transmission Control Protocol/Internet Protocol (TCP/IP).

Meanwhile, the tax office terminal and the client terminal communicate with the device 100 of the present invention through a separate messenger program, for example, a KakaoTalk channel. The tax firm terminal may send an information request message to the client terminal through a messenger program, or may receive a predetermined documentary evidence from the client terminal.

That is, the tax firm performs accounting processing tasks such as collection, analysis, and classification of tax documents related to clients using the services provided by the automated device and method of the present invention, and when information exchange with the client is required in the process of performing the tasks uses messenger. The client terminal communicates with the tax firm through a messenger without accessing the automated device and method of the present invention, and the tax firm manages all of the various information and documents received through the messenger using the automated device and automated method of the present invention .

3A is a flowchart sequentially illustrating a method for automating the collection and classification of tax documents according to an embodiment of the present invention, and FIG. 3B is a flowchart illustrating a method for automating collection and classification of tax documents according to an embodiment of the present invention. In addition, FIGS. 4A and 4B are flowcharts illustrating a method for irregularly collecting tax documents in response to a request for sending tax documents according to an embodiment of the present invention. With reference to these drawings, a tax document collection and classification automation method according to an embodiment of the present invention will be described.

First, the tax document collection and classification automation method according to the present invention can be divided into a tax document collection process and an electronic tax document conversion process. The tax document collection process includes a document collection step of collecting at least one tax document related to a client terminal and, the electronic tax document conversion process includes a document type classification step for specifying the type of tax document, a document information classification step for understanding the meaning of at least one information included in the tax document, and classifying an account subject for the tax document It includes the step of classifying the account subject.

Specifically, the method for automating tax document collection and classification according to an embodiment of the present invention includes a document collection step of collecting at least one tax document related to a client terminal from at least one web site by web crawling, the type of tax document Document type classification step of classifying; and a document information classification step of identifying a meaning of each of at least one string included in the tax document, and an account classification step of classifying an account subject with respect to the tax document.

[Tax Document Collection Process]

First, an embodiment of the present invention performs the step of collecting at least one tax document related to the client terminal (tax document collection process). The tax document collection step basically performs regular collection through web crawling, and may additionally include irregular collection.

go. regular collection

The regular collection of tax documents is performed by the web crawling performing unit 111 (refer to FIG. 2 ) of the tax document collecting unit 110 . The web crawling performing unit 111 crawls a web site by a web crawler and automatically collects tax documents related to the client terminal from the web site. The web crawling performing unit 111 automatically collects tax documents related to the client terminal regularly according to a predetermined cycle without a separate request from the tax firm terminal. The target of the web crawling may include all websites that can collect tax documents related to client terminals, such as home tax and credit finance association. Before performing web crawling, the tax firm terminal may request the client terminal to provide access right or ID and PW to a specific website, and the client terminal may access its tax information provided by the website. You can receive authority, or ID and PW. For example, the tax office terminal can smoothly access tax documents related to the client terminal by receiving the consent of Hometax's tax agent appointment in advance. In addition, the tax office terminal may request the client terminal to pre-register the web crawling target web site, and the tax office terminal may be set to perform web crawling only on some web sites selected from the client terminal. The tax corporation terminal collects, for example, (1) electronic tax invoice, (2) credit card sales and purchase data, (3) cash receipt sales and purchase data, etc. of the client terminal from the website. Data obtained from the web site through web crawling may be a string (eg, store name, transaction time, transaction address, etc.) for detailed information included in the tax document.

me. Irregular Collection

Meanwhile, according to an embodiment of the present invention, the method may further include the steps of requesting transmission of at least one required tax document to the client terminal and receiving the tax document from the client terminal. And, the step of requesting the tax document may be performed irregularly by the command of the tax firm terminal.

Irregular collection of tax documents is performed through the tax document request/receiver 113 of the tax document receiving unit, is performed through a messenger, and is performed irregularly according to a separate command or request from the tax corporation terminal.

First, in the document collection step, the tax firm terminal requests the client terminal to send predetermined tax documents. At this time, the tax documents required to be collected differ depending on the type of tax return, and information on the required tax documents for each type of tax return is predetermined, for example, as shown in Table 1 below. And this information may be stored in advance in the tax document information DB. The tax firm terminal fetches a list of tax documents from the tax document information DB and requests the client terminal to send the documents on the list. Communication between the tax firm terminal and the client terminal may be performed through a separate messenger program, and the tax firm terminal may transmit a tax document list to the client terminal through, for example, KakaoTalk messenger. Thereafter, a tax document is received from the client terminal.

Type of tax return required documents Tax documents (proofing documents) VAT 1. VAT return
2. Total table of tax invoices by vendor
3. Total table of tax invoices by customer
4. Credit card sales slip collection statement
5. Tax deduction report for electronic tax invoice issuance
6. Statement of non-deductible input tax
7. Statement of acquisition of depreciable assets, such as buildings
8. Other Income 1. Tax Invoice
2. bill
3. Credit card sales slip
4. Cash Receipt
5. Export declaration certificate
6. Sales history of overseas shopping malls (Paypal, eBay, etc.)
7. Open Market Sales, App Store Sales
8. Credit card sales approval history (sales provided by payment agencies, PG companies, etc.), Nice Pay Inesis Kakao Pay, etc.
9. Other sales approval details (payment agency, delivery agency, etc. Yogiyo)
10. Hometax unregistered card details
11. Bankbook deposit and withdrawal details
12. Fictitious input tax credit report
13. Statement of Cash Sales
14. Animal treatment service sales statement
15. Statement of Export Performance
16. Zero-tax sales statement
17. Declaration of deduction for purchase of recycled waste resources, etc.
18. Bad debt tax credit report

Meanwhile, referring to FIG. 4A or FIG. 4B , after the step of receiving the tax document, a document omission determination step of determining whether the receipt of the tax document is omitted by comparing the received at least one tax document with the requested tax document is further performed. may include And, when there is a missing tax document, by the instruction of the tax firm terminal, a message requesting transmission of the missing tax document may be transmitted to the client terminal through a messenger program pre-installed in the tax firm terminal.

This step is performed by the document omission determination unit 160 , and compares at least one tax document received in the tax office terminal with the tax document requested by the tax corporation terminal to determine whether a missing document exists. Through this, as shown in FIG. 4A , after collecting all the requested tax documents, a document type classification step to be described later is performed.

Meanwhile, the step of determining whether a document is missing may be performed after the step of classifying the document type (refer to FIG. 4B ). Specifically, the type of the received tax document may be specified in advance, or the received tax document may be received in an unspecified state. For example, if the file name of the tax document received from the client terminal is cash receipt.pdf or simplified receipt.pdf, the tax document type is specified in advance, and in the case of an arbitrary file name, the type is unspecified. If the type of document can be specified, even if the type of tax document is not further classified, the tax document can be compared with the tax document requested by the terminal of the tax firm. do. However, if the type of the tax document cannot be specified because the received tax document has an arbitrary file name, it is necessary to determine the type of the tax document to determine whether it is omitted. Accordingly, as shown in FIG. 4B , the document type classification step is first performed before the document omission determination step. The document type classification method will be described later.

