WO2022065363A1

WO2022065363A1 - Fraudulent expense detection program

Info

Publication number: WO2022065363A1
Application number: PCT/JP2021/034797
Authority: WO
Inventors: 綾子澤田
Original assignee: Ａｓｓｅｓｔ株式会社
Priority date: 2020-09-24
Filing date: 2021-09-22
Publication date: 2022-03-31

Abstract

[Problem] To highly accurately and automatically distinguish the likelihood of fraud with less reliance on manpower. [Solution] A fraudulent expense detection program for detecting fraudulent expense, the fraudulent expense detection program being characterized in causing a computer to execute: an information acquisition step for monthly acquiring outsourcing information comprising outsourcing expenses; and a distinguishing step for distinguishing the likelihood of fraud by utilizing three or more levels of relevance between reference outsourcing information comprising outsourcing expenses acquired monthly in the past and the likelihood of fraud, and on the basis of the reference outsourcing information corresponding to the outsourcing information acquired in the information acquisition step, with priority given to higher levels of relevance.

Description

Fraud detection program

The present invention relates to a fraudulent expense detection program that detects fraudulent expenses from book data.

Traditionally, it should not be included in expenses, but the so-called tax evasion act of recording personally spent money in the books as expenses and reducing taxes is endless. It would be good if the tax accountant could find out the fraudulent expenses recorded on such books and urge the client to correct them, but it is not always possible to completely detect the fraudulent expenses. In addition, it is difficult for various institutions to completely find out fraudulent accounting for all of the huge number of companies because it requires considerable effort. Although there is a growing social demand to curb tax evasion by including such fraudulent expenses in the books, the current situation is that no technology has been devised to meet this demand.

For this reason, a system that can detect fraudulent expenses with high accuracy without relying on manual labor has been desired.

Therefore, the present invention has been devised in view of the above-mentioned problems, and an object thereof is to be able to detect fraudulent expenses with high accuracy without relying on manual labor. To provide a fraud detection program.

The fraudulent expense detection program according to the present invention is a fraudulent expense detection program for detecting fraudulent expenses. Using the three or more levels of association between the outsourcing information and the possibility of fraud, priority is given to the one with the higher degree of association based on the reference outsourcing information according to the outsourcing information acquired in the above information acquisition step. In the information acquisition step, the time-series change tendency of the outsourcing cost is acquired, and this is applied to a pre-categorized change tendency pattern. Using the above outsourced information, in the above determination step, there are three or more stages of association between the above-mentioned reference outsourced information in which the time-series change tendency of past outsourcing costs is applied to a pre-categorized change tendency pattern and the possibility of fraud. It is characterized by having a computer perform the use of degrees.

Even if you do not have any special skills or experience, anyone can easily detect fraudulent expenses with high accuracy without relying on manual work by humans.

It is a block diagram which shows the whole structure of the system to which this invention is applied. It is a figure which shows the specific configuration example of a search device. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention. It is a figure for demonstrating the operation of this invention.

Hereinafter, the fraudulent expense detection program to which the present invention is applied will be described in detail with reference to the drawings.

The first embodiment is a block diagram showing an overall configuration of a fraudulent expense detection system 1 in which a fraudulent expense detection program as the first embodiment is implemented. The fraudulent expense detection system 1 includes an information acquisition unit 9, a discrimination device 2 connected to the information acquisition unit 9, and a database 3 connected to the discrimination device 2.

The information acquisition unit 9 is a device for a person using this system to input various commands and information, and specifically, is composed of a keyboard, buttons, a touch panel, a mouse, a switch, and the like. The information acquisition unit 9 is not limited to a device for inputting text information, and may be configured by a device such as a microphone that can detect voice and convert it into text information. Further, the information acquisition unit 9 may be configured as an image pickup device capable of taking an image of a camera or the like. The information acquisition unit 9 may be configured by a scanner having a function of recognizing a character string from a paper-based document. Further, the information acquisition unit 9 may be integrated with the discrimination device 2 described later. The information acquisition unit 9 outputs the detected information to the discrimination device 2. Further, the information acquisition unit 9 may be configured by means for specifying the position information by scanning the map information. Further, the information acquisition unit 9 may be composed of an illuminance sensor for measuring a temperature sensor, a humidity sensor, and a wind direction sensor. Further, the information acquisition unit 9 may be configured by a communication interface for acquiring data about the weather from the Japan Meteorological Agency or a private weather forecast company. Further, the information acquisition unit 9 may be composed of a body sensor that is attached to the body to detect body data, and the body sensor detects, for example, body temperature, heart rate, blood pressure, number of steps, walking speed, and acceleration. It may be composed of a sensor for the purpose. Further, the body sensor may acquire biological data of not only humans but also animals. Further, the information acquisition unit 9 may be configured as a device for acquiring information such as drawings by scanning or reading from a database. In addition to these, the information acquisition unit 9 may be configured by an odor sensor that detects odors and scents.

Database 3 stores various information necessary for detecting fraudulent expenses. Information necessary for detecting fraudulent expenses includes outsourced information for reference, which consists of monthly outsourced expenses acquired monthly in the past, travel traffic information for reference, which consists of travel expenses and transportation expenses acquired monthly in the past, and monthly in the past. A data set of reference entertainment information consisting of entertainment expenses acquired for each and the possibility of fraud actually made for these is stored.

That is, in the database 3, any one or more of such reference outsourcing information, reference travel traffic information, and reference entertainment information and the possibility of fraud are stored in association with each other.

The discrimination device 2 is composed of, for example, an electronic device such as a personal computer (PC), but is embodied in any other electronic device such as a mobile phone, a smartphone, a tablet terminal, a wearable terminal, etc., in addition to the PC. It may be the one to be converted. The user can obtain a search solution by the discrimination device 2.

FIG. 2 shows a specific configuration example of the discrimination device 2. The discrimination device 2 performs wired communication or wireless communication with a control unit 24 for controlling the entire discrimination device 2 and an operation unit 25 for inputting various control commands via an operation button, a keyboard, or the like. A communication unit 26 for the purpose, a determination unit 27 for making various judgments, and a storage unit 28 for storing a program for performing a search to be executed represented by a hard disk or the like are connected to the internal bus 21, respectively. .. Further, a display unit 23 as a monitor that actually displays information is connected to the internal bus 21.

The control unit 24 is a so-called central control unit for controlling each component mounted in the discrimination device 2 by transmitting a control signal via the internal bus 21. Further, the control unit 24 transmits various control commands via the internal bus 21 according to the operation via the operation unit 25.

The operation unit 25 is embodied by a keyboard or a touch panel, and an execution command for executing a program is input from the user. When the execution command is input by the user, the operation unit 25 notifies the control unit 24 of the execution command. Upon receiving this notification, the control unit 24, including the discrimination unit 27, executes a desired processing operation in cooperation with each component. The operation unit 25 may be embodied as the information acquisition unit 9 described above.

The discrimination unit 27 discriminates the search solution. The discriminating unit 27 reads out various information stored in the storage unit 28 and various information stored in the database 3 as necessary information when executing the discriminating operation. The discriminating unit 27 may be controlled by artificial intelligence. This artificial intelligence may be based on any well-known artificial intelligence technology.

