JP2023165922A - Accounting audit support device, accounting audit support method and accounting audit support program - Google Patents

Abstract

To provide an accounting audit support device, an accounting audit support method and an accounting audit support program that present a method for estimating that there is an abnormality in accounting data.SOLUTION: An accounting audit support device comprises: a generation unit which generates a first test model by learning fraudulent accounting data at the time when fraud is committed in accounting, and normal accounting data in which fraud is not committed in accounting, and generates a second test model on the basis of the first test model and a calculated value used when generating the first test model; an acquisition unit which acquires target accounting data related to the accounting of an audited company; and an estimation unit which estimates the possibility of fraud in the accounting of the audited company, on the basis of the target accounting data and the second test model.SELECTED DRAWING: Figure 1

Description

The present invention relates to an accounting audit support device, an accounting audit support method, and an accounting audit support program.

2. Description of the Related Art Devices for performing financial analysis have been proposed in the field of accounting. The device described in Patent Document 1 is a financial analysis device that detects abnormalities included in accounting data. That is, the device described in Patent Document 1 generates a first vector based on the fluctuation values of a plurality of account items of accounting data, and estimates the fluctuation values of the plurality of account items based on the first vector. The device described in Patent Document 1 detects a residual difference between an estimated fluctuation value and an actual fluctuation value, and identifies account items for which a value correlated with the residual exceeds a threshold value. Furthermore, the device described in Patent Document 1 specifies journal entries of accounting data, generates a second vector based on the fluctuation values of a plurality of account items for each journal entry, and calculates residuals from among the second vectors. Extract journal entries that include accounts whose correlated values exceed a threshold. The device described in Patent Document 1 extracts the extracted journal entry when at least one account item included in the extracted journal entry has the following items.

Patent No. 6345856

An object of the present invention is to provide an accounting audit support device, an accounting audit support method, and an accounting audit support program that present a method for estimating that there is an abnormality in accounting data.

In one embodiment, an accounting audit support device generates a first test model by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting, and a generation unit that generates a second test model based on the test model and the calculated values used in generating the first test model; an acquisition unit that obtains target accounting data regarding the audited company's accounting; and a target accounting data. and an estimating unit that estimates the possibility that fraud is being committed in the accounting of the audited company based on the second test model and the second test model.
An accounting audit support device according to one aspect includes a storage unit that stores a test model generated by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting. , an acquisition unit that acquires target accounting data regarding the audited company's accounting; an estimation unit that estimates the possibility that fraud is occurring in the audited company's accounting based on the target accounting data and the test model; and a display section that displays the results estimated by the section. The estimation unit estimates the "possibility that fraud is being committed" and the "factors for fraud (falsification content)."

An accounting audit support device according to one embodiment includes a generation unit that generates virtual fraudulent accounting data according to details of anticipated fraud, and a storage unit that performs learning based on fraudulent accounting data, virtual fraudulent accounting data, and normal accounting data. The test model generated by performing the above may be stored.

In one aspect of the accounting audit support device, the storage unit stores subject teacher data regarding account subjects by learning a plurality of account subjects of accounting data and a standard account subject that is a basis for subject names of accounting data. The acquisition unit may allocate the account items described in the target accounting data to the reference account items based on the target accounting data and the subject teacher data.

One aspect of the accounting audit support method is to store in a computer a test model generated by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting. a storage step for storing target accounting data on the audited company's accounting, an acquisition step for acquiring target accounting data regarding the audited company's accounting, and estimating the possibility that fraud is occurring in the audited company's accounting based on the target accounting data and the test model. and a display step of displaying the result estimated by the estimation step on a display unit.

In one aspect of the accounting audit support program, a computer stores a test model generated by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting. A memory function, an acquisition function that acquires target accounting data regarding the audited company's accounting, and an estimation function that estimates the possibility that fraud is occurring in the audited company's accounting based on the target accounting data and the test model. and a display function that displays the results estimated by the estimation function.

An accounting audit support device in one embodiment estimates the possibility that fraud is being committed in the accounting of an audited company based on target accounting data and a test model, and displays the estimated result on a display unit. A method for estimating that there is an abnormality in data can be presented.
Further, the accounting audit support method and the accounting audit support program of one embodiment can have the same effects as the accounting audit support device of one embodiment described above.

