Program, information processing device, and data analysis method

The program and information processing device address the limitations of existing technologies by estimating precision matrices to reliably evaluate differences in variable group relationships across datasets, improving audit efficiency and reliability.

JP2026182219AActive Publication Date: 2026-11-19DELOITTE TOHMATSU LLC
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2025078258
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2025-05-08
Publication Date
2026-11-19
Estimated Expiration
2045-05-08

AI Technical Summary

Technical Problem

Existing technologies, such as those described in Patent Document 1, are inadequate for objectively evaluating the homogeneity of accounting data across different periods due to unsuitable comparative analysis methods and lack statistically reliable parameter estimation, particularly when analyzing relationships between variables in multiple datasets.

Method used

A program and information processing device that estimate precision matrices for two different datasets to minimize an information criterion, using a likelihood term and a penalty term based on sparsity, allowing for statistically reliable evaluation of differences in variable group relationships.

Benefits of technology

Enables statistically reliable estimation of differences in variable group relationships between two datasets, facilitating efficient identification of structural changes and enhancing the reliability of audit processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2026182219000001_ABST
    Figure 2026182219000001_ABST
Patent Text Reader

Abstract

We will perform statistically reliable estimations of the differences in the relationships between variable groups in two different datasets. [Solution] The reception means 111 receives operations from the user via the operation unit 14. The acquisition means 112 acquires journal entry data and stores it in the journal entry DB 121. The preprocessing means 113 reads the journal entry data from the journal entry DB 121, performs preprocessing, generates a matrix which is an accounting dataset, and stores it in the matrix DB 122. The estimation means 114 reads two different accounting datasets from the matrix DB 122 and estimates a first precision matrix corresponding to the first dataset and a second precision matrix corresponding to the second dataset based on the read first and second datasets, so as to minimize a predetermined information criterion.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical Field]

[0001] This invention relates to a method for analyzing the homogeneity of two datasets. [Background technology]

[0002] Audit work is burdensome for humans because it requires detecting fraudulent records and unusual transactions from a vast and complex amount of accounting information. Therefore, attempts are being made to utilize artificial intelligence (AI) in this audit work. In addition to reducing the burden on humans through automation, the use of AI in audit work is also expected to enhance detection capabilities, such as finding things that are difficult for humans to find.

[0003] Patent Document 1 describes an accounting information processing device that obtains a trained model by performing a graphical lasso regression analysis while discretely changing the L1 regularization term using training data, and then uses this trained model to calculate the anomaly score of records included in an unknown dataset and detect records with a high probability of fraud. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Patent No. 6667865 [Overview of the Initiative] [Problems that the invention aims to solve]

[0005] However, the technology described in Patent Document 1 above analyzes the relationships between variables in a single dataset. On the other hand, even with accounting information from the same business entity, changes in the business environment or accounting policies may occur if the period being analyzed is different. Objectively evaluating the homogeneity of data between such periods, or efficiently identifying the points where structural changes have occurred, can be important in audits. However, the technology in Patent Document 1 is not suitable for comparative analysis of periods using two such datasets.

[0006] Furthermore, it has been pointed out that the parameter estimation process for multivariate normal distributions that introduces the L1 regularization term into the objective function, as employed by the method described in Patent Document 1, does not have theoretically guaranteed statistically good properties, and therefore there may be issues with the statistical reliability of the estimation results obtained.

[0007] This invention has been made in view of the above circumstances, and one of its objectives is to provide a program, an information processing device, and a data analysis method that perform statistically reliable estimations of differences in the relationships between groups of variables in two different datasets. [Means for solving the problem]

[0008] In one aspect, the present invention provides a program that causes a computer to execute an acquisition step of acquiring a first data set and a second data set different from the first data set; based on the first data set and the second data set acquired in the acquisition step, a likelihood term based on a first likelihood for the first data set and a second likelihood for the second data set, and sparsity indicated by the number of at least one non-zero element in a first precision matrix in the first data set, a second precision matrix in the second data set, and a difference between the first precision matrix and the second precision matrix an estimation step of estimating the first precision matrix and the second precision matrix so as to minimize an information criterion defined based on a penalty term for evaluating; and an output step of outputting information indicating a difference in relationship of variable groups between the first data set and the second data set based on the first precision matrix and the second precision matrix estimated in the estimation step.

Effects of the Invention

[0009] The present invention can perform estimation with statistical reliability regarding the difference in the relationship of variable groups between two different data sets.