The display unit 23 is configured by a graphic controller that creates a display image based on the control by the control unit 24. The display unit 23 is realized by, for example, a liquid crystal display (LCD) or the like.

When the storage unit 28 is composed of a hard disk, predetermined information is written to each address based on the control by the control unit 24, and is read out as needed. Further, the storage unit 28 stores a program for executing the present invention. This program will be read and executed by the control unit 24.

The operation in the fraudulent expense detection system 1 having the above-mentioned configuration will be described.

In the fraud expense detection system 1, for example, as shown in FIG. 3, it is premised that the degree of association between the outsourced information for reference and the possibility of fraud is set in advance.

Reference outsourcing information is information related to outsourcing costs in the counterpart account (counterpart sub-account) in accounting software. This reference outsourcing information may be information on the breakdown (subcontractor, etc.) in addition to the outsourcing cost itself. This information may be extracted directly from the data of the counterpart account or the description in the accounting software, or actually extracted directly from the electronic data or paper data of the books, the electronic data or the paper data of the invoice. You may. When extracting from paper data, an image of the paper data may be read, and a character string may be extracted from the image using OCR technology and incorporated into text information.

By acquiring such reference outsourcing information on a monthly basis, it is also possible to acquire the changing trend of outsourcing costs on a monthly basis. This monthly change trend itself may be used as reference outsourcing information.

In addition, the outsourcing information for reference may be configured as a set by combining each breakdown (subcontractor, etc.) in addition to the monthly outsourcing cost itself. This makes it possible to classify and learn monthly outsourcing costs for each subcontractor.

The possibility of fraud here indicates the possibility of fraudulent expenses. If it is outsourced expenses, the expenses are recorded as outsourced expenses from a company that can not be considered as an outsourced company in the first place, or the way of increasing expenses is irregular, or the outsourced expenses are suddenly sharply close to the closing date to clearly reduce taxes. There is a high possibility of fraud, such as when the number is increasing. In addition, if a relative's company or the like has dropped a huge amount of expenses that cannot be considered in the normal market price, the possibility of fraud is high.

This possibility of fraud may be indicated by two stages of whether or not there is fraud, or is expressed by a ranking evaluated by the system side or the user side in five stages or ten or more stages. There may be. Alternatively, it may be simply expressed as extremely suspicious, suspicious, slightly suspicious, no problem, or the like.

The possibility of fraud may be based on the frequency of expenses formed on the book data and accounting data. In other words, the frequency of expenses is that in cases where outsourcing expenses are usually 100,000 yen per month, if outsourcing expenses are 2 million yen per month for a certain month, that 2 million yen. It can be determined that the month when the outsourcing cost is incurred is infrequent. Similarly, if the book data shows that there is a transaction with a company ●● as a subcontractor every month, but there is a transaction with another △△ company only in one month, the frequency of the subcontractor is in that month. It can be determined to be low. The possibility of fraud may be calculated based on the frequency of such expenses. In such a case, for example, the lower the frequency, the higher the possibility of fraud may be set.

The possibility of fraud may be judged based on the previous experience of an evaluator, a tax expert (tax accountant, accountant, etc.), or a person who has worked at a tax office, etc. It is also possible to extract the case judged to be, and judge from the actual book data and accounting data about it. In such a case, a plurality of inspectors who judge the possibility of fraud may evaluate the possibility of fraud in multiple stages for each preset item and statistically analyze them to obtain an evaluation value of the possibility of fraud. .. It may be judged through analysis. It was

In the example of FIG. 3, it is assumed that the input data is, for example, reference outsourcing information P01 to P03. The reference outsourcing information P01 to P03 as such input data is linked to the possibility of fraud as an output. In this output, the possibility of fraud as an output solution is displayed.

The outsourced information for reference is related to each other through the degree of association of 3 or more levels with respect to the possibility of fraud A to D as the output solution. The outsourced information for reference is arranged on the left side through this degree of association, and each possibility of fraud is arranged on the right side through this degree of association. The degree of association indicates the degree of fraud possibility and the degree of relevance to the outsourced reference information arranged on the left side. In other words, this degree of association is an indicator of what kind of fraud possibility each reference outsourced information is likely to be associated with, and is used to select the most probable fraud possibility from the reference outsourced information. It shows the accuracy in. In the example of FIG. 3, w13 to w19 are shown as the degree of association. These w13 to w19 are shown in 10 stages as shown in Table 1 below, and the closer to 10 points, the higher the degree of relevance of each combination as an intermediate node to the possibility of fraud as an output. On the contrary, the closer to one point, the lower the degree of relevance of each combination as an intermediate node to the possibility of fraud as an output.

The discrimination device 2 acquires in advance the degree of association w13 to w19 of three or more stages shown in FIG. That is, the discrimination device 2 accumulates the past data set, which of the outsourced information for reference and the possibility of fraud in that case is adopted and evaluated in discriminating the actual search solution, and stores these. By analyzing and analyzing, the degree of association shown in FIG. 3 is created.

For example, it is assumed that the fraud possibility A (for example, 70% fraud possibility) is highly evaluated as the fraud possibility for the reference outsourced information acquired from the book data and the accounting data in the past. By collecting and analyzing such data sets, the degree of association with the reference outsourced information becomes stronger.

This analysis may be performed by artificial intelligence. In such a case, for example, in the case of reference outsourcing information P01, analysis is performed from various data as a result of past evaluation of the possibility of fraud. If there are many cases of fraudulent possibility A in the case of outsourced information P01 for reference, the degree of association that leads to the evaluation of this fraudulent possibility is set higher, and fraudulent possibility B (for example, fraud possibility 20% 9) is set. When there are many cases of fraud, the degree of association that leads to the evaluation of this fraudulent possibility is set higher. For example, in the example of the reference outsourcing information P01, the fraudulent possibility A and the fraud possibility C are linked. From the previous case, the degree of association of w13 leading to the possibility of fraud A is set to 7 points, and the degree of association of w14 leading to the possibility of fraud C is set to 2 points.

Further, the degree of association shown in FIG. 3 may be composed of the nodes of the neural network in artificial intelligence. That is, the weighting coefficient for the output of the node of this neural network corresponds to the above-mentioned degree of association. Further, the network is not limited to a neural network, and may be composed of all decision-making factors constituting artificial intelligence.

In such a case, as shown in FIG. 4, reference outsourcing information is input as input data, rogue possibility is output as output data, and at least one hidden layer is provided between the input node and the output node. , Machine learning may be done. The above-mentioned degree of association is set in either one or both of the input node and the hidden layer node, and this is the weight of each node, and the output is selected based on this. Then, when this degree of association exceeds a certain threshold value, the output may be selected.

Such degree of association is what is called learned data in artificial intelligence. After creating such learned data through a data set of previously evaluated book data, accounting data, etc. and the possibility of fraud that was actually discriminated and evaluated, the possibility of fraud will be newly determined from now on. Above, the possibility of fraud is searched for using the above-mentioned trained data. In such a case, outsourcing information is newly acquired from the accounting data and book data to be discriminated. The newly acquired outsourced information is input by the above-mentioned information acquisition unit 9. This outsourcing information corresponds to the reference outsourcing information, and if the reference outsourcing information is a set of outsourcing costs and breakdown (subcontractor, etc.), the outsourcing to be acquired corresponding to this. Information is also obtained as a set of outsourcing costs and breakdown (subcontractors, etc.).