In both of the above cases, if a missing tax document is found, a message requesting sending the missing tax document is transmitted to the client terminal through a messenger program pre-installed in the tax office terminal by the command of the tax firm terminal; In other words, you can request tax documents again through messenger.

A step of determining whether a document is omitted after the above-described irregular collection step will be described with an example. 7 is an example illustrating a result of a tax corporation 　 terminal requesting a missing document from among the documents required for VAT payment to the client terminal. The required tax document LIST in the left column is a list of documents required for VAT payment, which is predefined in the tax document information DB. The tax firm terminal requests the client terminal to send some or all of these documents. The received tax document in the middle column is a list of tax documents received from the client terminal, and each received tax document corresponds to one of the required tax document LISTs. In this case, as described above, in the unspecified state of the tax document, the document type classification step may be performed first with respect to the received document. Information on the required document LIST and the received document LIST may be displayed in the form of a table on the terminal of the tax firm. The user of the tax firm terminal (the employee of the tax firm) can easily check whether there is a tax document that has not been received from the client terminal by checking the table. As described above, since the tax office terminal and the client terminal communicate through a separate messenger program, the tax office terminal sends a list of five tax documents, such as the export declaration certificate and other sales approval details that are not received in FIG. You can manually send a text message to the client terminal and manually request that the tax document be returned. Meanwhile, according to an embodiment of the present invention, request message transmission can be automated through the collection and classification automation method (program) of the present invention. That is, at the request of the tax office terminal, a message requesting transmission of the missing tax document may be transmitted to the client terminal through a messenger program pre-installed in the tax office terminal. In this case, the terminal of the tax firm can command all at once by i) selecting each missing tax document or ii) missing tax documents in bulk. For example, when the tax office terminal clicks (commands) the "Request" menu in the "Export declaration certificate" column in FIG. 7, a message requesting the sending of the document is automatically generated, and the message is displayed on the messenger program (KakaoTalk). This is a method in which the message is input as an outgoing message to the client and transmitted to the counterpart client terminal. According to this configuration, the tax office terminal can easily perform a request for missing documents by entering a simple command, and it is not necessary to manually type the missing documents request message in the message window.

As described above, tax documents are regularly/irregularly collected, and in the case of irregular collection, necessary tax documents can be secured by determining whether documents are omitted. After performing such a tax document collection process, an electronic tax document conversion process is performed.

[Electronic tax document conversion process]

The electronic tax document conversion process includes a document type classification step of classifying the types of tax documents, a document information classification step of understanding the meaning of each of at least one string included in the tax document, and an account classifying account subject for tax documents It includes the subject classification phase. The electronic tax document conversion process is performed by the electronic tax document conversion unit 150 (see Fig. 2), and all tax documents collected regularly or irregularly are processed in the document type classification stage, document information classification stage, and account subject classification stage. go through At this time, the three classifications are performed by the artificial intelligence decision-making unit 157 . Hereinafter, the process of converting the collected tax documents into electronic tax documents will be described in detail.

go. Document Type Classification Steps

The document type classification step is a process of classifying the tax document in the form of an input image into one of various tax document types, and the appearance of specific words among a plurality of strings included in the collected tax documents using an artificial intelligence model Classify the types of tax documents for each tax document based on their frequency. And, natural language processing technology and image processing technology are used here. Here, the type of tax document may be any document that can be used as proof when filing a tax return, such as receipts, bills, tax invoices, credit card sales slips, cash receipts, and bankbook deposit and withdrawal statements.

The document type classification step uses a natural language processing (NLP) model and a deep neural network (DNN) model to select one specific tax document type based on the frequency of appearance of characteristic information among a plurality of strings included in the collected tax documents. classified by type. Specifically, when a plurality of strings included in the collected tax documents are input into Transformer, a natural language processing deep learning model, characteristic information corresponding to each string is output, and when the characteristic information is input to a deep neural network (DNN) model, the appearance frequency Certain types of tax documents with high values are output as a result. That is, if the collected tax document includes a large number of words corresponding to characteristic information of a specific type of tax document, the type of the tax document may be specified based on this. For example, if a tax document includes a supplier and a recipient, the document type with the highest frequency of occurrence of these words is a tax invoice, so the tax document is classified as a tax invoice. As another example, if the received tax document contains words such as affiliate store and card number, the document type with the highest frequency of occurrence of these words is a credit card document, so the tax document is classified as a credit card document. In addition, as shown in FIG. 6c , the text of “tax invoice” is mostly directly exposed in the tax invoice, so the type of the document may be classified as a tax invoice from the string “tax invoice”.

At this time, in the case of tax documents collected irregularly rather than regularly collected through crawling, after the step of receiving the tax documents, a plurality of strings included in the tax documents are extracted through optical character recognition. Do the steps first. Optical character recognition technology is a technology that converts images of human or machine-printed characters into machine-readable characters, and uses this to extract character strings (text) inside tax documents. For example, from the credit card slip image of FIG. 5A , information such as the business number that issued the tax document, business address, industry type, transaction amount, issuance date and time, etc. may be extracted in a machine-readable form as shown in FIG. 5B . The result of optical character recognition is positional information (x-coordinate, y-coordinate, height, width) and a character string (text) existing in the corresponding positional space. Thereafter, the string extracted from the collected tax documents is compared with the feature information set to classify the tax document type.

On the other hand, prior to the document type classification step, the step of pre-defining a set of characteristic information in which characteristic information for distinguishing one tax document from another type of tax document is included for each tax document type; and learning the natural language processing artificial intelligence model "Transformer" by using the string included in various tax-related documents as input values. The characteristic information set contains specific characteristic information for each tax document type that can distinguish one tax document from another type of tax document, and this characteristic information set is predefined (see Table 2 below). For example, all tax invoices will contain the words 'tax invoice', 'supplier', 'supplier', etc. in common, and these words are not normally included in other types of tax documents, so it is possible to specify that they are tax invoices. are information. Therefore, "tax invoice", 'supplier', 'supplier', etc. are predefined as characteristic information included in the tax invoice.