Brief Description of the Drawings

[0010] [Figure 1] A diagram showing an example of the configuration of the information processing apparatus 1. [Figure 2] A diagram showing an example of the journal entry DB 121. [Figure 3] A diagram showing an example of the configuration of the matrix DB 122. [Figure 4] A diagram showing an example of the functional configuration of the information processing apparatus 1. [Figure 5] A flowchart showing an example of the operation flow of preprocessing. [Figure 6] A flowchart showing an example of the operation flow of outputting difference information. [Figure 7] A flowchart showing an example of the operation flow of estimation processing. [Figure 8] An example of output in which the difference in the relationship of variable groups is represented by a graph structure. [Modes for carrying out the invention]

[0011] <Embodiment> <Configuration of the information processing device> Figure 1 shows an example of the configuration of an information processing device 1. The information processing device 1 shown in Figure 1 includes a processor 11, memory 12, communication unit 13, operation unit 14, and display unit 15. These components are connected to each other in a way that allows them to communicate with one another, for example, by a bus.

[0012] The processor 11 controls each part of the information processing device 1 by reading and executing programs stored in the memory 12. The processor 11 is, for example, a CPU (Central Processing Unit).

[0013] The control unit 14 is equipped with various control elements such as control buttons, a keyboard, a touch panel, and a mouse for issuing various instructions. It receives operations and sends signals corresponding to the operations to the processor 11. These operations include, for example, pressing buttons or making gestures on the touch panel.

[0014] The display unit 15 has a display screen such as a liquid crystal display and displays images under the control of the processor 11. A transparent touch panel of the operation unit 14 may be placed on top of the display screen.

[0015] The communication unit 13 is a communication circuit that connects the information processing device 1 to an external device or the like via wired or wireless means, enabling communication between them.

[0016] Memory 12 is a storage means for storing the operating system, various programs, data, etc., that are loaded into the processor 11. Memory 12 has RAM (Random Access Memory) and ROM (Read Only Memory). Memory 12 may also have a solid-state drive, a hard disk drive, etc. Memory 12 also stores the journal entry DB 121 and the matrix DB 122.

[0017] <Configuration of the journal entry database> Figure 2 shows an example of the journal entry database 121. The journal entry database 121 is a database that stores journal entry information for multiple accounting transactions. The journal entry database 121 shown in Figure 2 has a journal entry ID list 1211 and a journal entry table 1212.

[0018] The journal entry ID list 1211 is a list that stores a unique journal entry ID for each journal entry, along with information about that journal entry. Each journal entry ID listed in the journal entry ID list 1211 is associated with a single journal entry sheet 1212.

[0019] Journal entry table 1212 is a table or record that stores specific journal entry data corresponding to the journal entry ID. The journal entry data includes essential items such as debit account title, debit amount, credit account title, and credit amount. Furthermore, journal entry table 1212 may also include items that indicate the entity involved in the transaction, such as department code, business unit code, and person in charge code; information indicating the counterparty to the transaction, such as customer code and supplier code; and additional information such as a description. Journal entry table 1212 may also include items or data that distinguish whether accounts such as "accounts receivable" and "accounts payable" arose from normal transactions or from the buying and selling of fixed assets and securities. The memory area associated with these items stores codes indicating accounts, strings indicating names, and numerical values ​​indicating amounts. Note that while journal entry DB 121 stores journal entry data, if the information processing device 1 is to analyze accounting data, it may have a database that stores accounting data instead of journal entry DB 121. Here, accounting data refers to all data that records the economic activities of a company or organization, specifically data that forms the basis for preparing financial statements, such as sales, purchases, expenses, assets, liabilities, and equity.

[0020] <Configuration of the matrix database> Figure 3 shows an example of the configuration of the matrix DB 122. The matrix DB 122 is a database that stores matrix-format data (also referred to as the "accounting dataset") generated by the pre-processing means 113, which will be described later, based on the journal entry data stored in the journal entry DB 121. In this matrix-format data, each column corresponds to a variable to be analyzed, such as an account title, and each row corresponds to an individual sample, such as an observed value or aggregated value for a specific period or transaction unit.

[0021] The matrix DB122 shown in Figure 3 has a matrix ID list 1221 and a matrix element value group 1222. The matrix ID list 1221 is a list that manages matrix IDs for uniquely identifying the matrix representing each accounting dataset. Matrix IDs are, for example, "X1", "X2", etc. Each matrix ID listed in the matrix ID list 1221 is associated with one matrix element value group 1222.