FIG. 1 is a block diagram for explaining an accounting audit support device according to an embodiment. FIG. 7 is a diagram illustrating an example of an image showing the degree of possibility of fraudulent accounting, which is displayed on the display unit. FIG. 3 is a diagram for explaining an example of a graph (diagram) displayed on a display unit and used to support accounting audits. 3 is a flowchart for explaining an example of a method for generating a test model. 3 is a flowchart for explaining an example of a method of generating subject teacher data. It is a flowchart for explaining an example of a method of supporting an accounting audit.

An embodiment of the present invention will be described below.
FIG. 1 is a block diagram for explaining an accounting audit support device 1 according to an embodiment.

The accounting audit support device 1 is a device that supports the work of an auditing person who conducts an accounting audit. Accounting audits here include audits by external auditors by auditing corporations and certified public accountants, as well as audits by corporate auditors and internal audits by internal auditors. The accounting audit support device 1 learns accounting data in advance to generate a test model, and when an auditor audits the company's accounts, the test model is applied to the target accounting data of the audited company, and the application As a result, the possibility of fraud in the audited company's accounting (for example, falsification of accounting data, etc.) and the estimated factors leading to the fraud (contents of falsification) are output. When generating a test model, the accounting audit support device 1 uses, as accounting data, accounting data created without a company committing fraud (normal accounting data without fraud), and accounting data created when a company commits accounting fraud. learning accounting data (fraudulent accounting data). Classifying accounting data as fraudulent accounting data requires human judgment, but in addition to actual fraud such as violations of accounting standards, analysis of the quality of profits in accounting data may result in profit adjustment actions. You may classify it into fraudulent accounting data based on your judgment. Furthermore, the accounting audit support device 1 requires various types of fraudulent accounting data in order to support accounting audits using machine learning. However, in general, there are relatively few cases where a company commits accounting fraud, and it is difficult for the accounting audit support device 1 to obtain a large amount of information about accounting fraud actually committed by a company (fraudulent accounting data). In some cases. For such cases, the accounting audit support device 1 generates virtual accounting data (virtual fraudulent accounting data) assuming that a company commits accounting fraud, and learns and verifies the fraudulent accounting data. Generate the model.

The accounting audit support device 1 will be described in detail below.
The accounting audit support device 1 includes an acquisition section 11 , a generation section 12 , an estimation section 13 , a storage section 14 , and a display section 15 .
In this embodiment, an example in which the acquisition unit 11, the generation unit 12, the estimation unit 13, the storage unit 14, and the display unit 15 are arranged in the accounting audit support device 1 will be described. The configuration may be independent from the device 1. That is, the generation unit 12 provided in an external information processing device or the like generates a test model and outputs it to the accounting audit support device 1, and the accounting audit support device 1 stores the test model in the storage unit 14. good.

In order to generate the test model, the generation unit 12 uses, as accounting data that is teaching data, accounting data that the company has normally created without committing any fraud in accounting (normal accounting data), and accounting data that has been falsified etc. with respect to accounting by the company. Acquire accounting data (fraudulent accounting data) when fraud was committed. The normal accounting data and the fraudulent accounting data may be stored in, for example, a server (not shown).

When the ratio between the number of normal accounting data and the number of fraudulent accounting data is less than or equal to a preset threshold, the generation unit 12 generates virtual fraudulent accounting data according to the expected details of fraud. The accounting audit support device 1 generates a test model by machine learning, applies the test model to accounting data to be audited (target accounting data), and outputs the possibility that there is accounting fraud in the target accounting data. If the number of fraudulent accounting data is relatively small compared to the number of normal accounting data, the generation unit 12 will not be able to generate an appropriate test model, and accordingly, the accounting audit support device 1 will: It may not be possible to estimate a company's accounting fraud. Therefore, if the ratio of the number of fraudulent accounting data to the number of normal accounting data is below the threshold, the generation unit 12 determines the method of falsification (details of fraud) that may be performed by the company regarding accounting. Then, fraudulent accounting data (virtual fraudulent accounting data) is generated based on the estimated fraud details. For example, auditors who perform audits, data managers who are in charge of acquiring and managing accounting data for generating test models, and data analysts who are responsible for managing and adjusting test models, etc. The content of the fraud is examined, and virtual fraudulent accounting data is generated by the generation unit 12 based on operating the input unit (not shown) of the accounting audit support device 1 according to the examination result.