Tax Document Types Feature information Feature information example Tax bill Feature information 1
Feature information 2
Feature information 3
Feature information 4 Tax bill
producer
recipient
… credit card slip Feature information 1
Feature information 2
Feature information 3
Feature information 4 card number
franchisee
Transaction Date
…

Meanwhile, according to the present invention, when classifying a document type, it is also possible to classify the document type by using an image processing technology, separately from the method of classifying the document type based on the frequency of appearance of the character string as described above. And, it uses image processing technology. Image processing technology is a technology for classifying document types through images of documents using a deep learning model (CNN, Convolutional Neural Network). The CNN model is a deep learning model with strength in image discrimination and learns image information of documents. For example, in the case of a credit card slip, items are arranged on the left and numbers are arranged on the right, and in the case of a tax invoice, learning to have a grid format as shown in FIG. 6C . In the process of training a deep learning model (CNN), such image characteristics are learned and when an image of a tax document is input, the type of tax document closest to the image of the input tax document is selected through machine learning among multiple tax document types. Determine it and print it out. In this case, the result of type classification may be provided as a probability.

Meanwhile, according to the present invention, the result of type classification using the above-described optical character recognition and natural language understanding technology (referred to as type classification based on the frequency of appearance of character strings) and the result of type classification using image processing technology (type classification based on images) ) can be taken into account, resulting in higher accuracy of document type determination. That is, when the result of the type classification based on the frequency of appearance of the character string matches the result of the type classification based on the image, the type of the document is determined. If they do not match, the document type classification process can be re-executed. For example, if the result of text-based classification is tax invoice/90%, and the result of image-based classification is tax invoice/60%, the tax document is classified as tax invoice. If the result of text-based classification is tax invoice/90% and the result of image-based classification is cash receipt/90%, the tax document classification process is re-executed.

me. Document Information Classification Steps

There is information that is essential for each tax document (hereafter, essential components). For example, in the case of a credit card slip, essential constituent items are the merchant name, business number, address, purchased product, purchase amount, and the like. The document information classification step includes matching each of a plurality of character strings included in the tax document to one of predefined required components for each tax document. Required components for each tax document are defined in advance and stored in the tax document information DB (refer to Table 3 below). For reference, since the above-described characteristic information consists of information having characteristics for distinguishing the types of tax documents, and the essential components refer to information that must be included once the types of tax documents are set, the essential components are It may not coincide with the above-described characteristic information.

Types of tax documents Required items Example of required configuration items Tax bill Required configuration item 1
Required configuration item 2
Required configuration item 3
Required configuration item 4
Required configuration items 5
Required configuration items 6
… Tax bill
Supplier registration number
Supplier name
Supplier business address
Supplier registration number
Supplier name
Supplier business address
… credit card slip Required configuration item 1
Required configuration item 2
Required configuration item 3
Required configuration item 4
Required configuration items 5
Required configuration items 6
… Store name
Company Registration Number
address
purchase
purchase amount
Sum
…

In this step, based on the linguistic characteristics of the string, the step of matching the meaning of the string to one of the essential constituent items is included. For example, if a tax document is classified as “credit card slip” in the document type classification step, referring to Table 3 above, the required components are 5 types of merchant name, business number, address, purchased item, and purchase amount. If the string as a result of optical character recognition of the tax document is “merchant name”, it is classified as a mandatory component of “merchant name”, and if the string is “business number”, it is classified as a required component of “business registration number”.

In this case, according to an embodiment of the present invention, natural language processing technology may be used. Specifically, depending on the character string, as in the case described above, the meaning of the character string is exposed to the character string, so that characteristic information can be grasped directly from the character string without separate interpretation of the meaning. In some cases, it is difficult to assign a character string, and the characteristic information contained in the string is classified only after understanding the meaning of the string. In the former case, for example, character strings such as “card number”, “address”, and “merchant name” are directly classified into characteristic information such as card number, characteristic information such as address, and characteristic information such as merchant name, respectively. do. In contrast, in the latter case, information is classified using the linguistic characteristics of character strings. And here, natural language understanding technology is used. Natural language understanding is a technology that makes a computer operate according to an input given in natural language, and is one of the elemental technologies of natural language processing. Natural language processing is a major field of artificial intelligence, and machine learning techniques, statistical natural language processing techniques, and deep learning techniques are used to understand natural language. For example, if, as a result of optical character recognition, the string “address” is not exposed and only the string “8 Daedo-dong, 353beon-gil, POSCO-daero, Nam-gu, Pohang-si, Gyeongsangbuk-do,” is exposed, the meaning of this string is determined based on linguistic characteristics. first understand In other words, the character string “8 Daedo-dong, 353beon-gil, POSCO-daero, Nam-gu, Pohang-si, Gyeongsangbuk-do,” is classified as “address” by applying natural language processing technology because its linguistic characteristics are close to addresses. On the other hand, since the same information may be expressed in a variety of different characters for each tax document, character strings meaning specific essential components (eg, transaction date) are opened (eg, transaction date, transaction date, transaction date, transaction time, sale date). , sales period, etc.).

In addition, according to an embodiment of the present invention, the step of classifying the information may include the step of matching the meaning of the corresponding character string to one of the essential components using image processing technology. Depending on the document type, there is a formal document in which specific information exists in a specific location, and in an informal document in which the location of specific information is different. For example, a receipt is an informal document that has no form, and a tax invoice is typically a tangible tax document with a common format. Therefore, for tax documents classified as tax invoices, it is possible to classify information using locational characteristics. For example, in a tax invoice, the total amount is usually placed in the lower right corner of the tax invoice (see FIG. 6c ). Therefore, it can be determined that the tax document includes information meaning “total amount” through such locational characteristics.

In order to do this, the relationship between information and location characteristics must be defined in advance (see Table 4 below). Specifically, when a positional characteristic (text box) obtained as a result of optical character recognition overlaps with at least one predefined positional characteristic (x-coordinate, y-coordinate, height, width), a character string existing in the positional characteristic is determined to indicate predefined characteristic information. For example, referring to Table 4 below, the results of OCR on the tax invoice [10, 10; 40, 60], since the corresponding positional characteristic overlaps with one of the predefined positional characteristics, the OCR result of 15,000 is predicted to mean the characteristic information of “total amount” and information to classify

Types of tax documents Geolocation: information example Tax bill [x1, y1; height, width]: info 1
[x2, y2; height, width]: info 2
… [10, 10 ; 40, 60] : “total amount”
[60, 10 ; 40, 60]: “business name”
…

On the other hand, if the tax documents were classified into one type in the previous step, it should be checked whether all the required components are included in the tax documents. For example, in the case of irregularly collected tax documents, the original image of the tax document contains all 5 required items, but in some cases, only 4 items were extracted due to a failure in optical character recognition, or the original image of the tax document itself There may be cases where one of the five essential items is missing. In this regard, according to an embodiment of the present invention, after the document information classification step, the information omission determination step of determining whether essential components missing in the tax document exist, and if the missing items exist, the tax It may further include the step of requesting input of information to the corporate terminal.