[0022] The matrix element value group 1222 is an area that stores the values ​​of each element (component) of the matrix data corresponding to the matrix ID, associating them with its row index and column index. For example, the matrix element value group 1222 shown in Figure 3 stores the values ​​of each component of the accounting dataset (size n1 × d) corresponding to the matrix ID "X1".

[0023] Note that the matrix DB122 may be allocated as a specific area within memory 12, or it may be implemented as a table on a relational database management system (RDBMS).

[0024] Furthermore, the matrix DB122 may be stored as a file in memory 12 or an external storage device in a specific file format, such as CSV format or Parquet format.

[0025] Furthermore, the matrix DB122 may store multiple accounting datasets as separate tables or files, or it may store them together in a single table or the like, along with information indicating the period and dataset classification.

[0026] <Functional Configuration of Information Processing Devices> Figure 4 shows an example of the functional configuration of the information processing device 1. The processor 11 of the information processing device 1 shown in Figure 4 functions as a receiving means 111, an acquisition means 112, a preprocessing means 113, an estimation means 114, and an output means 115 by executing a program stored in the memory 12.

[0027] The reception means 111 receives operations from the user via the operation unit 14. These operations include, for example, an operation to instruct the input of journal entry data, an operation to instruct the execution of preprocessing to generate matrices, an operation to select matrices representing two different accounting datasets and instruct the system to estimate the differences in the relationships between the groups of variables between them, and an operation to instruct the system to output the estimation results.

[0028] For example, when the receiving means 111 receives an operation from a user via the operation unit 14 to instruct the input of journal entry data, the acquisition means 112 acquires the input journal entry data and stores it in the journal entry DB 121. Also, for example, when the receiving means 111 receives an operation to specify a destination for acquiring journal entry data, the acquisition means 112 acquires the journal entry data from the specified destination via the communication unit 13 and stores it in the journal entry DB 121.

[0029] For example, when the reception means 111 receives an operation from a user via the operation unit 14 to instruct the execution of preprocessing to generate a matrix, the preprocessing means 113 reads journal entry data from the journal entry DB 121, executes the instructed preprocessing, and generates a matrix which is an accounting dataset. The preprocessing means 113 then stores the generated matrix in the matrix DB 122.

[0030] For example, if the reception means 111 receives instructions from the user via the operation unit 14 to select matrices representing two different accounting datasets (a first dataset and a second dataset) and to estimate the difference in their relationship, the estimation means 114 reads the corresponding two accounting datasets from the matrix DB 122.

[0031] The estimation means 114 then estimates a first precision matrix corresponding to the first dataset and a second precision matrix corresponding to the second dataset, based on the read first and second datasets, in order to minimize a predetermined information criterion. Details of this estimation process will be described later. The first and second datasets mentioned above are matrices generated by performing preprocessing on journal entry data read from the journal entry DB 121. In other words, these two datasets, whose precision matrices are estimated by the estimation means 114, are examples of the first and second datasets generated based on journal entry data.

[0032] The output means 115, for example, when the receiving means 111 receives an operation from the user to instruct the output of estimation results, generates information showing the difference in the relationships between the variable groups between the first dataset and the second dataset, based on the first and second precision matrices estimated by the estimation means 114. This information may include, for example, a list of variable pairs that have changed over different periods, or a graph structure showing the structural changes (differences) in the relationships. The output means 115 then outputs this generated information, for example, by displaying it on the display unit 15, or by transmitting it to an external device via the communication unit 13.

[0033] Therefore, the information processing device 1 shown in Figure 4 is an example of an information processing device having: an acquisition means for acquiring a first dataset and a second dataset different from the first dataset; an estimation means for estimating a first precision matrix and a second precision matrix so as to minimize an information criterion defined based on the first dataset and the second dataset acquired by the acquisition means, based on a likelihood term based on the first likelihood for the first dataset and the second likelihood for the second dataset, and a penalty term that evaluates sparsity indicated by the number of non-zero elements in the difference between the first precision matrix in the first dataset, the second precision matrix in the second dataset, and the difference between the first precision matrix and the second precision matrix; and an output means for outputting information indicating the difference in the relationship between the variable groups between the first dataset and the second dataset based on the first precision matrix and the second precision matrix estimated by the estimation means.

[0034] <Estimation Processing Algorithm> The algorithm for the estimation process performed by the estimation means 114 in this embodiment will be described below.

[0035] Equation (1) shows that the accounting dataset follows a multivariate normal distribution determined by the precision matrix Λ. In this example, the accounting dataset is precentered by preprocessing. That is, the accounting dataset is converted into deviations from the sample mean for each sample by preprocessing. Since the accounting dataset is assumed to follow a multivariate normal distribution in which the population mean vector is approximated by the sample mean vector, the population mean vector in equation (1) becomes the zero vector.