The generation unit 12 performs machine learning using normal accounting data and fraudulent accounting data as training data to generate a test model. Alternatively, the generation unit 12 performs machine learning using normal accounting data, fraudulent accounting data, and virtual fraudulent accounting data as training data to generate a test model. Therefore, the test model is a model that uses accounting data as input and can estimate whether the accounting data is normal, the possibility that fraud exists, and the factors that cause fraud (contents of falsification).
Here, a group of accounting data for grouping and managing accounting data that have something in common is called a "data set." For example, if data are grouped by fraud type, data sets such as "normal accounting data" and "fraudulent accounting data" can be created. As another example, if accounting data is grouped by industry, a dataset such as "construction" or "construction industry" can be created. Further, the data set may have a hierarchical structure and may include more detailed data sets. This allows the generation unit 12 to generate a test model according to the hierarchy (according to the industry). For example, the dataset is stratified in the order of "all industries,""construction," and "general construction," and "general construction" is divided into "general contractor A,""general contractor B," and "general contractor C," respectively. The data may be hierarchically divided into normal accounting data (or fraudulent accounting data) and details of accounting fraud estimated when the industry is "general construction" (virtual fraudulent accounting data).

For example, the generation unit 12 may generate a plurality of test models. For example, the generation unit 12 may generate a plurality of test models based on data appropriately selected from among a plurality of normal accounting data, a plurality of fraudulent accounting data, and a plurality of hypothetical fraudulent accounting data. Further, the generation unit 12 may generate one or more test models for each industry, for example.

Further, after generating the test model (test model 1), the generation unit 12 uses the generated test model (test model 1) and the calculated value when generating the test model (test model 1) to generate the test model (test model 1). One or more new test models (test model 2) may be generated, and one or more new test models (test model 3) may be generated. That is, the generation unit 12 may generate a multi-stage learning model.

The generation unit 12 may register a test index as a variable when performing machine learning. The test index is a variable used when calculating the possibility of accounting fraud. For example, the generation unit 12 calculates a significant difference in each test index between a plurality of data and a correlation between test indices within the same data set, and registers the test index based on the result of the calculation. For example, the generation unit 12 compares the test indices calculated between the datasets, and selects those with statistically significant differences (larger differences), and when the test indices within the datasets are compared with each other. The index with lower correlation is registered as a test index. The test index may be various, such as the ratio of working capital to average total assets. The generation unit 12 can perform calculations using statistical theory that generally describes "significant difference" and "correlation."
Note that when registering the test index as described above, the generation unit 12 may standardize the test index. For example, the generation unit 12 standardizes the test index by performing at least one of normalizing the value range, standardizing the variance, and standardizing the average of normal accounting data, fraudulent accounting data, and hypothetical fraudulent accounting data. You may do so.
After generating the test model, the generation unit 12 stores the generated test model in the storage unit 14 .

Further, the generation unit 12 may generate subject teacher data regarding account subjects (subject names). Specific examples of account items may include "assets," "liabilities," "net assets," "revenue," and "expenses." The subject teacher data is generated by performing learning based on a plurality of account subjects (subject names) of accounting data and a reference account subject (subject name) that serves as a basis for subject names of accounting data. Corporate accounting data may have different item names (account items) depending on the industry and company. If the test model's account (standard account) differs from the account of the target accounting data related to the audited company's accounting, the test model cannot be applied to the target accounting data and the possibility of accounting fraud cannot be estimated. there is a possibility. For this reason, the generation unit 12 generates subject teacher data in order to match the account subject of the target accounting data with the account subject of the test model (standard account subject). The generation unit 12 generates subject teacher data by acquiring a plurality of account subjects and learning which standard account subject these account subjects correspond to. After generating the subject teacher data, the generating unit 12 stores the generated subject teacher data in the storage unit 14 . The standard account is the account used in the test model.