Specifically, since the required constituent items are defined in advance and stored in the tax document information DB as shown in Table 3, it is possible to set a goal that all required constituent items should be included for a specific tax document classified by type. The information omission determination step is to compare essential components matched to each of the plurality of strings with pre-defined essential components to check whether essential components are missing. As a result of comparison, if it is judged that the electronic tax document is inappropriate because some essential components are omitted, information input is requested to the terminal of the tax firm. For example, a window for manually inputting missing information may be displayed to be exposed. The information manual input window pops up on the terminal of the tax firm to manually input information from the tax accountant.

All. Account Subject Classification Steps

After the document type classification step and the information classification step, an account subject classification step of classifying account items for tax documents is performed. Categorizing an account subject specifies the purpose of the tax document in question. And, after the account purpose classification is completed, it may become data used for tax return.

Specifically, in the account subject classification step, the purpose of the tax document is sales, non-operating income (above, revenue account), cost of sales, sales and administrative expenses, and non-operating expenses based on at least one essential component included in the tax document. , and classifying it as one of corporate tax expenses (above, expense accounts). To this end, the present invention utilizes a deep learning deep neural network model (DNN, Deep Neural Network).

Specifically, the step of classifying the account subject includes the step of classifying the received tax document into one of revenue or expense for purpose. Specifically, the tax documents collected through the document collection step may be typically classified into one type of a tax invoice, a cash receipt, and a card transaction statement. And, these types can be divided into either a revenue account or an expense account.

First, depending on the document, the tax invoice may be divided into a revenue account or an expense account. In this step, it can be divided into revenue or expense accounts based on supplier and recipient information among the essential components of client information and tax invoice. For example, in a tax invoice collected from client A, if A corresponds to a supplier among the required components, the tax invoice is classified as a revenue account, and conversely, if A corresponds to a supplier, it is classified as an expense account. Meanwhile, the cash receipt may be divided into a revenue account or an expense account. Since the cash receipt-related data collected through web crawling includes whether the purpose of the cash receipt is purchase or sales, in this step, the purpose is classified based on the web crawling collected data. On the other hand, card transaction statements are divided into expense accounts.

Next, based on the required component information of each tax document, the tax documents of the income account are classified into one of the accounts of sales or non-operating income, and the tax documents of the expense account are one of the account subjects of cost of goods sold, SG&A expenses, or corporate tax expenses classified as For example, the account subject of the corresponding tax document may be classified based on information on at least one of a transaction occurrence point, transaction time, transaction location, same transaction frequency, and company characteristics. In addition, deep learning natural language processing model (Transformer) and deep learning deep neural network model (DNN, Deep Neural Network) are used here. For example, in the case of the above-mentioned transaction place, by using a deep learning natural language processing model, the transaction place, that is, the spending store information, is classified into several types, and the classified information of the spending store, the spending time period, the spending place, the spending frequency, the company's By entering the personality into a deep learning deep neural network model (DNN, Deep Neural Network), the account subject is determined in the transaction history.

In the case of a tax invoice, the account items may be classified based on at least one of supplier information or supplier information, a transaction date, and a transaction item name among the required components of the tax invoice. For example, a transaction date may be used to classify a revenue account or an expense account, and the transaction item may be used to classify an account item into SG&A expenses. On the other hand, in the case of a card transaction statement, account subjects can be classified based on at least one information among the transaction date and time and merchant name among the essential components. Account subjects may be classified based on one piece of information.

Meanwhile, a step of classifying whether the transaction on the tax document is a normal transaction or an abnormal transaction may be performed. Abnormal transactions refer to transactions in which the authenticity of sales or purchased transactions is questionable, or sales and purchases that are not related to the industry. In the abnormal transaction determination stage, a deep learning deep neural network model (DNN) or an anomaly detection model is used. This artificial intelligence model determines whether the transaction is abnormal based on characteristic information such as the client's business type and business type, the customer's business type and business type, transaction goods, supply price, and transaction time.

In addition, thereafter, the step of determining whether the transaction details on the tax document are non-deductible may be performed. In other words, it is checked whether the relevant transaction is not subject to deduction. Non-deductible items occur only in purchase transactions, and items such as entertainment expenses, private use, and vehicle fuel expenses are treated as non-deductible. Deduction of non-deductible items is made through rule-based judgment determined by the tax law, deep learning deep neural network model (DNN), or anomaly detection model (Anomaly Detection Model). Here, the rule-based determination refers to a method of determining in advance matters pertaining to unfair trade for each business type of the client terminal and determining whether a transaction target of the corresponding tax document corresponds to this. For example, if the client terminal (customer) is a gas station and the transaction item is fuel cost, it may be determined that the transaction item is non-deductible. In addition, it can be judged as the amount of private use or entertainment expenses by considering the business type and type of business of the customer, the goods to be traded, the supply price, and the transaction time.

Meanwhile, a step of determining whether the tax document is subject to penalty tax may be performed. Here, rule-based judgment is used. Rule-based judgment is a step in which a tax filing period is set in advance and the subject of penalty tax is determined based on whether the transaction date and time of the relevant tax document falls within the reporting period. For example, if the transaction is processed after the tax return period has elapsed, a tag stating that it is subject to penalty for non-compliance with the tax return period is attached.

[Electronic tax document storage process]

After the account subject of the tax document is classified through the above-described steps, the step of storing the electronic tax document in the electronic tax document DB may be performed.

Specifically, first, the document form is called from the tax document information DB (see FIG. 2 ) in which the document form according to the type of tax document is stored in advance. In the tax document information DB, document forms of various supporting documents are stored. Representatively, a document format for a receipt as one type of tax document is exemplarily shown in FIG. 6A . Then, the tax document is converted into an electronic tax document by filling in the string in the designated field in the document form. There is at least one field in the tax document form. Looking at the receipt document form of FIG. 6A , a plurality of blank fields such as a business registration number field, a trade name field, a business location field, a business type field, and an item field exist. 6B is an exemplary view of an electronic document of a receipt, and the texts of FIG. 5B (eg, business number, business address, industry type, transaction amount, issuance date and time, etc.) of a plurality of correspondence in the receipt document form of FIG. 6A By filling in the fields to create one electronic tax document.

Next, the electronic tax document generated as above is stored in the electronic tax document DB. At this time, as shown in Fig. 6b, not only the electronic tax document in the form of an image, but also the document type and document information classified to generate Fig. 6b are stored together in the form of Table or JSON in the electronic document DB.

The generated electronic tax document has three characteristics. First, the user can access the electronic document through the tax corporation (tax agent) terminal and the client terminal. Second, electronic tax documents are classified by type and purpose. Third, e-tax documents contain information about mandatory items for each type of document. For example, if the credit card sales slip is electronic, the electronic tax document includes information on the type of document called 'credit card sales slip', information on the purpose of proof of sales for VAT payment, and information on essential items such as business name, business number, and sales. do.