[0036]

number

[0037] In this equation, x is a sample vector corresponding to one row of the dataset X, and if X has d variables, then the number of dimensions is d.

[0038] Equation (2) is the objective function E based on the information criterion. IC This demonstrates that information criteria are standards for evaluating the quality and validity of statistical models, and are defined based on, for example, AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). In this embodiment, the information criterion is defined based on either AIC or BIC. In other words, this information criterion is an example of an information criterion defined based on either the Akaike Information Criterion or the Bayesian Information Criterion. Note that the information criterion is not limited to the AIC and BIC mentioned above, but may also be other information criteria such as DIC (Deviation Information Criterion).

[0039] Objective function E IC Λ is the first precision matrix corresponding to the first dataset X1. (1) And Λ, which is the second precision matrix corresponding to the second dataset X2. (2)and are variables. L is a likelihood term, and R is a penalty term. The likelihood term is an index representing how well the model fits the data. The penalty term is a term that imposes a penalty on the complexity of the model, such as the number of parameters and degrees of freedom.

[0040]

Mathematical Formula

[0041] Equation (3) represents the likelihood term. The likelihood term is a term obtained by multiplying the sum of the respective log-likelihood functions of the first dataset X1 and the second dataset X2 by a negative coefficient (here, -1). Note that in Equation (3), p(X k |Λ (k) ) represents the first likelihood for the first dataset when k=1, and represents the second likelihood for the second dataset when k=2. Therefore, this likelihood term is an example of a likelihood term based on the first likelihood for the first dataset and the second likelihood for the second dataset.

[0042]

Mathematical Formula

[0043] Equation (4) and Equation (5) described later show examples of specific forms of the penalty term. The penalty term is the first precision matrix Λ (1) sparsity, the second precision matrix Λ (2) sparsity, and is configured as a term that evaluates at least one of the sparsity related to the difference between them. Here, the sparsity is evaluated based on the number of non-zero elements (L0 norm). That is, this penalty term is an example of a penalty term that evaluates sparsity represented by the number of non-zero elements of at least one of the first precision matrix in the first dataset, the second precision matrix in the second dataset, and the difference between the first precision matrix and the second precision matrix.

[0044] For example, R which is an example of the penalty term shown in Equation (4) A has the first term as Λ (1)The L0 norm of , the second term is Λ (1) and Λ (2) The difference between, and Λ (2) This is the L0 norm of the Hadamard product between two matrices. The Hadamard product is a matrix product determined by multiplying matrices of the same size element by element. An example of a penalty term of this form is a penalty term that includes a first penalty term based on the number of non-zero elements of the first precision matrix, and a second penalty term based on the number of non-zero elements of the Hadamard product of the difference between the first precision matrix and the second precision matrix and the second precision matrix.

[0045] Note that here, λ1 and λ in equation (4) d2 , and λ1, λ2, λ in equation (5) described later d is a positive constant that adjusts the weight of each penalty element. These are determined, for example, depending on whether AIC or BIC is adopted as the information criterion, and the overall size of the penalty term is adjusted. For example, if BIC is adopted as the information criterion, n is the sample size of the k-th dataset. k The logarithm of, i.e., log n k etc. λ d2 It will be set to this.

[0046]

number

[0047] Furthermore, another form of the penalty term is R shown in equation (5) below. B As shown above, it can also be constructed as a weighted sum of three terms: the L0 norm of the first precision matrix, the L0 norm of the second precision matrix, and the L0 norm of the difference matrix. An example of a penalty term in this form is a penalty term that includes a first penalty term based on the number of non-zero elements in the first precision matrix, a second penalty term based on the number of non-zero elements in the second precision matrix, and a third penalty term based on the number of non-zero elements in the difference between the first precision matrix and the second precision matrix.

[0048]

number

[0049] Equation (6) shows the log-likelihood function used to derive the likelihood term described above. This log-likelihood function is the log-likelihood function of a multivariate normal distribution. The log-likelihood function of the k-th dataset is its precision matrix Λ (k) From the logarithmic matrix of S, (k) Λ (k) It is proportional to the value obtained by subtracting the trace. Here S (k) This is the sample covariance matrix for the k-th dataset.