Alternatively, if the generation unit 12 can indirectly associate the account item of the target accounting data with the account item of the test model, the generation unit 12 may store the association in the storage unit 14. For example, if the account item "Long-term interest-bearing debt" can be calculated from "Total liabilities", "Current liabilities" and "Fixed liabilities"("Long-term interest-bearing debt" = "Total liabilities" - "Current liabilities" - "Fixed liabilities") ), the calculation formula etc. may be stored in the storage unit 14.
Alternatively, the generation unit 12 may associate the account items of the target accounting data and the account items of the test model based on a preset degree of similarity, and store the results of the association in the storage unit 14. good.

The storage unit 14 stores the test model. The storage unit 14 may be, for example, a storage device such as a hard disk or a memory. As described above, the test model is a model generated by learning fraudulent accounting data when fraud is committed in accounting, and normal accounting data in which fraud is not committed in accounting. Alternatively, the test model is a model generated by learning fraudulent accounting data, virtual fraudulent accounting data, and normal accounting data. The test model is a model for obtaining the possibility that fraud such as falsification has been carried out in the accounting data (target data) of an audited company whose accounting is to be audited.

The storage unit 14 also stores subject teacher data regarding account subjects. As described above, the subject teacher data is data generated by learning a plurality of account subjects of accounting data and a reference account subject that is a basis for subject names of accounting data. The subject training data is a model for estimating which account of the test model the account of the target accounting data corresponds to.

The acquisition unit 11 acquires target accounting data regarding the audited company's accounting. The acquisition unit 11 acquires the target accounting data via a communication network, or by reading the target accounting data stored in an external memory, for example.

The acquisition unit 11 assigns the account items described in the target accounting data to the reference account items based on the target accounting data and the subject teacher data. That is, upon acquiring the target accounting data, the acquisition unit 11 assigns the account items described in the target accounting data to the account items (reference account items) of the test model based on the subject teacher data. Alternatively, upon acquiring the target accounting data, the acquisition unit 11 changes the account item (standard account item) to the account item of the target accounting data based on the result of associating the account item of the target accounting data with the account item of the test model in advance. Calculate from. That is, the estimating unit 13 assigns the account item of the target accounting data to the account item of the test data, or renames the account item of the target accounting data to the account item of the test model.
Further, if the acquisition unit 11 cannot calculate the standard account item from the account item of the target accounting data based on the subject teacher data, the acquisition unit 11 may set the account item of the target accounting data as the standard account item, and The items may be set as standard account items after being modified or otherwise changed as appropriate.

The estimation unit 13 estimates the possibility of fraud in the audited company's accounting based on the target accounting data and the test model. That is, the estimation unit 13 applies the test model to the target accounting data of the audited company, and obtains the degree of possibility of fraudulent accounting (a numerical value, for example). The estimating unit 13 obtains "unconditional probability", "fraud execution probability", and "F score" as the height of the possibility of fraud. The unconditional probability is the median value of the numerical values obtained by the estimation unit 13. The fraud probability indicates the result estimated by the estimation unit 13. If the probability of fraud is higher than the unconditional probability, it indicates that there is a high possibility that fraud has been committed in the audited company's accounting. The F-score is the ratio between the unconditional probability and the probability of fraud.
Here, when a plurality of test models are generated in the generation section 12, the estimation unit 13 provides target accounting data to each of the plurality of test models selected from the plurality of test models, and the estimation unit 13 provides target accounting data to each of the plurality of test models selected from the plurality of test models, and It may also be used to obtain a high level of sexuality.
In this case, the estimation unit 13 calculates the probability of accounting fraud based on the maximum value (or minimum value) of the results (numeric values) obtained by applying each of the plurality of test models to the target accounting data. It is also possible to obtain
Alternatively, the estimation unit 13 may obtain the probability of fraudulent accounting based on the average value of the results (numeric values) obtained by applying each of the plurality of test models to the target accounting data.
Estimating the "possibility of fraud" by the "estimation unit" of the present invention includes estimating "factors for fraud (falsification content)".

The display unit 15 displays the results estimated by the estimation unit 13. That is, the display unit 15 displays the degree of possibility of fraudulent accounting acquired by the estimation unit 13.