[Tax document collection and classification automation device]

2 is a block diagram schematically illustrating the configuration of a tax document collection and classification automation apparatus according to an embodiment of the present invention. Referring to the drawings, the automated device 100 of the present invention includes a tax document collection unit 110 , an electronic tax document conversion unit 150 , and an electronic tax document storage unit 180 .

The tax document collection unit 110 includes a web crawling unit 111, and the web crawling unit 111 collects at least one tax document related to the client terminal from at least one web site by web crawling. carry out The web crawling performing unit 111 automatically collects tax documents related to the client terminal regularly according to a predetermined cycle without a separate request from the tax firm terminal. The target of the web crawling may include all websites that can collect tax documents related to client terminals, such as home tax and credit finance association. Before performing web crawling, the tax firm terminal may request the client terminal to provide access right or ID and PW to a specific website, and the client terminal may access its tax information provided by the website. You can receive authority, or ID and PW. For example, the tax office terminal can smoothly access tax documents related to the client terminal by receiving the consent of Hometax's tax agent appointment in advance. In addition, the tax office terminal may request the client terminal to pre-register the web crawling target web site, and the tax corporation terminal may be set to perform web crawling only on some web sites selected from the client terminal. The tax corporation terminal collects, for example, (1) electronic tax invoice, (2) credit card sales and purchase data, (3) cash receipt sales and purchase data, etc. of the client terminal from the website. Data obtained from the web site through web crawling may be a string (eg, store name, transaction time, transaction address, etc.) for detailed information included in the tax document.

On the other hand, the tax document collecting unit 110 may further include a document requesting/receiving unit 113, wherein the step of requesting the transmission of at least one required tax document to the client terminal, receiving the tax document from the client terminal Follow the steps. And, the step of requesting the tax document may be performed irregularly by the command of the tax firm terminal. First, the document requesting/receiving unit 113 requests the client terminal to transmit predetermined tax documents. In this case, information on tax documents required for each type of tax return is predetermined, for example, as shown in Table 1 above, and this information may be stored in advance in the tax document information DB. The document requesting/receiving unit 113 retrieves a list of tax documents from the tax document information DB and requests the client terminal to transmit the documents on the list. The tax office terminal and the client terminal may communicate through a separate messenger program, and the document requesting/receiving unit 113 may transmit a tax document list to the client terminal through, for example, KakaoTalk messenger. Thereafter, the tax document is received from the client terminal.

On the other hand, the apparatus 100 according to an embodiment of the present invention compares at least one tax document received by the document requesting/receiving unit 113 with the requested tax document to determine whether the receipt of the tax document is omitted. The omission determination unit 160 may be further included. The document omission determination unit 160 is configured to, when there is a missing tax document, transmit a message requesting transmission of the missing tax document to the client terminal through a messenger program pre-installed in the tax firm terminal by the instruction of the tax firm terminal do.

As described above, the document type classification step to be described later may be performed after all requested tax documents are collected through the above-described reception omission determination of the document omission determination unit 160 (see FIG. 4a ), and also, document omission Determination of reception omission by the determination unit 160 may be performed after the document type classification step (refer to FIG. 4B ). Specifically, the type of the received tax document may be specified in advance, or the document type may be received in an unspecified state. If the type of document can be specified, the tax document type is not further classified. Even if the document omission determination unit 160 can compare the tax document with the tax document requested by the tax firm terminal, as shown in FIG. 4A , the document omission determination can be performed prior to document type classification. However, if the type of document cannot be specified, it is necessary to determine what type of tax document the tax document is in order to determine whether the document is missing. Accordingly, as shown in FIG. 4B , the reception omission determination by the document omission determining unit 160 may be performed after the document type classification step. In both cases, if a missing tax document is found, the document omission determination unit 160 transmits a message requesting transmission of the missing tax document to the tax firm terminal by the command of the tax firm terminal in advance. The tax document can be requested again by sending it to the client terminal through the program, that is, through the messenger. A method of determining whether a document is omitted by the document omission determining unit 160 is the same as described above.

On the other hand, the device 100 according to an embodiment of the present invention includes an electronic tax document conversion unit 150, which includes a document type classification unit 151 for classifying types of tax documents, at least included in the tax document A document information classifying unit 153 for understanding the meaning of each character string and an account subject classifying unit 155 for classifying account subjects for tax documents are included.

go. Document type classification unit (151)

The document type classification unit 151 performs a step of classifying the input tax document in the form of an image into one of various tax document types. Here, the type of tax document may be any document that can be used as proof when filing a tax return, such as receipts, bills, tax invoices, credit card sales slips, cash receipts, and bankbook deposit and withdrawal statements.

Specifically, the document type classification unit 151 classifies the tax document type for each tax document based on the frequency of occurrence of specific words among a plurality of character strings included in the collected tax documents using an artificial neural network. As will be described later, the artificial neural network generates learning data by setting a plurality of strings and the type of tax document corresponding to the string, and trains the artificial neural network (ANN) using the learning data in advance. is learned

Meanwhile, the document type classification unit 151 uses a natural language processing technology and an image processing technology. As a result of comparing the plurality of strings of the collected tax documents with the set of characteristic information (see Table 1), if the tax document contains a number of words corresponding to characteristic information of a specific type of tax document, based on this, the tax document type can be specified.

At this time, in the case of tax documents irregularly collected by the document requesting/receiving unit 113 , rather than the tax documents regularly collected through crawling by the web crawling execution unit of the tax document collection unit 110 , the tax documents are received. After that, the process of extracting a plurality of strings included in the tax document through optical character recognition is performed first. Extract the character string (text) inside the tax document using optical character recognition technology (see Fig. 5a). The result of optical character recognition is positional information (x-coordinate, y-coordinate, height, width) and a character string (text) existing in the corresponding positional space. Thereafter, the string extracted from the collected tax documents is compared with the feature information set to classify the tax document type.

On the other hand, the electronic tax document conversion unit 150 further includes an artificial intelligence decision-making unit 157, wherein the characteristic information for distinguishing one tax document from another type of tax document is a characteristic information set in which each tax document type is included. is defined in advance, and the natural language processing model 'Transformer' is trained using strings included in various tax-related documents as input values. When a plurality of strings included in the collected tax documents are input into Transformer, a natural language processing deep learning model, characteristic information corresponding to each string is output. In addition, a characteristic information storage unit in which a characteristic information set in which characteristic information for distinguishing one tax document from another type of tax document is included for each type of tax document is defined in advance may be further included. The artificial intelligence decision-making unit 157 classifies the tax document type through the artificial neural network learned together with the document type classification unit 151 .