[0050]

number

[0051] Equation (7) shows an approximation of the number of non-zero elements. The penalty term described above evaluates the sparsity of the precision matrix based on its number of non-zero elements. Therefore, the problem of minimizing the information criterion including this penalty term becomes a combinatorial optimization problem, which presents difficulties in terms of computation time and numerical handling. To mitigate this combinatorial optimization problem into a continuous optimization problem, the information processing device 1 introduces a proxy function that substitutes for the penalty term in its estimation algorithm.

[0052]

number

[0053] The ||z||0 shown in equation (7) is called the quasi-norm or L0 norm and represents the number of non-zero elements in z. If z is a matrix or vector, ||z||0 is the total number of non-zero elements that make up z. Also, if z is a scalar, ||z||0 is 0 for z=0 and 1 for z≠0. For simplicity, equation (7) is explained by assuming z is a scalar. In this case, the L0 norm of z on the left side is approximated by the right side. The right side consists of the absolute value of z and a small positive constant η (η>0). This approximation is justified by the fact that the right side is 0 when z=0, approaches 1 rapidly as the absolute value of z increases, and is convex downwards with respect to the absolute value of z. The fact that the right side is convex downwards is shown by the fact that the second derivative of the function g(|z|) that represents the right side is negative over the entire domain.

[0054] Equation (8) shows an inequality that gives an upper bound on the approximation of the number of non-zero elements. The fact that g(|z|) is convex downwards as described above means that the tangent line at any point is greater than or equal to this function over the entire domain. In this equation, the left side is the above function and the right side is the tangent line to that function at any point |z0|, and it is shown that the right side is always greater than or equal to the left side.

[0055]

number

[0056] Equation (9) is the inequality obtained by rearranging equation (8). Here, the right-hand side showing the tangent line to equation (8) is equal to the substitute function q(z|z0) in equation (9). Therefore, ||z||0 can be bounded from above by the substitute function q(z|z0) determined by the variable z when z0 is the estimated value.

[0057]

number

[0058] Equation (10) is an extension of the inequality in equation (9) to the precision matrix Λ. In equation (10), |Λ uv| is the absolute value of the element at row u and column v in the precision matrix Λ, η is a minute positive constant, and Λ (t) uv is the estimated value of Λ in each of the iterative processes uv . That is, when the right-hand side shown in Equation (10) is used for the penalty term described above, the information processing apparatus 1, in the iterative process, for Λ uv by Λ (t) uv updating, can bring the precision matrix Λ closer to the optimal value. Here, the right-hand side shown in Equation (10) is an example of a surrogate function that is constructed based on the estimated value of the precision matrix in the iteration and takes a value equal to or greater than the penalty term.

[0059] By thus replacing the penalty term of the information criterion adopted in the objective function with a surrogate function, the information processing apparatus 1 solves a continuous optimization problem instead of a combinatorial optimization problem. That is, this process of estimating the precision matrix performs a process of minimizing a surrogate objective function obtained by replacing the penalty term with a surrogate function that takes a value equal to or greater than the penalty term, so that execution is easier than in the case where no replacement is performed.

[0060] [Mathematical Formula]

[0061] Note that Λ is a symmetric matrix, and the diagonal components are positive. Therefore, in order to evaluate the sparsity of this precision matrix Λ, it is sufficient to sum only the number of non-zero elements in the strictly upper triangular components obtained by excluding the diagonal components from the upper triangular components. For this reason, Equation (10) limits the components to be counted to the strictly upper triangular components (1≦u<v≦d). Of course, since Λ is a symmetric matrix, the components to be counted may be limited to the strictly lower triangular components (1≦v<u≦d) obtained by excluding the diagonal components from the lower triangular components.

[0062] The precision matrix estimated in this way indicates that when its component is 0, the corresponding two variables are statistically independent. For example, of the precision matrix estimated for the dataset X, Λ uvWhen x is 0 (u≠v), the two variables x of dataset X u and x v They are statistically independent.

[0063] <Operation of the Information Processing Device> <How to retrieve journal entry data> Figure 5 is a flowchart illustrating an example of the operation flow for acquiring journal entry data. The processor 11 of the information processing device 1 determines whether or not it has received an instruction from the user to acquire journal entry data or to perform preprocessing (step S001). While it is determined that neither of these instructions has been received (step S001; NO), the processor 11 continues this determination process.

[0064] If it is determined that an instruction to acquire journal entry data has been received (step S001; YES), the processor 11 acquires the journal entry data in accordance with the instruction, either from a user operation from the operation unit 14 or from an external device via the communication unit 13 (step S002).

[0065] Furthermore, if the processor 11 determines that it has received an instruction to perform preprocessing in step S001, it reads journal entry data from the journal entry DB 121 in accordance with the instruction and generates matrices and rows, which are accounting datasets.