The above-mentioned learning data may be composed of only one company, or may be trained including those of other companies. Further, it is desirable that the company that newly detects fraud and the company that constitutes the learning data are the same, but the company is not limited to this.

Based on the newly acquired outsourcing information in this way, the possibility of fraud is determined. In such a case, the degree of association shown in FIG. 3 (Table 1) acquired in advance is referred to. For example, when the newly acquired outsourced information is the same as or similar to P02, the fraud possibility B is associated with w15 and the fraud possibility C is associated with the linkage degree w16 via the degree of association. In such a case, the fraud possibility B having a high degree of association is preferentially selected. That is, the higher the degree of association, the higher the priority of selection.

In this way, the most suitable possibility of fraud can be searched for and displayed to the user from the newly acquired outsourced information. By looking at the search results, the tax inspector can detect fraudulent expenses based on the searched fraudulent possibility, and can easily estimate the possibility of illegal activity.

In the above-mentioned example, the case where only the outsourcing cost is used as the reference outsourcing information and the outsourcing information, or the case where the outsourcing cost and the breakdown (subcontractor, etc.) are configured as a set is explained as an example, but the description is not limited to this. For example, the example of FIG. 5 shows a case where a combination of reference outsourcing cost information and reference breakdown information is formed. Here, the reference outsourcing cost information is composed only of the actual amount of the outsourcing cost, and the reference breakdown information is information showing the breakdown of the outsourcing cost, for example, showing the subcontractor.

By combining the reference breakdown information in addition to the reference outsourcing cost information, the possibility of fraud can be determined with higher accuracy. Therefore, in addition to the reference outsourcing cost information, the reference breakdown information is combined to form the above-mentioned degree of association.

In the example of FIG. 5, it is assumed that the input data is, for example, reference outsourcing cost information P01 to P03 and reference breakdown information P14 to 17. The intermediate node shown in FIG. 5 is a combination of reference breakdown information and reference outsourcing information as such input data. Each intermediate node is further linked to the output. In this output, the possibility of fraud as an output solution is displayed.

Each combination (intermediate node) of the outsourcing cost information for reference and the breakdown information for reference is related to each other through three or more levels of association with the possibility of fraud as this output solution. The reference outsourcing cost information and the reference breakdown information are arranged on the left side through this degree of association, and the possibility of fraud is arranged on the right side through this degree of association. The degree of association indicates the degree of high possibility of fraud and relevance to the reference outsourcing cost information and the reference breakdown information arranged on the left side. In other words, this degree of association is an index showing what kind of fraud possibility each reference outsourcing cost information and reference breakdown information is associated with, and is a reference subcontracting cost information and reference breakdown. It shows the accuracy in selecting the most probable fraud possibility from the information. Therefore, the optimum fraud possibility is searched for by combining the reference outsourcing cost information and the reference breakdown information.

In the example of FIG. 5, w13 to w22 are shown as the degree of association. As shown in Table 1, these w13 to w22 are shown in 10 stages, and the closer to 10 points, the higher the degree of relevance of each combination as an intermediate node to the output, and conversely, 1 point. The closer they are, the less relevant each combination as an intermediate node is to the output.

The discrimination device 2 acquires in advance the degree of association w13 to w22 of three or more stages shown in FIG. That is, the discrimination device 2 accumulates past data as to which of the reference outsourcing cost information, the reference breakdown information, and the possibility of fraud in that case is suitable for discriminating the actual search solution. By analyzing and analyzing these, the degree of association shown in FIG. 5 is created.

For example, it is assumed that the outsourcing cost information for reference in the actual case in the past suddenly increases at the end of the fiscal year. Further, it is assumed that the breakdown information for reference is the subcontractor K company, which has no transaction so far. In such a case, the fraudulent possibility indicating how much the fraudulent possibility was actually learned is learned as a data set, and is defined in the form of the above-mentioned degree of association. In addition, such reference outsourcing cost information and reference breakdown information should be extracted from accounting software, accounting data held by each tax accountant office, accounting office, and databases held by the tax office. You may do it.

This analysis may be performed by artificial intelligence. In such a case, for example, in the case of the reference outsourcing cost information P01 and the reference breakdown information P16, the possibility of fraud is analyzed from the past data. If there are many cases of fraudulent possibility A, the degree of association leading to this fraudulent possibility A is set higher, and if there are many cases of fraudulent possibility B and there are few cases of fraudulent possibility A, it is fraudulent. The degree of association leading to possibility B is set high, and the degree of association leading to possibility A is set low. For example, in the example of the intermediate node 61a, it is linked to the output of the possibility of fraud A and B, but from the previous case, the degree of association of w13 leading to the possibility of fraud A is set to 7 points, and the degree of association of w13 leading to the possibility of fraud B is set to 7 points. The degree of association is set to 2 points.

Further, the degree of association shown in FIG. 5 may be composed of the nodes of the neural network in artificial intelligence. That is, the weighting coefficient for the output of the node of this neural network corresponds to the above-mentioned degree of association. Further, the network is not limited to a neural network, and may be composed of all decision-making factors constituting artificial intelligence. Other than that, the configuration related to artificial intelligence is the same as the description in FIG.

In the example of the degree of association shown in FIG. 5, the node 61b is a node in which the reference breakdown information P14 is combined with the reference outsourcing cost information P01, the linkage degree of the fraud possibility C is w15, and the fraud possibility E. The degree of association is w16. The node 61c is a node that is a combination of the reference breakdown information P15 and P17 with respect to the reference outsourcing cost information P02, and the degree of association of the possibility of fraud B is w17 and the degree of association of the possibility of fraud D is w18. There is.

Such degree of association is what is called learned data in artificial intelligence. After creating such learned data, when actually determining the possibility of fraud from now on, the above-mentioned learned data will be used. In such a case, input or select outsourcing cost information and breakdown information for which the possibility of fraud is actually determined.

Based on the newly acquired outsourcing cost information and breakdown information in this way, search for the optimal possibility of fraud. In such a case, the degree of association shown in FIG. 5 (Table 1) acquired in advance is referred to. For example, when the newly acquired outsourcing cost information is the same as or similar to P02 and the breakdown information is P17, the node 61d is associated via the degree of association. The node 61d is associated with the possibility of fraud C by w19 and the possibility of fraud D by the degree of association w20. In such a case, the fraud possibility C having a higher degree of association is selected as the optimum solution.

Table 2 below shows an example of the degree of association w1 to w12 extending from the input.

The intermediate node 61 may be selected based on the degree of association w1 to w12 extending from this input. That is, the larger the degree of association w1 to w12, the heavier the weighting in the selection of the intermediate node 61 may be. However, the association degrees w1 to w12 may all have the same value, and the weightings in the selection of the intermediate node 61 may all be the same.