FIG. 2 is a diagram illustrating an example of an image showing the degree of possibility of fraudulent accounting, which is displayed on the display unit 15.
The display unit 15 displays an image as illustrated in FIG. 2 as an image of the high possibility of fraudulent accounting. The display unit 15 displays an image that includes the name of the company to be audited, the accounting period to be audited, the audit method (type of verification index), and the like. In addition, the display unit 15 displays the degree of possibility of fraud in the image as a numerical value. As a specific example, the display unit 15 displays "unconditional probability", "fraud implementation probability", and "F score" as the height of the possibility of fraud.
The unconditional probability is the median value of the results (numerical values) estimated by the estimation unit 13. In the case illustrated in FIG. 2, the unconditional probability is 0.004, and this value is the median value when determining whether or not fraud is occurring in accounting.
The fraud probability indicates the result estimated by the estimation unit 13. If the probability of fraud is higher than the unconditional probability, it indicates that there is a high possibility that fraud has been committed in the audited company's accounting. In other words, if the probability of fraud is lower than the unconditional probability, this indicates that there is a low possibility that fraud is occurring in the audited company's accounting.
The F-score is the ratio between the unconditional probability and the probability of fraud. If the F-score value is higher than 1, it indicates that there is a high possibility that fraud has been committed in the audited company's accounting.

FIG. 3 is a diagram for explaining an example of a graph (diagram) displayed on the display unit 15 and used to support accounting audit.
In addition to the numerical values illustrated in FIG. 2, the display unit 15 also displays a graph comparing the numerical value of any one of the plurality of account items of the target accounting data with the reference numerical value, as illustrated in FIG. may also be displayed.
The solid lines shown in FIG. 3 indicate the numerical values of multiple account items A to E of the target accounting data. The broken lines shown in FIG. 3 are numerical values that serve as standards for account items A to E, such as the average value of the industry to which the company that created the target accounting data belongs. In this way, the display unit 15 can show, for example, the account items A to E of the audited company and the standard values of the account items A to E to the person in charge of auditing. The auditor refers to Figure 3, and if there is a solid line inside the broken line, the audited company's accounting data is considered to be within the standards, and there is a possibility that fraud has occurred in the audited company's accounting data. can be judged to be low. On the other hand, the auditor refers to Figure 3, and if there is a solid line outside the broken line, the audited company's accounts are not within the standards, and there is a possibility that fraud has been committed in the non-standard accounts. It can be determined that there is.

Note that the graph (diagram) displayed by the display unit 15 is not limited to the example shown in FIG. 3, and various graphs (diagrams) may be displayed. For example, the display unit 15 may display not only a two-dimensional graph (diagram) as shown in FIG. 3 but also a three-dimensional graph (diagram).
In addition, for example, when target accounting data for multiple years is acquired from the audited company, the display unit 15 displays the return on equity (ROE) recorded in the target accounting data and the return on equity. It may also be possible to display trends over multiple years with net income, equity capital, etc. obtained from.

Next, an accounting audit support method according to an embodiment will be described.
The accounting audit support method includes, for example, a step of generating a test model to estimate the possibility of fraud such as falsification of a company's accounting, and a step of generating subject teacher data to unify account subjects (subject names). A step of generating the target accounting data and a step of supporting the accounting audit of the target accounting data of the audited company are performed.

First, an example of a method for generating a test model will be described.
FIG. 4 is a flowchart for explaining an example of a method for generating a test model.

In step ST101, the generation unit 12 acquires a plurality of normal accounting data. For example, the generation unit 12 may acquire normal accounting data stored in a server (not shown) or the like. Normal accounting data is accounting data that is normally created by a company without committing any accounting fraud.

In step ST102, the generation unit 12 acquires a plurality of pieces of fraudulent accounting data. For example, the generation unit 12 may acquire fraudulent accounting data stored in a server (not shown) or the like. Fraudulent accounting data is accounting data obtained when a company commits fraud such as falsification of accounting.

In step ST103, the generation unit 12 determines whether the ratio between the number of normal accounting data acquired in step ST101 and the number of fraudulent accounting data acquired in step ST102 is less than or equal to a threshold value. The threshold value is appropriately set, for example, by a person in charge of auditing. If the number of fraudulent accounting data is relatively small, the generation unit 12 may not be able to generate an appropriate test model. Therefore, if the number of fraudulent accounting data is less than the threshold, in step ST104 described later, the generation unit 12 estimates the method of falsification (details of fraud) that the company may perform regarding accounting, and estimates the Fraud accounting data (virtual fraud accounting data) is generated based on the details of the fraud committed.