[0066] If the processor 11 receives instructions to perform preprocessing along with instructions to acquire journal entry data, it acquires the journal entry data and performs preprocessing based on this data to generate matrices that constitute accounting datasets (step S003). The processor 11 then stores the generated matrices in memory 12 (step S004).

[0067] <Operation of outputting difference information> Figure 6 is a flowchart showing an example of the operation flow for outputting difference information. Processor 11 determines whether or not it has received an instruction from the user to output difference information (step S101). While it is determined that it has not received an instruction to output difference information (step S101; NO), processor 11 continues this determination process.

[0068] When it determines that it has received an instruction to output difference information (step S101; YES), the processor 11 retrieves two accounting datasets (matrices) stored in memory 12 according to the instruction (step S102). These two accounting datasets are the first dataset corresponding to the first period and the second dataset corresponding to a second period different from the first period. Therefore, this process is an example of a retrieval step to obtain the first dataset and the second dataset which is different from the first dataset. Here, the first period and the second period can be, for example, last month and this month, the previous period and this period, etc. The period can be, for example, a week, a month, a quarter, a year, etc. These two periods can be continuous or discontinuous. For example, the two periods can be this quarter and the same quarter of the previous year, etc.

[0069] Then, processor 11 estimates the first precision matrix from the first dataset corresponding to the first period, and estimates the second precision matrix from the second dataset corresponding to a second period different from the first period (step S200). In the example shown in Figure 6, these two estimation processes are performed in parallel. Details of the estimation process will be described later.

[0070] Once both estimation processes are complete, the processor 11 compares the sparsity patterns of the respective estimated precision matrices and detects points where the relationship between the variable groups between the first and second datasets has changed, or changed variable pairs (step S103). If the comparison reveals that the sparsity patterns of the precision matrices corresponding to the two periods are almost identical, the processor 11 determines that the accounting datasets for these two periods are almost identical.

[0071] The processor 11 then outputs information indicating the change (difference) in the relationship between the variable groups between the two accounting datasets to the display unit 15 or the like (step S104). Therefore, this process is an example of an output step that outputs information indicating the difference in the relationship between the variable groups between the first dataset and the second dataset, based on the estimated first precision matrix and second precision matrix.

[0072] The processor 11 may output a list of variable pairs that show differences in the relationships between the groups of variables in the two accounting datasets. In this case, the above process, which outputs the differences in relationships, is an example of an output step that outputs a list of variable pairs that show differences in relationships.

[0073] The processor 11 may output the difference in the relationships between the two accounting datasets as a graph structure, as information indicating the difference in the relationships between the groups of variables. In this case, the above process of outputting the difference in relationships is an example of an output step that outputs the relationships between the groups of variables or the difference therebetween as a graph structure.

[0074] <How the estimation process works> Figure 7 is a flowchart showing an example of the operation flow of the estimation process. The processor 11 initializes the variable to be estimated, i.e., the precision matrix (step S201). The precision matrix is ​​initialized, for example, by substituting a d x d square identity matrix.

[0075] Next, processor 11 searches for the optimal value of the precision matrix using the coordinate descent method. That is, processor 11 selects one of the variables that are elements of these precision matrices (step S202), and while keeping the other variables fixed, it changes the selected variable and evaluates the surrogate objective function that replaces the objective function described above each time. Then, processor 11 updates the one variable that was selected so that the surrogate objective function is optimized (step S203).

[0076] If the surrogate objective function no longer changes, the processor 11 interrupts the updating of the single variable mentioned above and determines whether it has completed updating all variables (step S204). If it determines that all variables have not been updated (step S204; NO), the processor 11 returns to step S202 and selects the unupdated variables.

[0077] On the other hand, if it is determined that all variables have been updated (step S204; YES), the processor 11 determines whether the change in the surrogate objective function has fallen within a certain range compared to before the completion of one cycle of updating all variables, that is, whether the surrogate objective function has converged (step S205). If it is determined that it has not converged (step S205; NO), the processor 11 resets the variable selection history, for example, and returns to step 202. As a result, the processor 11 starts the next cycle of updating all variables.

[0078] On the other hand, if it is determined that convergence has occurred (step S205; YES), the processor 11 stores the estimation result at that point, i.e., the precision matrix, in memory 12 (step S206) and terminates the process.

[0079] Therefore, the operation of this estimation process is an example of an estimation step in which the first and second precision matrices are estimated to minimize an information criterion defined based on the likelihood term and the penalty term.