Further, in the present invention, with respect to the above-mentioned outsourcing information and reference outsourcing information (outsourcing cost information, reference outsourcing cost information), a time-series change tendency is acquired, and the change tendency pattern categorized in advance is obtained. It may be configured by applying to. The time-series change tendency includes, but is not limited to, a monthly change tendency, and may be any time-series such as weekly or daily.

FIG. 6 shows an example of a time-series change trend pattern categorized in advance, with the horizontal axis representing time (month, week, etc.) and the vertical axis representing outsourcing costs. For example, FIG. 6A shows a pattern in which up and down movements are repeated while the outsourcing cost rises. Further, FIG. 6B shows a pattern in which the outsourcing cost exceeds the upper resistance line after the conflict continues for a long time. FIGS. 6 (c) and 6 (d) show a pattern in which outsourcing costs fluctuate sharply.

When applying the acquired time-series change tendency to a change tendency pattern categorized in advance, a judgment model generated by machine learning as shown in FIG. 7 may be used. In this determination model, an image of a change tendency pattern is used as teacher data. The input is a time-series change tendency acquired, and the output is a typified pattern. When a new time-series change tendency of outsourcing costs is acquired, fitting is performed based on the judgment model generated from this machine learning, and it is determined what type of change tendency pattern is applied.

As a result, this reference outsourcing information will be represented by a typified change trend pattern. By learning the possibility of fraud for such a change tendency pattern, the above-mentioned degree of association is formed.

Next, even when a new trend of change in outsourcing costs is actually acquired over time, it is determined what type the acquired trend of change applies to, for example, through the determination model shown in FIG. 7, and each reference is made. We will apply the outsourced information to the types of change trend patterns. As a result, this outsourced information will be represented by a typified change trend pattern. It is determined through the above-mentioned degree of association that the type of change tendency is applicable to what type of change tendency pattern of the outsourced information for reference. Then, using the three or more levels of association between the type of change tendency pattern of reference outsourced information corresponding to the type of change tendency pattern of outsourced information and each possibility of fraud, priority is given to the one with higher degree of association. , Judge the possibility of fraud.

In the above example, outsourcing costs are taken as an example for explanation, but the description is not limited to this, and any other offsetting account can be applied in the same manner.

For example, in the fraudulent expense detection system 1, as a substitute for the reference outsourced information, the reference travel traffic information consisting of the travel expense and transportation expenses acquired monthly in the past and the degree of association between the possibility of fraud and three or more levels are used. You may.

The reference travel traffic information referred to here is information related to travel expenses and transportation expenses in the counterpart account (counterpart sub-account) in the accounting software. In addition to the travel expenses themselves, this reference travel transportation information includes, for example, the type of transportation used (train, taxi), the name of the employee who actually used the transportation, and the details of the transportation used ( For example, if it is a train, it may be information about the Shinkansen, some private railway lines, etc.), the regular fee, the prepaid card, and the like. This information may be obtained by directly extracting the data of the counterpart account and the description in the accounting software, or actually from the electronic data and paper data of books, receipts and receipts, and electronic data and paper data of invoices. It may be extracted directly. When extracting from paper data, an image of the paper data may be read, and a character string may be extracted from the image using OCR technology and incorporated into text information.

By acquiring such reference travel traffic information on a monthly basis, it is also possible to acquire the trend of changes in outsourcing costs on a monthly basis. This monthly change tendency itself may be used as reference travel traffic information.

In addition, the reference travel traffic information may be configured as a set by combining each breakdown with the monthly travel expenses and transportation expenses themselves. This makes it possible to classify and learn monthly travel expenses and transportation expenses by breakdown.

When using such reference travel traffic information, as a substitute for the reference outsourced information shown in FIGS. 3 and 4, the reference travel traffic information is similarly learned in relation to the possibility of fraud. After creating such learned data through a data set of previously evaluated book data, accounting data, etc. and the possibility of fraud that was actually discriminated and evaluated, the possibility of fraud will be newly determined from now on. Above, the possibility of fraud is searched for using the above-mentioned trained data. In such a case, travel traffic information is newly acquired from the accounting data and book data to be discriminated. The newly acquired travel traffic information is input by the above-mentioned information acquisition unit 9. This travel traffic information corresponds to the reference travel traffic information, and if the reference travel traffic information is a set of the travel expenses and the breakdown, the travel traffic to be acquired corresponding to this. Information is also obtained as a set of travel expenses, transportation expenses and breakdown. Then, the possibility of fraud with a higher degree of association is derived as a search solution for the reference travel traffic information according to the acquired travel traffic information. Since the details of these processing operations are the same as in the case of the above-mentioned travel traffic information and reference travel traffic information, the description below will be omitted by quoting the explanations of FIGS. 3 and 4.

Similarly, when using the reference travel traffic information and the travel traffic information, the explanation will be given by taking as an example the case where only the travel expense transportation cost is used as the reference travel traffic information and the travel traffic information, or the travel expense transportation cost and the breakdown are configured as a set. However, it is not limited to this. For example, as shown in the example of FIG. 5, a combination of the reference travel expense information and the reference breakdown information may be used. Here, the reference travel expense transportation expense information is composed only of the actual amount of the travel expense transportation expense, and the reference breakdown information is information showing the breakdown of the outsourcing expense, for example, indicating the subcontractor. The processing operation of the case is also described below by replacing the reference outsourcing cost information shown in FIG. 5 with the reference travel cost transportation cost information.

Further, in the case of reference travel traffic information, as shown in FIG. 6, it may be applied to a pre-categorized time-series change tendency pattern. Further, when applying the acquired time-series change tendency to the change tendency pattern categorized in advance, a determination model generated by machine learning as shown in FIG. 7 may be used.

The reference travel traffic information will be represented by a categorized change trend pattern. By learning the possibility of fraud for such a change tendency pattern, the above-mentioned degree of association is formed.

Next, even when a new time-series change tendency of travel expenses and transportation expenses is actually acquired, it is determined what type the acquired change tendency applies to, for example, through the determination model shown in FIG. 7, and each of them is determined. We will apply the reference travel traffic information to the types of changing trend patterns. As a result, this travel traffic information will be represented by a typified change trend pattern. It is determined through the above-mentioned degree of association that the type of change tendency is applicable to what type of change tendency pattern of the reference travel traffic information. Then, using the three or more levels of association between the type of change tendency pattern of the reference travel traffic information corresponding to the type of change tendency pattern of travel traffic information and each possibility of fraud, priority is given to the one with higher degree of association. And determine the possibility of fraud.

For example, in the fraudulent expense detection system 1, as a substitute for the reference outsourced information, the reference entertainment information consisting of the entertainment expenses acquired monthly in the past and the degree of association with the possibility of fraud are used in three or more stages. May be good.

The reference entertainment information referred to here is information related to entertainment expenses in the counterpart account (counterpart sub-account) in the accounting software. The entertainment information for reference may be, for example, information on the business partner, information on the business partner, and information on the restaurant used, as a breakdown, in addition to the entertainment expenses themselves. This information may be extracted directly from the data of the counterpart account or the description in the accounting software, or actually extracted directly from the electronic data or paper data of the books, the electronic data or the paper data of the invoice. You may. When extracting from paper data, an image of the paper data may be read, and a character string may be extracted from the image using OCR technology and incorporated into text information.