In step ST104, the generation unit 12 generates virtual accounting data. For example, when a data analyst or the like infers the content of falsification when a company commits accounting fraud, the generation unit 12 generates the estimated content as virtual accounting data based on the operations of the data analyst or the like.

In step ST105, the generation unit 12 generates a test model based on learning the normal accounting data acquired in step ST101, the fraudulent accounting data acquired in step ST102, and the virtual fraudulent accounting data generated in step ST104. do. For example, the generation unit 12 generates one or more test models by performing machine learning on normal accounting data labeled as "normal" and fraudulent accounting data and virtual accounting data labeled as "fraudulent." generate.
The generation unit 12 may store the generated test model in the storage unit 14.

Next, an example of a method for generating subject teacher data will be described.
FIG. 5 is a flowchart for explaining an example of a method for generating subject teacher data.

Corporate accounting data may have different item names (account items) depending on the industry and company. If the test model's account (standard account) differs from the account of the target accounting data related to the audited company's accounting, the test model cannot be applied to the target accounting data and the possibility of accounting fraud cannot be estimated. there is a possibility. For this reason, the generation unit 12 generates subject teacher data in order to match the account subject of the target accounting data with the account subject of the test model (standard account subject).

In step ST201, the generation unit 12 acquires a plurality of account items (item names). For example, the generation unit 12 may acquire a plurality of pieces of accounting data and acquire account items described in the accounting data.

In step ST202, the generation unit 12 acquires a standard account item (item name). The standard account item is, for example, an account item (item name) used in the test model. The standard account items are set in advance and stored, for example, in the storage unit 14 or the like.

In step ST203, the generation unit 12 generates subject teacher data by learning the account subjects acquired in step ST201 and the reference account subjects acquired in step ST202. For example, the generation unit 12 generates subject teacher data by acquiring a plurality of account subjects and performing machine learning on which standard account subject they correspond to. The generation unit 12 may store the generated subject teacher data in the storage unit 14.

Next, an example of a method for supporting accounting audit will be explained.
FIG. 6 is a flowchart for explaining an example of a method for supporting accounting audit.

In step ST301, the acquisition unit 11 acquires target accounting data regarding the audited company's accounting. The acquisition unit 11 acquires the target accounting data via a communication network, or by reading the target accounting data stored in an external memory, for example.

In step ST302, the acquisition unit 11 applies the subject teacher data generated in step ST203 to the target accounting data acquired in step ST301, and converts the account subject (subject name) of the target accounting data into the reference account subject (subject name). Assign to.

In step ST303, the estimation unit 13 applies the test model generated in step ST105 to the target accounting data corresponding to the standard account item assigned in step ST302, and determines whether fraud such as falsification has been performed on the target accounting data. Get the height of sex. That is, the estimation unit 13 estimates the possibility of accounting fraud.

In step ST304, the display unit 15 displays the degree of possibility of fraud obtained in step ST303. The degree of possibility of fraud is indicated by, for example, a numerical value. For example, if the value is relatively high, there is a high possibility that fraud has been committed in the target accounting data. In other words, if the value is relatively low, there is a low possibility that fraud has occurred in the target accounting data.
In addition to the above-mentioned numerical values, the display unit 15 may also display a graph comparing the numerical value of any one of the plurality of account items of the target accounting data with a reference numerical value. The reference value may be, for example, the average value of the industry to which the company that created the target accounting data belongs.
In addition, for example, when target accounting data for multiple years is acquired from the audited company, the display unit 15 displays the return on equity (ROE) recorded in the target accounting data and the return on equity. It may also be possible to display trends over multiple years with net income, equity capital, etc. obtained from.
By referring to the probability of fraud displayed on the display unit 15, graphs, and trends in ROE, etc., the auditor can determine whether or not there is fraud in the accounting of the company being audited, and in which accounts there is a possibility of fraud. It is possible to determine whether there is a gender.
This step ST304 may be referred to as a fraud factor analysis step that supports factor analysis of fraudulent manipulation of accounting data based on the result estimated in step ST303.