Meanwhile, the document type classification unit 151 according to the present invention may classify the document type by using an image processing technology, separately from the method of classifying the document type based on the frequency of occurrence of the character string as described above. And, it uses image processing technology. Image processing technology is a technology for classifying document types through images of documents using a deep learning model (CNN, Convolutional Neural Network). The CNN model is a deep learning model with strength in image discrimination and learns image information of documents. For example, in the case of a credit card slip, items are arranged on the left and numbers are arranged on the right, and in the case of a tax invoice, learning to have a grid format as shown in FIG. 6C . In the process of training a deep learning model (CNN), such image characteristics are learned and when an image of a tax document is input, the type of tax document closest to the image of the input tax document is selected through machine learning among multiple tax document types. Determine it and print it out. In this case, the document type classification unit 151 may provide the result of the type classification as a probability.

On the other hand, the document type classification unit 151 according to the present invention provides a type classification result using the above-described natural language understanding technology (referred to as a type classification based on the frequency of appearance of a character string) and a type classification result using an image processing technology (in images). It is possible to perform high-accuracy document type discrimination by considering all of the based type classification). That is, if the results of the two types of classification match, the document type is determined, and if they do not match, the document type classification process may be re-executed.

me. Document information classification unit (153)

The document information classifying unit 153 matches each of the plurality of character strings included in the tax document to one of predefined essential components for each tax document. There is information that is essential for each tax document (hereafter, essential components). For example, in the case of a credit card slip, essential constituent items are the merchant name, business number, address, purchased product, purchase amount, and the like. Required components for each tax document are defined in advance and stored in the tax document information DB (see Table 3 above). For reference, since the above-described characteristic information consists of information having characteristics for distinguishing the types of tax documents, and the essential components refer to information that must be included once the types of tax documents are set, the essential components are It may not coincide with the above-described characteristic information.

The document information classification unit 153 performs a step of matching the meaning of the character string to one of essential constituent items based on the linguistic characteristics of the character string using natural language processing technology. For example, if a tax document is classified as “credit card slip” in the document type classification step, referring to Table 3 above, the required components are 5 types of merchant name, business number, address, purchased item, and purchase amount. If the string as a result of optical character recognition of the tax document is “merchant name”, it is classified as a mandatory component of “merchant name”, and if the string is “business number”, it is classified as a required component of “business registration number”.

In this case, the document information classification unit 153 may use natural language processing technology. More specifically, depending on the character string, as in the case described above, the meaning of the character string is exposed to the character string, so that characteristic information can be grasped directly from the character string without separate interpretation of the meaning. In some cases, it is difficult to assign a character string, and the characteristic information contained in the string is classified only after understanding the meaning of the string. In the former case, for example, character strings such as “card number”, “address”, and “merchant name” are directly classified into characteristic information such as card number, characteristic information such as address, and characteristic information such as merchant name, respectively. do. In contrast, in the latter case, information is classified using the linguistic characteristics of character strings. And, natural language understanding technology is used here, and natural language processing is a major field of artificial intelligence, and machine learning techniques, statistical natural language processing techniques, and deep learning techniques are used for natural language understanding. As a specific example, the foregoing in relation to “address” is cited. On the other hand, since the same information may be expressed in a variety of different characters for each tax document, character strings meaning specific essential components (eg, transaction date) are opened (eg, transaction date, transaction date, transaction date, transaction time, sale date). , sales period, etc.).

Meanwhile, the document information classification unit 153 may further perform a step of matching the meaning of the string to one of the essential constituent items based on the positional characteristics of the string using image processing technology. Depending on the document type, there is a formal document in which specific information exists in a specific location, and in an informal document in which the location of specific information is different. In the case of tangible tax documents that have a common format, such as a tax invoice, information can be classified using location characteristics for the tax documents. For example, in a tax invoice, the total amount is usually placed in the lower right corner of the tax invoice (see FIG. 6c ). Therefore, it can be determined that the tax document includes information meaning “total amount” through such locational characteristics. To do this, a relationship between information and location characteristics should be defined in advance, and a detailed description is the same as described in relation to Table 2 above.

Meanwhile, the apparatus 100 according to the present invention further includes an information omission determining unit 170 . If the tax documents were classified as one type in the previous step, it should be checked whether the tax documents contain all the required components. For example, in the case of irregularly collected tax documents, the original image of the tax document contains all 5 required items, but in some cases, only 4 items were extracted due to a failure in optical character recognition, or the original image of the tax document itself There may be cases where one of the five essential items is missing. Required configuration items may be defined in advance as shown in Table 3 above and stored in the tax document information DB. The information omission determining unit 170 checks whether information is missing by comparing the essential configuration items matched to each of the plurality of character strings with the pre-designated essential configuration items. As a result of comparison, if it is judged that the electronic tax document is inappropriate due to the omission of some essential components, information input is requested to the terminal of the tax firm. For example, a window for manually inputting missing information may be displayed to be exposed. The information manual input window pops up on the terminal of the tax firm to manually input information from the tax accountant.

All. Account Subject Classification Unit (155)

The account subject classification unit 155 performs an account subject classification step of classifying account subjects with respect to tax documents. Categorizing an account subject specifies the purpose of the tax document in question. And, after the account purpose classification is completed, it may become data used for tax return.

Specifically, the account subject classification unit 155 classifies the received tax document as one of revenue or expense, and based on the required component information of each tax document, the tax document of the revenue account is one of sales or non-operating income. The tax documents of the expense account are classified as one of the account items of cost of goods sold, SG&A expenses, or corporate tax expenses. To this end, the present invention utilizes a deep learning deep neural network model (DNN, Deep Neural Network).

First, in order to classify the account subject in the account subject classification unit 155, the purpose of classifying the received tax document into one of revenue or expense is performed. The collected tax documents may be typically classified into one type of tax invoices, cash receipts, and card transaction statements, and these types may be classified as either income accounts or expense accounts. First, depending on the document, the tax invoice may be divided into a revenue account or an expense account. In this step, it can be divided into revenue or expense accounts based on supplier and recipient information among the essential components of client information and tax invoice. Meanwhile, the cash receipt may be divided into a revenue account or an expense account. Since the cash receipt-related data collected through web crawling includes whether the purpose of the cash receipt is purchase or sales, in this step, the purpose is classified based on the web crawling collection data. On the other hand, card transaction statements are divided into expense accounts.

Next, the account subject classification unit 155 classifies the tax document of the income account into one of the account subjects of sales or non-operating income based on the required component information of each tax document, and the tax document of the expense account is cost of sales, It is classified as one of the accounts for SG&A expenses or corporate tax expenses. For example, the account subject of the corresponding tax document may be classified based on information on at least one of a transaction occurrence point, transaction time, transaction location, same transaction frequency, and company characteristics. In addition, deep learning natural language processing model (Transformer) and deep learning deep neural network model (DNN, Deep Neural Network) are used here. For example, in the case of the above-mentioned transaction place, by using a deep learning natural language processing model, the transaction place, that is, the spending store information, is classified into several types, and the classified information of the spending store, the spending time period, the spending place, the spending frequency, the company's By entering the personality into a deep learning deep neural network model (DNN, Deep Neural Network), the account subject is determined in the transaction history. In the case of a tax invoice, the account items may be classified based on at least one of supplier information or supplier information, a transaction date, and a transaction item name among the required components of the tax invoice. On the other hand, in the case of a card transaction statement, account subjects can be classified based on at least one information among the transaction date and time and merchant name among the essential components. Account subjects may be classified based on one piece of information.