By acquiring such reference entertainment information on a monthly basis, it is also possible to acquire the trend of changes in outsourcing costs on a monthly basis. This monthly change trend itself may be used as reference entertainment information.

In addition, the entertainment information for reference may be configured as a set by combining each breakdown in addition to the monthly entertainment expenses themselves. This makes it possible to classify and learn monthly entertainment expenses by breakdown.

When using such reference entertainment information, as an alternative to the reference outsourcing information shown in FIGS. 3 and 4, the reference entertainment information is similarly learned in relation to the possibility of fraud. After creating such learned data through a data set of previously evaluated book data, accounting data, etc. and the possibility of fraud that was actually discriminated and evaluated, the possibility of fraud will be newly determined from now on. Above, the possibility of fraud is searched for using the above-mentioned trained data. In such a case, entertainment information is newly acquired from the accounting data and book data to be discriminated. The newly acquired entertainment information is input by the above-mentioned information acquisition unit 9. This entertainment information corresponds to the entertainment information for reference, and if the entertainment information for reference corresponds to the entertainment expenses and the breakdown, the entertainment information to be acquired is also the entertainment entertainment. Get expenses and breakdown as a set. Then, the possibility of fraud with a higher degree of association is derived as a search solution for the entertainment information for reference according to the acquired entertainment information. Since the details of these processing operations are the same as in the case of the entertainment information and the entertainment information for reference described above, the description below will be omitted by quoting the explanations of FIGS. 3 and 4.

Similarly, when using the entertainment information for reference and the entertainment information, the case where only the entertainment expenses for reference and the entertainment information are used, or the entertainment expenses and the breakdown are configured as a set has been described as an example. Not limited to. For example, as shown in the example of FIG. 5, a combination of the entertainment expense information for reference and the breakdown information for reference may be used. Here, the reference entertainment expense information is composed only of the actual amount of the entertainment expense, and the reference breakdown information is information showing the breakdown of the outsourcing expense, for example, indicating the subcontractor. The processing operation of the case will also be described below by replacing the reference outsourcing cost information shown in FIG. 5 with the reference entertainment cost information.

Further, in the case of entertainment information for reference, as shown in FIG. 6, it may be applied to a time-series change tendency pattern categorized in advance. Further, when applying the acquired time-series change tendency to the change tendency pattern categorized in advance, a determination model generated by machine learning as shown in FIG. 7 may be used.

The entertainment information for reference will be represented by a categorized change tendency pattern. By learning the possibility of fraud for such a change tendency pattern, the above-mentioned degree of association is formed.

Next, even when a new trend of change in entertainment expenses over time is actually acquired, it is determined what type the acquired trend of change applies to, for example, through the determination model shown in FIG. 7, and each of them is determined. We will apply the entertainment information for reference to the types of change trend patterns. As a result, this entertainment information will be represented by a typified change trend pattern. It is determined through the above-mentioned degree of association that the type of change tendency is applicable to what type of change tendency pattern of the entertainment information for reference. Then, using the three or more levels of association between the type of the change tendency pattern of the entertainment information for reference corresponding to the type of the change tendency pattern of the entertainment information and each possibility of fraud, priority is given to the one with the higher degree of association. , Judge the possibility of fraud.

The example of FIG. 8 shows a case where a combination of reference outsourcing cost information and reference travel traffic information is formed.

By combining the reference travel traffic information in addition to the reference outsourcing cost information, the possibility of fraud can be determined with higher accuracy. Therefore, in addition to the reference outsourcing cost information, the reference travel traffic information is combined to form the above-mentioned degree of association.

In the example of FIG. 8, it is assumed that the input data is, for example, reference outsourcing cost information P01 to P03 and reference travel traffic information P14 to 17. The intermediate node shown in FIG. 8 is a combination of reference travel traffic information and reference outsourced information as such input data. Each intermediate node is further linked to the output. In this output, the possibility of fraud as an output solution is displayed.

Each combination of reference outsourcing cost information and reference travel traffic information (intermediate node) is associated with each other through three or more levels of association with the possibility of fraud as this output solution. The reference outsourcing cost information and the reference travel traffic information are arranged on the left side through this degree of association, and the possibility of fraud is arranged on the right side through this degree of association. In particular, fraudulent expenses may correlate with outsourcing expenses and travel expenses, so the possibility of fraud can be determined with high accuracy by learning by combining these.

The discrimination device 2 acquires in advance the degree of association w13 to w22 of three or more stages shown in FIG. That is, the discrimination device 2 accumulates past data as to which of the reference outsourcing cost information, the reference travel traffic information, and the possibility of fraud in that case is suitable for discriminating the actual search solution. By analyzing and analyzing these, the degree of association shown in FIG. 8 is created.

In the example of the degree of association shown in FIG. 8, the node 61b is a node of the combination of the reference travel traffic information P14 with respect to the reference outsourcing cost information P01, and the degree of linkage of the fraud possibility C is w15 and the fraud possibility. The degree of association of E is w16. The node 61c is a node that is a combination of the reference travel traffic information P15 and P17 with respect to the reference outsourcing cost information P02, and the degree of association of the possibility of fraud B is w17 and the degree of association of the possibility of fraud D is w18. ing.

Such degree of association is what is called learned data in artificial intelligence. After creating such learned data, when actually determining the possibility of fraud from now on, the above-mentioned learned data will be used. In such a case, input or select outsourcing cost information and travel traffic information for which the possibility of fraud is actually determined.

Based on the newly acquired outsourcing cost information and travel traffic information in this way, search for the optimal possibility of fraud. The details of this search method are the same as those described in FIG.

It should be noted that the combination of the reference outsourced information and the reference entertainment information is not limited to the case where the degree of association of the possibility of fraud with the combination of the reference outsourced information and the reference travel traffic information is learned in advance. The degree of association of the possibility of fraud with the reference may be learned in advance, or the degree of association of the possibility of fraud with the combination of the reference travel traffic information and the reference entertainment information may be learned in advance. Then, the outsourcing information, the entertainment information, and the travel traffic information corresponding to each reference information constituting these combinations are acquired from the book data and the accounting data of the company to be determined, and the determination is made in the same manner.

In addition, the degree of association of the possibility of fraud with the combination of the reference outsourcing information, the reference entertainment information, and the reference travel traffic information may be learned in advance. In such a case, outsourcing information, entertainment information, and travel traffic information are acquired from the book data and accounting data of the company to be determined, and the determination is made in the same manner.

In the above-described embodiment, the case of learning outsourced information, entertainment information, and travel traffic information has been described as an example, but the description is not limited to this, and any counterpart account (for example, meeting fee, etc.) Personnel costs, newspaper book costs, consumables costs, communication costs, payment fees, membership fees, office supplies costs, welfare costs, etc.) should be used as reference information to learn in relation to the possibility of fraud. You may do it. As a result, it is possible to determine the possibility of fraud by inputting the information of the counterpart account according to the learned reference information and performing the solution search in the same manner.

Further, according to the present invention, there is a feature that the optimum solution search is performed through the degree of association set in three or more stages. The degree of association can be described by a numerical value from 0 to 100%, for example, in addition to the above-mentioned 10 levels, but is not limited to this, and any stage can be described as long as it can be described by a numerical value of 3 or more levels. It may be configured.