The effects of this embodiment will be explained below.
The accounting audit support device 1 includes a storage unit 14 that stores a verification model used in accounting audits, an acquisition unit 11 that acquires target accounting data regarding the audited company's accounting, and based on the target accounting data and the verification model, The system includes an estimating section 13 that estimates the possibility that fraud is being committed in the audited company's accounting, and a display section 15 that displays the results estimated by the estimating section 13.
As a result, the accounting audit support device 1 can present to the auditor the possibility that accounting fraud, such as falsification, has occurred in the target accounting data based on machine learning, and can estimate the possibility of such fraud. can.
In addition, when generating a test model, the accounting audit support device 1 may, for example, classify fraudulent accounting data and hypothetical fraudulent accounting data labeled "fraud" and normal accounting data labeled "normal". Since learning is performed based on the learned fraud content, it is possible to estimate whether or not accounting fraud such as falsification has occurred in the target accounting data by focusing on the details of the learned fraud. Therefore, when the target accounting data is tampered with, the accounting audit support device 1 can estimate the tampered part (the tampered contents) by retracing the audit determination route of the test model.

Furthermore, if the ratio between the number of normal accounting data and the number of fraudulent accounting data is less than or equal to a preset threshold, the accounting audit support device 1 generates virtual fraudulent accounting data according to the content of the expected fraud. It is also possible to include a generation unit 12 that performs the following steps.
As a result, even if there is relatively little fraudulent accounting data when a company actually commits accounting fraud, the accounting audit support device 1 can estimate the falsification method used when a company commits accounting fraud and determine the virtual uncertainty. Accounting data can be generated. Note that the accounting audit support device 1 includes, for example, an auditor who conducts audits, a data manager who is in charge of acquiring and managing accounting data for generating a verification model, and a processing person who manages and adjusts the verification model. Based on the operations of a data analyst or the like, the generation unit 12 can generate virtual fraudulent accounting data. Therefore, the accounting audit support device 1 applies verification data generated based on learning virtual fraudulent accounting data, etc. to actual target accounting data, thereby detecting the possibility that fraud has been committed in the target accounting data. gender can be estimated.

In addition, the accounting audit support device 1 is based on subject accounting data and subject accounting data generated by learning a plurality of account subjects of accounting data and a standard account subject that is a basis for subject names of accounting data. Alternatively, the account items listed in the target accounting data may be assigned to the standard account items.
As a result, even if the account subject (subject name) of the target accounting data and the account subject (subject name) of the verification data are not the same, the accounting audit support device 1 can perform the target accounting based on the subject teacher data learned in advance. It is possible to change the account item (subject name) of accounting data to the corresponding account item (subject name) of verification data. Therefore, the accounting audit support device 1 can provide verification data to the target accounting data after changing the account item (subject name), and estimate the possibility that accounting fraud has occurred in the target accounting data. It is also possible to estimate where accounting fraud is occurring (contents of falsification).

The accounting audit support method includes a storing step of storing a verification model used for an accounting audit in the storage unit 14, an acquisition step of acquiring target accounting data regarding the accounting of the audited company, and a step of storing the target accounting data and the verification model in a computer. Based on this, an estimating step of estimating the possibility that fraud is being committed in the accounting of the audited company, and a displaying step of displaying the result estimated by the estimating step on the display unit 15 are executed.
As a result, the accounting audit support method can present to the auditor the possibility that accounting fraud, such as falsification, has occurred in the target accounting data based on machine learning, and can estimate the possibility of such fraud. .
In addition, when generating a test model, the accounting audit support method uses, for example, fraudulent accounting data and hypothetical fraudulent accounting data labeled "fraud" and normal accounting data labeled "normal". Since learning is performed based on the learned fraud, it is possible to estimate whether or not accounting fraud such as falsification has occurred in the target accounting data by focusing on the details of the learned fraud. Therefore, in the accounting audit support method, if fraud has been committed in the target accounting data, it is possible to estimate the tampered part (the content of the tampering) by retracing the audit determination route of the test model.