[0080] Furthermore, as described above, this estimation process iteratively minimizes a surrogate objective function, which is obtained by replacing a portion of the objective function with a surrogate function. In other words, this process is an example of an estimation step that performs estimation by iteratively executing a process to minimize a surrogate objective function, which is constructed based on the estimated values ​​of the first and second precision matrices in each iteration, and takes a value greater than or equal to the penalty term, and which is included in the information criterion.

[0081] Furthermore, the estimation process described in this embodiment employs coordinate descent as the optimization algorithm. Therefore, this process is an example of a process that minimizes a surrogate objective function using coordinate descent.

[0082] <Example of output using graph structure> Figure 8 shows an example of an output that represents the differences in the relationships between a group of variables using a graph structure. Points P0 to P9 in Figure 8 represent 10 variables in accounting information, corresponding to account titles such as "accounts receivable," "sales," "purchases," "accounts payable," "uncollected income," and "accounts payable." Note that the number of account titles is not limited to the example shown in Figure 8 and is generally greater, but for the sake of explanation, we will use the above 10 in Figure 8. The straight lines connecting these points are drawn based on the components of the estimated precision matrix.

[0083] The graph structure shown in Figure 8(a) illustrates, for example, the independence between variables in the first dataset observed during the first period. In this graph structure, the number of connections each point has to other points (hereinafter also referred to as the number of connections) is 6 for point P0, 3 for point P4, 5 for point P5, 4 for point P6, 2 for point P7, 4 for point P8, and 5 for point P9, while all other points (i.e., points P1, P2, and P3) have a connection number of 1. Two points connected by a line indicate a relationship between the two variables they represent. Therefore, points not connected by a line are conditionally independent.

[0084] On the other hand, the graph structure shown in Figure 8(b) illustrates, for example, the independence between variables in the second dataset observed in a second period that differs from the first period. In this graph structure, the number of connections to point P0 decreases by one, from 6 to 5. Also, the number of connections to point P4 increases by two, from 3 to 5. As a result, when moving from the first period to the second period, the lines connecting points P0 and P7, and points P5 and P8 disappear, while lines connecting points P4 and P6, and points P4 and P9 are created. In this way, the information processing device 1 represents the difference in the relationships between groups of variables in two different periods using points and lines. Therefore, a user viewing this can grasp the creation and disappearance of variable pairs. The information processing device 1 may also change the thickness and shape of the lines connecting the points depending on the strength of the relationships, for example, the magnitude of the absolute values ​​of the corresponding elements in the precision matrix.

[0085] Based on the operations described above, this information processing device 1 can make theoretically and practically valid estimations of the differences in the relationships between variable groups in accounting datasets corresponding to two different periods. In particular, because this information processing device 1 employs an information criterion in its objective function, it can obtain estimation results that are close to the true model when the sample size is large.

[0086] In particular, the information processing device 1 described above analyzes account titles as variables. This information processing device 1 can identify changes that need to be considered in order to evaluate the homogeneity of accounting datasets obtained from journal entry data such as journal vouchers, based on the relationship structure between account titles, in audit procedures that anticipate the risk of excessively inflating sales or profits. The user of the information processing device 1 can, for example, identify pairs of account titles extracted as changed variable groups as the cause of changes in journal vouchers and use this information for qualitative justification in audits.

[0087] <Variation> The above describes the embodiment, but the contents of this embodiment can be modified as follows. Furthermore, the following modifications may be combined with each other.

[0088] <1> In the embodiment described above, the information processing device 1 had a processor 11 composed of a CPU, but the control means for controlling the information processing device 1 may have other configurations.

[0089] In other words, the information processing device 1 may have various processors as its processor 11, such as a GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), or programmable logic device, in addition to a CPU.

[0090] <2> The processor operation in the above-described embodiment may not be performed by a single processor, but may also be performed by multiple processors located in physically separate locations working together. In this case, the information processing device 1 may be configured as a server-client system, or it may be implemented in a cloud computing environment.

[0091] Furthermore, the order of the processor's operations is not limited to the order described in the embodiments described above, but may be changed as appropriate.

[0092] <3> In the embodiments described above, each dataset was generated based on journal entry data, which is one of the accounting data subject to audit. However, it may also be generated based on other data. In other words, the information processing device 1 is applicable to analyses other than those related to audits. For example, the dataset may be generated based on biosensor data that monitors health status, sensor data that monitors abnormalities in the manufacturing lines of various products, purchase history data in e-commerce, website access log data, etc. In this case, the information processing device 1 outputs information that shows changes (differences) in health status or the product manufacturing environment over different periods.