By discriminating the most probable fraud possibility based on the degree of association expressed by the numerical values of three or more stages, the degree of association is high under the situation where there are multiple possible candidates for the search solution. It is also possible to search and display in order. If the user can be displayed in descending order of the degree of association in this way, it is possible to preferentially display more probable search solutions.

In addition to this, according to the present invention, it is possible to judge without overlooking the discrimination result of the extremely low output such as the degree of association of 1%. Remind the user that even a judgment result with an extremely low degree of association is connected as a slight sign, and may be useful as the judgment result once every tens or hundreds of times. be able to.

Further, according to the present invention, there is a merit that the search policy can be determined by the method of setting the threshold value by performing the search based on the degree of association of three or more stages. If the threshold value is lowered, even if the above-mentioned degree of association is 1%, it can be picked up without omission, but it is unlikely that a more appropriate discrimination result can be detected favorably, and a lot of noise may be picked up. be. On the other hand, if the threshold value is raised, there is a high possibility that the optimum search solution can be detected with high probability, but the degree of association is usually low and it is passed through, but it is suitable to appear once in tens or hundreds of times. Sometimes the solution is overlooked. It is possible to decide which one should be emphasized based on the ideas of the user side and the system side, but it is possible to increase the degree of freedom in selecting the points to be emphasized.

Further, in the present invention, the above-mentioned degree of association may be updated. This update may reflect information provided, for example, via a public communication network such as the Internet. In addition, when each reference information such as reference outsourcing information is acquired and knowledge, information, and data regarding the possibility of fraud and improvement measures are acquired, the degree of association is increased or decreased according to these.

In other words, this update is equivalent to learning in terms of artificial intelligence. It can be said that it is a learning act because it acquires new data and reflects it in the learned data.

In addition, this update of the degree of association is done by the system side or the user side based on the contents of research data, treatises, conference presentations, newspaper articles, books, etc. by experts, except when it is based on information that can be obtained from public communication networks. It may be updated artificially or automatically. Artificial intelligence may be utilized in these update processes.

In addition, the process of first creating a trained model and the above-mentioned update may use not only supervised learning but also unsupervised learning, deep learning, reinforcement learning, and the like. In the case of unsupervised learning, instead of reading and training the data set of input data and output data, information corresponding to the input data is read and trained, and the degree of association related to the output data is self-formed from there. You may let it.

Second Embodiment Hereinafter, the second embodiment will be described. In executing this second embodiment, the fraudulent expense detection system 1, the information acquisition unit 9, the discrimination device 2, and the database 3 used in the first embodiment are similarly used. The description of each of these configurations will be omitted below by citing the description of the first embodiment.

In the second embodiment, for example, as shown in FIG. 9, it is premised that three or more levels of association with the possibility of fraud are preset for the combination of the reference attribute information and the reference breakdown information. Become. It was

In the example of FIG. 9, it is assumed that the input data is, for example, reference attribute information P01 to P03 and reference breakdown information P14 to 17. The intermediate node shown in FIG. 9 is a combination of the reference attribute information and the reference breakdown information as such input data. Each intermediate node is further linked to the output. In this output, the possibility of fraud as an output solution is displayed.

The reference attribute information referred to here is information related to the attributes of the company that acquired the book data in the past. The attributes of this company are industry, business content, business content, capital, address, year of establishment, number of employees, sales transition, profit transition, and other financial information (settlement date / month, sales, operating profit, ordinary income). Profit, net income, net income per share, dividend per share, shareholders'equity per share, number of issued shares, total assets, shareholders'equity, capital, interest-bearing debt, gains / losses carried forward, retained earnings, shareholders' equity ratio, unrealized gains / losses , ROA, ROE, total asset ordinary profit margin), etc. This reference attribute information relates to the attribute of the company that acquired the reference breakdown information.

Each combination of reference attribute information and reference breakdown information (intermediate node) is associated with each other through three or more levels of association with the possibility of fraud as this output solution. The reference attribute information and the reference breakdown information are arranged on the left side through this degree of association, and the possibility of fraud is arranged on the right side through this degree of association. The degree of association indicates the degree of high possibility of fraud and relevance to the reference attribute information and the reference breakdown information arranged on the left side. In other words, this degree of association is an index showing what kind of fraudulent possibility each reference attribute information and reference breakdown information are associated with, and is based on the reference attribute information and reference breakdown information. It shows the accuracy in selecting the most probable fraudulent possibility. Therefore, the optimum fraud possibility is searched for by combining the reference attribute information and the reference breakdown information.

In the example of FIG. 9, w13 to w22 are shown as the degree of association. As shown in Table 1, these w13 to w22 are shown in 10 stages, and the closer to 10 points, the higher the degree of relevance of each combination as an intermediate node to the output, and conversely, 1 point. The closer they are, the less relevant each combination as an intermediate node is to the output.

The discrimination device 2 acquires in advance the degree of association w13 to w22 of three or more stages shown in FIG. That is, the discriminating device 2 accumulates past data as to which of the reference attribute information, the reference breakdown information, and the possibility of fraud in that case is suitable for discriminating the actual search solution. By analyzing and analyzing these, the degree of association shown in FIG. 9 is created.

For example, in the reference attribute information in the actual case in the past, when the company of the book data is a retail store and the business format mainly sells clothes, the subcontractor is the breakdown of the subcontracting cost. For example, in the case of a food processing factory, if the business format and the subcontractor are clearly inconsistent, the possibility of fraud increases. On the other hand, if the subcontractor is a subcontractor of the cloth for the clothes or a contractor who specializes in embroidery, the possibility of fraud is low because the business format and the subcontractor match. In any case, the fraudulent possibility, which indicates how much the fraudulent possibility was actually, is learned as a data set and defined in the form of the above-mentioned degree of association. It should be noted that such reference attribute information and reference breakdown information should be extracted from accounting software, accounting data held by each tax accountant office, accounting office, and a database held by the tax office. You may.

This analysis may be performed by artificial intelligence. In such a case, for example, in the case of the reference attribute information P01 and the reference breakdown information P16, the possibility of fraud is analyzed from the past data. If there are many cases of fraudulent possibility A, the degree of association leading to this fraudulent possibility A is set higher, and if there are many cases of fraudulent possibility B and there are few cases of fraudulent possibility A, it is fraudulent. The degree of association leading to possibility B is set high, and the degree of association leading to possibility A is set low. For example, in the example of the intermediate node 61a, the output of the possibility of fraud A and B is linked. The degree of association is set to 2 points.

Further, the degree of association shown in FIG. 9 may be composed of the nodes of the neural network in artificial intelligence. That is, the weighting coefficient for the output of the node of this neural network corresponds to the above-mentioned degree of association. Further, the network is not limited to a neural network, and may be composed of all decision-making factors constituting artificial intelligence. Other than that, the configuration related to artificial intelligence is the same as the description in FIG.