In addition, the account subject classification unit 155 may perform a step of classifying whether the transaction on the tax document is a normal transaction or an abnormal transaction. Abnormal transactions refer to transactions in which the authenticity of sales or purchased transactions is questionable, or sales and purchases that are not related to the industry. In the abnormal transaction determination stage, a deep learning deep neural network model (DNN) or an anomaly detection model is used. This artificial intelligence model determines whether the transaction is abnormal based on characteristic information such as the client's business type and business type, the customer's business type and business type, transaction goods, supply price, and transaction time.

In addition, the account subject classification unit 155 may perform a step of determining whether the transaction details on the tax document are non-deductible. That is, it is checked whether the relevant transaction is not subject to deduction. Non-deductible items occur only in purchase transactions, and items such as entertainment expenses, private use, and vehicle fuel expenses are treated as non-deductible. Deduction of non-deductible items is made through rule-based judgment determined by the tax law, deep learning deep neural network model (DNN), or anomaly detection model (Anomaly Detection Model). Here, the rule-based determination refers to a method of determining in advance matters pertaining to unfair trade for each business type of the client terminal and determining whether a transaction target of the corresponding tax document corresponds to this. For example, if the client terminal (customer) is a gas station and the transaction item is fuel cost, it may be determined that the transaction item is non-deductible. In addition, it can be judged as the amount of private use or entertainment expenses by considering the business type and type of business of the customer, the goods to be traded, the supply price, and the transaction time.

Meanwhile, the account subject classification unit 155 may perform a step of determining whether the tax document is subject to penalty tax. Here, rule-based judgment is used. Rule-based judgment is a step in which a tax filing period is set in advance, and the subject of penalty tax is determined based on whether the transaction date and time of the relevant tax document falls within the reporting period. For example, if the transaction is processed after the tax return period has elapsed, a tag stating that it is subject to penalty for non-compliance with the tax return period is attached.

On the other hand, the automated device 100 of the present invention includes an electronic tax document storage unit (180). The electronic tax document storage unit 180 performs a step of storing the information of the tax document as an electronic tax document and storing the information in the electronic tax document DB.

Specifically, first, the document form is called from the tax document information DB (refer to FIG. 2 ) in which the document form according to the type of tax document is stored in advance. In the tax document information DB, document forms of various supporting documents are stored. Illustratively, FIG. 6A illustrates a document format for a receipt as one type of tax document. Then, the tax document is converted into an electronic tax document by filling in the string in the specified field in the document form. There is at least one field in the tax document form. Referring to the receipt document form of FIG. 6A , a plurality of blank fields such as a business registration number field, a trade name field, a business location field, a business status field, and an item field exist. 6B is an exemplary view of an electronic document of a receipt, and the texts of FIG. 5B (eg, business number, business address, business type, transaction amount, issuance date and time, etc.) of a plurality of correspondence in the receipt document form of FIG. 6A One electronic tax document has been created by filling in the fields to

Next, the electronic tax document generated as above is stored in the electronic tax document DB. At this time, as shown in Fig. 6b, not only the electronic tax document in the form of an image, but also the type of tax document classified to generate Fig. 6b, information, and account subjects may be stored together in the electronic document DB in the form of Table or JSON.

As described above, according to the method and apparatus for automating the collection and classification of tax documents according to the present invention, the process of collecting tax documents related to the client terminal is basically performed by a regular collection method through web crawling, and the terminal of the tax firm The occasional collection method of requesting the transmission of tax documents and receiving and collecting tax documents from the client terminal is carried out in parallel. By adopting this method, it is possible to automate the collection of tax documents through regular collection of web crawling, and furthermore, the problem of omission of tax documents accompanying the regular collection method can be supplemented with irregular collection. Accordingly, it provides the advantage of enabling the terminal of a tax firm to easily collect documents necessary for tax filing.

On the other hand, by using artificial intelligence to automatically perform a series of processes of classifying document types, classifying information contained in documents, and assigning account subjects to the collected tax documents using artificial intelligence, conventional human labor Compared to the method of classifying tax documents with this method, it has the effect of significantly improving work efficiency.

On the other hand, there is an advantage in that the tax office terminal easily grasps the document collection status by comparing the tax documents requested by the tax corporation terminal with the tax documents received from the client and displaying the missing documents on the tax corporation terminal. In addition, when there are missing documents, a message requesting transmission of the missing tax documents under the command of the tax office terminal can be easily transmitted to the client terminal through a messenger program pre-installed in the tax office terminal. Accordingly, we monitor the situation in which the required documents for tax return are received from the customer, and if any documents are found, contact the customer individually through e-mail, etc. There is an advantage in that it can greatly reduce the stress and work load caused by the conventional work method in which the above series of tasks must be repeated again.

What has been described above is only one embodiment for implementing the tax document collection and classification automation apparatus and method according to the present invention, and the present invention is not limited to the above-described embodiment, and as claimed in the claims Likewise, without departing from the gist of the present invention, it will be said that the technical spirit of the present invention exists to the extent that various modifications can be made by anyone with ordinary knowledge in the field to which the invention pertains.

100: tax document collection and classification automation device 110: tax document collection unit
111: web crawling performing unit 113: document requesting/receiving unit
150: electronic tax document conversion unit 151: document type classification unit
153: document information classification unit 155: account subject classification unit
157: artificial intelligence decision-making unit 160: document omission determination unit
170: information omission determination unit 180: electronic tax document storage unit

Claims (25)