In the accounting audit support program, the computer has a memory function for storing the verification model used in the accounting audit, an acquisition function for acquiring target accounting data regarding the accounting of the audited company, and based on the target accounting data and the verification model, An estimation function for estimating the possibility of fraud being committed in the accounting of an audited company and a display function for displaying the results estimated by the estimation function are realized.
As a result, the accounting audit support program can present to auditors the possibility that accounting fraud, such as falsification, has occurred in the target accounting data based on machine learning, and can estimate the possibility of such fraud. .
In addition, when generating a test model, the accounting audit support program uses, for example, fraudulent accounting data and hypothetical fraudulent accounting data labeled "fraud" and normal accounting data labeled "normal". Since learning is performed based on the learned fraud, it is possible to estimate whether or not accounting fraud such as falsification has occurred in the target accounting data by focusing on the details of the learned fraud. Therefore, when the target accounting data is tampered with, the accounting audit support program can estimate the tampered part (the tampered contents) by retracing the audit determination route of the test model.

Each part of the accounting audit support device 1 described above may be realized as a function of a computer processing unit or the like. That is, the generation unit 12, the acquisition unit 11, and the estimation unit 13 of the accounting audit support device 1 may be realized as a generation function, an acquisition function, and an estimation function, respectively, by an arithmetic processing device of a computer.
The accounting audit support program can cause a computer to implement each of the functions described above. The accounting audit support program may be recorded on a computer-readable non-transitory recording medium such as an external memory or an optical disk.
Furthermore, as described above, each part of the accounting audit support device 1 may be realized by a computer processing device or the like. The arithmetic processing device and the like are constituted by, for example, an integrated circuit or the like. Therefore, each part of the accounting audit support device 1 may be realized as a circuit that constitutes an arithmetic processing device or the like. That is, the generation unit 12, acquisition unit 11, and estimation unit 13 of the accounting audit support device 1 may be realized as a generation circuit, an acquisition circuit, and an estimation circuit that constitute an arithmetic processing unit of a computer.
Further, the storage unit 14 and the display unit 15 of the accounting audit support device 1 may be realized as a storage circuit and a display circuit by being configured with an integrated circuit or the like, for example. Further, the storage unit 14 and the display unit 15 of the accounting audit support device 1 may be realized as a storage function and a display function, for example. Further, the storage unit 14 and the display unit 15 of the accounting audit support device 1 may be configured as a storage device and a display device by being configured with a plurality of devices, for example.

1 Accounting audit support device 11 Acquisition unit 12 Generation unit 13 Estimation unit 14 Storage unit 15 Display unit

Claims (7)

A first test model is generated by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting, and the first test model and the first test model are combined. a generation unit that generates a second test model based on the calculated values used when generating the second test model;
an acquisition department that acquires target accounting data regarding the audited company's accounting;
an estimation unit that estimates the possibility that fraud is being committed in the audited company's accounting based on the target accounting data and the second test model;
Accounting audit support device equipped with. The generation unit generates a multi-stage learned test model based on the first test model,
The accounting audit according to claim 1, wherein the estimation unit estimates the possibility that fraud is being committed in the audited company's accounting based on the target accounting data and the multi-stage learned test model. Support equipment. The accounting audit support device according to claim 1, wherein the generation unit registers a test index used when calculating the possibility of fraudulent accounting when performing machine learning to generate a test model. The accounting system according to claim 3, wherein the generation unit calculates a significant difference in each test index between a plurality of data and a correlation between test indices within the same data set, and registers the test index based on the result of the calculation. Audit support equipment. The accounting audit support device according to claim 1, further comprising a display unit that displays the result estimated by the estimation unit. The computer is
A first test model is generated by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting, and the first test model and the first test model are combined. a generation step of generating a second test model based on the calculated values used in the generation;
an acquisition step of acquiring target accounting data regarding the audited company's accounting;
an estimating step of estimating the possibility that fraud is being committed in the audited company's accounting based on the target accounting data and the second test model;
A method to support accounting audits. to the computer,
A first test model is generated by learning fraudulent accounting data when fraud is committed in accounting and normal accounting data in which fraud is not committed in accounting, and the first test model and the first test model are combined. a generation function that generates a second test model based on the calculated values used during generation;
an acquisition function that acquires target accounting data regarding the audited company's accounting;
an estimation function that estimates the possibility that fraud is being committed in the audited company's accounting based on the target accounting data and the second test model;
Accounting audit support program that realizes.

Images