In the example of the degree of association shown in FIG. 9, the node 61b is a node in which the reference attribute information P01 is combined with the reference breakdown information P14, the degree of association of the possibility C is w15, and the degree of association E is the possibility E. The degree of association is w16. The node 61c is a node that is a combination of the reference breakdown information P15 and P17 with respect to the reference attribute information P02, and the degree of association of the possibility of fraud B is w17 and the degree of association of the possibility of fraud D is w18. ..

Such degree of association is what is called learned data in artificial intelligence. After creating such learned data, when actually determining the possibility of fraud from now on, the above-mentioned learned data will be used. In such a case, input or select the breakdown information of the subcontractor from the attribute information of the company that actually tries to determine the possibility of fraud and the book data of the company.

Based on the newly acquired attribute information and breakdown information in this way, search for the optimal possibility of fraud. In such a case, the degree of association shown in FIG. 9 (Table 1) acquired in advance is referred to. For example, when the newly acquired attribute information is the same as or similar to P02 and the breakdown information is P17, the node 61d is associated via the degree of association, and this node. In 61d, the fraud possibility C is associated with w19, and the fraud possibility D is associated with the degree of association w20. In such a case, the fraud possibility C having a higher degree of association is selected as the optimum solution.

In the present invention, the database 3 in which the correspondence between each subcontractor and its type may be referred to may be referred to. In this database 3, the types to which all the traders belong are stored in correspondence with each other. The type referred to here may indicate, for example, an industry or what kind of industry it belongs to, from rough ones such as logistics, retail, and manufacturing, to clothing, food, and even in the same retail industry. It may be subdivided into household goods, golf equipment, etc., and even within the same manufacturing industry, from the middle classification of electronic equipment, mold makers, furniture makers, etc., even if the same furniture makers, beds, tons, tables, etc. It may be subdivided to a level.

By referring to such a database 3, the type may be acquired based on the subcontractor actually described in the reference breakdown information, and this may be included in the reference breakdown information.

In such a case, the subcontractor is extracted, the type associated with the subcontractor's vendor is read from the above database 3, and this is included in the reference breakdown information.

As a result, if the subcontractor is a subcontractor of furniture manufacturing even though the company is an electronic device manufacturer from the attribute information, an inconsistency will obviously occur, and there is a possibility of fraud. The higher the weight, the heavier the weighting of the degree of association. If the consistency cannot be checked in relation to the attribute information only with the company name of the subcontractor described in the breakdown information, the type of subcontractor described in the breakdown information by referring to this database 3 Can be determined.

In constructing this breakdown information and the breakdown information for reference, the type of the subcontractor referred from the database 3 may be acquired each time and included.

In FIG. 10, in addition to the above-mentioned reference attribute information and reference breakdown information, a combination of reference external environment information and three or more levels of association with each exchange increase / decrease data for the combination are set. An example is shown. External environmental information for reference is GDP, employment statistics, industrial production index, capital investment, labor force survey, economic trend index, consumer spending, new car sales, consumer price index, Nikkei average stock price, exchange rate outside the company. Includes various data on politics, economy, society, technology, etc.

In such a case, as shown in FIG. 10, the degree of association is such that the set of combinations of the reference attribute information, the reference breakdown information, and the reference external environment information is set as the nodes 61a to 61e of the intermediate node as described above. It will be expressed.

For example, in FIG. 10, in FIG. 10, the reference attribute information P02 is associated with the association degree w3, the reference breakdown information P15 is associated with the association degree w7, and the reference external environment information P19 is associated with the association degree w11. Similarly, in the node 61e, the reference attribute information P03 is associated with the association degree w5, the reference breakdown information P15 is associated with the association degree w8, and the reference external environment information P18 is associated with the association degree w10.

Similarly, even when such a degree of association is set, the search solution is determined based on the newly acquired attribute information, the breakdown information, and the external environment information.

In determining this search solution, refer to the degree of association shown in FIG. 10 acquired in advance. For example, when the acquired attribute information is the same as or similar to the reference attribute information P02, the acquired breakdown information corresponds to the reference breakdown information P15, and the acquired external environment information corresponds to the reference external environment information P19. In the combination, the node 61c is associated with the node 61c, and the increase / decrease data A2 is associated with the association degree w17, and the increase / decrease data A4 is associated with the association degree w18. As a result of such a degree of association, a search solution is actually obtained based on w17 and w18.

As an alternative to the external environment information in FIG. 10, the above-mentioned reference outsourcing cost information may be combined as shown in FIG. As a result, it is possible to make a highly accurate determination in consideration of the subcontractor, its breakdown, and the attributes of the company, including the subcontracting cost itself.

Further, regarding this outsourcing cost, a time-series change trend may be acquired, and as shown in FIGS. 6 and 7, this may be categorized into each pattern to form outsourcing cost information and reference outsourcing cost information. Of course.

In this second embodiment, the description is given by taking as an example the outsourcing cost in the offsetting account item as the reference breakdown information and the breakdown information, but the description is not limited to this, and the above-mentioned travel expense and transportation expense are described. The same applies when the breakdown is configured as breakdown information.

In such a case, there are three or more stages: a combination of reference attribute information related to the attributes of the company acquired in the past, reference breakdown information consisting of the breakdown of travel expenses and transportation expenses in the book data of the company, and the possibility of fraud. The degree of association is formed as shown in FIG. 9 by using the degree of association of. Then, the breakdown information consisting of the breakdown of the travel expenses and the transportation expenses in the book data of the company to be determined for fraud is actually acquired, and the possibility of fraud as a search solution is obtained as in the explanation in FIG. Travel Expenses In the breakdown of transportation expenses, travel routes and means of transportation may be inconsistent in relation to the location of the company, and the destination may clearly show inconsistency in relation to the actual industry of the company. Therefore, it is possible to detect the possibility of fraud with high accuracy based on this information.

At this time, refer to the database that stores the correspondence between each transportation name and its type, acquire the type based on the transportation name as the breakdown of the above travel expenses, and include this in the breakdown information. You may do so. Further, based on the transportation name as the breakdown of the travel expenses and transportation expenses in the book data of the company, the type may be acquired and used as the reference breakdown information. For example, the transportation names include the Shinkansen, JR Yamanote Line, private railway Tokyu Line, buses, etc., and the breakdown includes stop stations, stop times, departure and arrival times, and the like.

In this second embodiment, the breakdown information for reference and the breakdown information are described by taking the outsourcing cost in the offsetting account item as an example, but the description is not limited to this, and the above-mentioned entertainment transportation cost is described. The same applies when the breakdown is configured as breakdown information.

In such a case, there are three or more stages of the combination of the reference attribute information related to the attributes of the company acquired in the past and the reference breakdown information consisting of the breakdown of entertainment expenses in the book data of the company, and the possibility of fraud. The degree of association is formed as shown in FIG. 9 by using the degree of association of. Then, the breakdown information consisting of the breakdown of the entertainment expenses in the book data of the company to be determined for fraud is actually acquired, and the possibility of fraud as a search solution is obtained as in the explanation in FIG. In the breakdown of entertainment expenses, the entertainment destination may be inconsistent in the relationship between the business type and type of the company, and the restaurant used for entertainment clearly shows inconsistency in the relationship with the actual business type of the company. In some cases, it is possible to detect the possibility of fraud with high accuracy based on this information. Further, even if the restaurant used for entertainment is a sex shop, a fast food restaurant, or the like, which is not suitable for the actual entertainment, it can be detected from the breakdown information.