A tax document collection and classification automation method performed by a tax document collection and classification automation device, the method comprising:
a document collecting step of collecting, by the tax document collecting unit of the device, at least one tax document related to the client terminal from at least one web site by web crawling;
a document type classification step in which the document type classification unit of the device classifies the type of the tax document;
a document information classification step in which the document information classification unit of the device grasps the meaning of each of the at least one character string included in the tax document; and
an account subject classification step in which the account subject classification unit of the device classifies the account subject for the tax document;
including,
The document type classification step is a step of classifying the type of the collected tax document into one of a plurality of tax document types. Classification of a type of tax document containing a number of item names,
After the document information classification step,
an information omission determination step in which the information omission determining unit of the device compares essential components matched to each of a plurality of character strings with predefined essential components to determine whether essential components missing in the tax document exist; and
If there is a missing item, requesting the input of information to the tax firm terminal;
further comprising,
The step of determining whether the information is missing is characterized in that the essential constituent items matched to each of the plurality of character strings are compared with a pre-designated essential constituent item to check whether information is missing,
How to automate the collection and classification of tax documents. According to claim 1,
The document collection step includes:
Further comprising the steps of requesting the client terminal to send at least one required tax document and receiving the tax document from the client terminal,
The step of requesting the tax document is characterized in that it is performed irregularly by the command of the tax firm terminal,
How to automate the collection and classification of tax documents. 3. The method of claim 2,
After receiving the tax document, the method further comprising the step of extracting, by the device, a plurality of character strings included in the tax document through optical character recognition,
How to automate the collection and classification of tax documents. According to claim 1,
Prior to the document type classification step,
predefining, by the artificial intelligence decision-making unit of the device, a feature item information set in which feature item information capable of distinguishing one tax document from another type of tax document is included for each tax document type; and learning the natural language processing model by using the item names included in various tax-related documents as input values;
How to automate the collection and classification of tax documents. 4. The method of claim 3,
The document information classification step includes matching each of a plurality of character strings included in the tax document to one of the required components for each predefined tax document,
How to automate the collection and classification of tax documents. 6. The method of claim 5,
The document information classification step includes matching the meaning of the string to one of the essential constituent items based on the linguistic characteristics of the string using natural language processing technology,
How to automate the collection and classification of tax documents. 7. The method of claim 6,
The document information classification step further comprises the step of matching the meaning of the character string to one of the essential constituent items based on the positional characteristics of the character string using image processing technology,
How to automate the collection and classification of tax documents. delete 3. The method of claim 2,
After the step of receiving the tax document, a document omission determination step in which the document omission determining unit of the device compares the received at least one tax document with the requested tax document to determine whether the receipt of the tax document is omitted further comprising: do,
When there is a missing tax document, according to the instruction of the tax firm terminal, a message for requesting transmission of the missing tax document is transmitted to the client terminal through a messenger program pre-installed in the tax firm terminal,
How to automate the collection and classification of tax documents. According to claim 1,
The step of classifying the account subject is,
purpose-classifying the received tax document as either revenue or expense; and
Based on the required component information of each tax document, the tax documents of the income account are classified into one of the accounts of sales or non-operating income, and the tax documents of the expense account are classified into one of the accounts of cost of goods sold, SG&A expenses, or corporate tax expenses. to do;
containing,
How to automate the collection and classification of tax documents. 11. The method of claim 10,
For tax invoices,
In the step of classifying the purpose, based on the supplier and recipient information among the client information and the required components of the tax invoice, it is divided into revenue or expense accounts,
The step of classifying into account subjects is to classify account subjects based on at least one information of supplier information or supplier information, transaction date, and transaction item name among the essential components of the tax invoice,
How to automate the collection and classification of tax documents. 11. The method of claim 10,
In the case of a credit card statement, the above purpose is divided into expense accounts;
The step of classifying into account subjects is to classify the account subjects based on at least one information of the transaction date and merchant name among the essential components of the card transaction statement,
How to automate the collection and classification of tax documents. 11. The method of claim 10,
For cash receipts,
The step of classifying into account subjects is to classify the account subjects based on at least one information of a transaction date and time, a transaction item name, and a trade name among essential components of a cash receipt,
How to automate the collection and classification of tax documents. a tax document collecting unit including a web crawling performing unit that collects at least one tax document related to a client terminal from at least one web site by web crawling;
A document type classification unit for classifying the type of the tax document, a document information classification unit for understanding the meaning of each of the at least one character string included in the tax document, and an account subject classification unit for classifying account subjects with respect to the tax document an electronic tax document conversion unit;
including,
The document type classification unit classifies the type of the collected tax document into one of a plurality of tax document types, and uses a pre-learned natural language processing model and a deep neural network model to name items included in the collected tax documents classified as a type of tax document that includes a number of
The document information classification unit matches each of the plurality of strings included in the tax document to one of the required components for each predefined tax document,
The tax document further comprises an information omission determination unit for determining whether there is an essential component missing from the tax document, wherein the information omission determination unit sets the required component items matched to each of the plurality of character strings in advance. to check whether information is omitted by comparing with
Tax documentation collection and classification automation device. 15. The method of claim 14,
The tax document collection unit,
Further comprising a document requesting/receiving unit for requesting transmission of at least one necessary tax document to the client terminal and receiving the tax document from the client terminal,
The document request/receiver unit will request the tax document irregularly by the command of the tax firm terminal,
Tax documentation collection and classification automation device. 15. The method of claim 14,
The electronic tax document conversion unit pre-defines a feature item information set in which feature item information that can distinguish one tax document from another type of tax document is included for each tax document type, and inputs item names included in various tax documents It further includes an artificial intelligence decision-making unit that trains a natural language processing model with a value,
The electronic tax document conversion unit further comprises a characteristic information storage unit in which a characteristic information set in which characteristic information for distinguishing one tax document from another type of tax document is included for each tax document type is predefined,
Tax documentation collection and classification automation device. delete 15. The method of claim 14,
The document information classification unit matches the meaning of the character string to one of the essential components based on the linguistic characteristics of the character string using natural language processing technology,
Tax documentation collection and classification automation device. 19. The method of claim 18,
The document information classification unit matches the meaning of the character string to one of the essential components based on the positional characteristics of the character string using image processing technology,
Tax documentation collection and classification automation device. delete 16. The method of claim 15,
A document omission determination unit further comprising a document omission determining unit that compares the received at least one tax document with the requested tax document to determine whether a tax document has been received or not,
The document omission determination unit, when there is a missing tax document, transmits a message requesting transmission of the missing tax document to the client terminal through a messenger program pre-installed in the tax firm terminal by the instruction of the tax firm terminal characterized,
Tax documentation collection and classification automation device. 15. The method of claim 14,
The account subject classification unit,
categorize the received tax documents as either revenue or expenses;
Based on the required component information of each tax document, the tax documents of the income account are classified into one of the accounts of sales or non-operating income, and the tax documents of the expense account are classified into one of the accounts of cost of goods sold, SG&A expenses, or corporate tax expenses. characterized in that
Tax documentation collection and classification automation device. 23. The method of claim 22,
For tax invoices,
Classify the purpose into revenue or expense accounts based on supplier and supplier information among the required components of client information and tax invoice;
Characterized in classifying the account subject based on at least one of supplier information or supplier information, transaction date, and transaction item name among the mandatory components of the tax invoice,
Tax documentation collection and classification automation device. 23. The method of claim 22,
The card transaction statement is divided by purpose into expense accounts,
Characterized in classifying the account subject on the basis of at least one of the transaction date and merchant name among the essential components of the card transaction statement,
Tax documentation collection and classification automation device. 23. The method of claim 22,
For cash receipts,
Characterized in classifying the account subject on the basis of at least one information of the transaction date and time, the transaction item name, and the trade name among the essential components of the cash receipt,
Tax documentation collection and classification automation device.

Images