At this time, referring to the database 3 in which the correspondence between each entertainment destination and its type is stored, the type is acquired based on the entertainment destination as the breakdown of the entertainment expenses, and this is included in the breakdown information. You may. Further, the database 3 may be referred to, and the type may be acquired based on the restaurant used for entertainment as a breakdown of the entertainment expenses in the book data of the company. Thereby, the type of business and the type of the entertainment destination company can be easily specified from the database 3, and the type of the restaurant used for the entertainment can be easily specified.

Further, in the second embodiment, any reference information corresponding to the input of the neural network described in the first embodiment may be combined to search for the possibility of fraud corresponding to the output.

1 Fraudulent expense detection system 2 Discrimination device 21 Internal bus 23 Display unit 24 Control unit 25 Operation unit 26 Communication unit 27 Discrimination unit 28 Storage unit 61 Node

Claims

In a fraudulent expense detection program that detects fraudulent expenses
An information acquisition step to acquire outsourcing information consisting of outsourcing costs on a monthly basis,
Based on the reference outsourcing information according to the outsourcing information acquired in the above information acquisition step, using the reference outsourcing information consisting of outsourcing costs acquired monthly in the past and the degree of association with the possibility of fraud in three or more stages. , It has a discrimination step of discriminating the possibility of fraud by giving priority to the one with a higher degree of association.
In the above information acquisition step, the time-series change tendency of the outsourcing cost is acquired, and this is applied to the change tendency pattern categorized in advance to be used as the outsourcing information.
In the above discrimination step, the above-mentioned reference outsourcing information in which the time-series change tendency of the past outsourcing cost is applied to the change tendency pattern categorized in advance and the degree of association between the possibility of fraud and the three or more levels are used. A fraudulent expense detection program characterized by having a computer run.
In the above information acquisition step, outsourcing information consisting of outsourcing costs and their subcontractors is acquired monthly.
The fraudulent expense detection program according to claim 1, wherein the determination step uses the above-mentioned degree of association with the outsourcing expenses acquired monthly in the past and the reference outsourcing information consisting of the subcontractors.
In the information acquisition step and the discrimination step, the change tendency pattern is based on the judgment model generated by machine learning as the input is the time-series change tendency of the outsourcing cost and the output is the categorized change tendency pattern. The fraudulent expense detection program according to claim 1 or 2, characterized in that it applies to.
In a fraudulent expense detection program that detects fraudulent expenses
An information acquisition step to acquire monthly outsourcing cost information consisting of outsourcing costs and breakdown information consisting of the breakdown of the outsourcing costs, and
Using the combination of the reference outsourcing cost information consisting of outsourcing costs acquired monthly in the past and the reference breakdown information consisting of the breakdown of the outsourcing costs, and the degree of association with the possibility of fraud at three or more levels, Based on the outsourcing cost information for reference according to the outsourcing cost information acquired in the above information acquisition step and the breakdown information for reference according to the breakdown information, the one with the higher degree of association is prioritized to determine the possibility of fraud. Has a discrimination step and
In the above information acquisition step, the time-series change tendency of the outsourcing cost is acquired, and this is applied to the change tendency pattern categorized in advance to obtain the outsourcing cost information.
In the above discrimination step, the above-mentioned reference outsourcing cost information in which the time-series change tendency of the past outsourcing cost is applied to the change tendency pattern categorized in advance and the degree of association between the possibility of fraud and three or more levels are used. A fraudulent expense detection program characterized by having a computer do this.
In a fraudulent expense detection program that detects fraudulent expenses
Information acquisition step to acquire travel traffic information consisting of travel expenses and transportation expenses on a monthly basis,
Reference travel according to the travel traffic information acquired in the above information acquisition step by using the reference travel traffic information consisting of travel expenses and transportation expenses acquired monthly in the past and the degree of association with the possibility of fraud in three or more stages. Based on the traffic information, it has a discrimination step to determine the possibility of fraud by giving priority to the one with the higher degree of association.
In the above information acquisition step, the time-series change tendency of the above travel expenses and transportation expenses is acquired, and this is applied to the change tendency pattern categorized in advance to obtain the above travel traffic information.
In the above discrimination step, the above-mentioned reference travel traffic information that applies the time-series change tendency of the past travel expenses and transportation expenses to the pre-categorized change tendency pattern and the degree of association of three or more levels of the possibility of fraud are used. A fraud detection program characterized by having a computer do what it does.
In the information acquisition step and the discrimination step, the change tendency is based on the judgment model generated by machine learning as the input is the time-series change tendency of the travel expense and the travel expense and the output is the typified change trend pattern. The fraudulent expense detection program according to claim 5, characterized in that it fits into a pattern.
In a fraudulent expense detection program that detects fraudulent expenses
An information acquisition step to acquire entertainment information consisting of entertainment expenses on a monthly basis,
Using the entertainment information for reference consisting of entertainment expenses acquired monthly in the past and the degree of association with the possibility of fraud at three or more levels, the entertainment information for reference according to the entertainment information acquired in the above information acquisition step Based on this, it has a discrimination step of determining the possibility of fraud by giving priority to the one having the higher degree of association.
In the above information acquisition step, the time-series change tendency of the entertainment entertainment expenses is acquired, and this is applied to the change tendency pattern categorized in advance to be used as the entertainment information.
In the above discrimination step, the above-mentioned reference entertainment information in which the time-series change tendency of past entertainment expenses is applied to a pre-categorized change tendency pattern and the degree of association with the possibility of fraud are used in three or more stages. A fraudulent expense detection program characterized by having a computer do this.
In the information acquisition step and the discrimination step, the change tendency is based on the judgment model generated by machine learning as the input is the time-series change tendency of the entertainment expense and the output is the typified change tendency pattern. The fraudulent expense detection program according to claim 7, characterized in that it fits into a pattern.
In a fraudulent expense detection program that detects fraudulent expenses
An information acquisition step to acquire attribute information related to the attributes of a company that detects fraudulent expenses and breakdown information consisting of a breakdown of outsourced expenses in the book data of that company.
Utilizing a combination of reference attribute information related to the attributes of a company acquired in the past, reference breakdown information consisting of a breakdown of outsourcing costs in the book data of that company, and a degree of association with three or more levels of fraud possibility. , Based on the reference attribute information according to the attribute information acquired in the above information acquisition step and the reference breakdown information according to the breakdown information, the one with the higher degree of association is prioritized to determine the possibility of fraud. A fraud detection program characterized by having a computer perform a determination step.
In the information acquisition step, the database that stores the correspondence between each subcontractor and its type is referred to, and based on the subcontractor as the breakdown of the above subcontracting expenses, the type is acquired and this is included in the breakdown information.
The fraudulent expense detection according to claim 1, characterized in that, in the above determination step, the type is acquired and used as the reference breakdown information based on the subcontractor as the breakdown of the outsourcing expenses in the book data of the company. program.
The fraudulent expense detection program according to any one of claims 1 to 10, wherein in the determination step, the degree of association corresponding to the weighting coefficient of each output of the node of the neural network in artificial intelligence is used.