[0093] Furthermore, although the two datasets were selected to correspond to different time periods, the differences between the datasets are not limited to time; they may also include differences in environment, conditions, location, etc. For example, the information processing device 1 may analyze two datasets generated from biosensor data of different individuals. Alternatively, for example, the information processing device 1 may analyze two datasets generated from monitoring data of different manufacturing lines.

[0094] <4> In the embodiment described above, the information processing device 1 used the coordinate descent method as the optimization algorithm for the estimation process, but other algorithms such as the proximity gradient method may also be used.

[0095] <5> The information processing using the information processing device 1 described above is conceived as a method for analyzing the differences in the relationships between the respective variable groups in two datasets. In other words, this method acquires a first dataset and a second dataset different from the first dataset, estimates a first precision matrix and a second precision matrix based on the acquired first and second datasets, estimates an information criterion defined based on the first likelihood for the first dataset and the second likelihood for the second dataset, and a penalty term that evaluates sparsity indicated by the number of non-zero elements in the difference between the first precision matrix and the second precision matrix in the second dataset, and outputs information showing the differences in the relationships between the variable groups in the first and second datasets based on the estimated first and second precision matrices.

[0096] <6> In the embodiment described above, the program read by the processor 11 of the information processing device 1 is an example of a program that causes the computer to execute an acquisition step of acquiring a first dataset and a second dataset different from the first dataset; an estimation step of estimating a first precision matrix and a second precision matrix so as to minimize an information criterion defined based on the first dataset and the second dataset acquired in the acquisition step, based on a likelihood term based on the first likelihood for the first dataset and the second likelihood for the second dataset, and a penalty term that evaluates sparsity indicated by the number of non-zero elements in the difference between the first precision matrix and the second precision matrix in the first dataset and the second precision matrix; and an output step of outputting information indicating the difference in the relationship between the variable groups between the first dataset and the second dataset based on the first precision matrix and the second precision matrix estimated in the estimation step.

[0097] This program may be provided stored on a computer-readable recording medium, such as magnetic tapes and magnetic disks, optical disks, magneto-optical recording media, or semiconductor memory. Alternatively, this program may be downloaded via a communication line, such as the Internet. [Explanation of Symbols]

[0098] 1... Information processing device, 11... Processor, 111... Reception means, 112... Acquisition means, 113... Preprocessing means, 114... Estimation means, 115... Output means, 12... Memory, 121... Journal entry DB, 1211... Journal entry ID list, 1212... Journal entry table, 122... Matrix DB, 1221... Matrix ID list, 1222... Matrix element value group, 13... Communication unit, 14... Operation unit, 15... Display unit.

Claims

1. On the computer, An acquisition step to acquire a first dataset and a second dataset different from the first dataset as the subject of audit, An estimation step in which the first and second precision matrices are estimated such that an information criterion is minimized based on the first and second precision matrices obtained in the acquisition step, the first precision matrix in the first dataset, the second precision matrix in the second dataset, and a penalty term that evaluates sparsity indicated by the number of non-zero elements in the difference between the first and second precision matrices, the first and second precision matrices obtained in the acquisition step, An output step that outputs information showing the difference in the relationship between the variable groups between the first dataset and the second dataset, based on the first and second precision matrices estimated in the estimation step, A program that executes the command.

2. The estimation step described above is: The estimation is performed by iteratively executing a process to minimize a surrogate objective function, which is obtained by replacing the penalty term included in the information criterion with a surrogate function constructed based on the estimated values ​​of the first and second precision matrices in each iteration, and which takes a value greater than or equal to the penalty term. The program according to feature 1.

3. The program according to claim 1, characterized in that the information criterion is defined based on the Akaike information criterion or the Bayesian information criterion.

4. The program according to claim 1, characterized in that the penalty term includes a first penalty term based on the number of non-zero elements of the first precision matrix, a second penalty term based on the number of non-zero elements of the second precision matrix, and a third penalty term based on the number of non-zero elements of the difference between the first precision matrix and the second precision matrix.

5. The program according to claim 1, characterized in that the penalty term includes a first penalty term based on the number of non-zero elements of the first precision matrix, and a second penalty term based on the number of non-zero elements of the Hadamard product of the difference between the first precision matrix and the second precision matrix and the second precision matrix.

6. The program according to claim 2, characterized in that the process of minimizing the surrogate objective function is performed using the coordinate descent method.

7. The aforementioned proxy function is |Λ uv | is the absolute value of the element in row u, column v of the precision matrix, η is an infinitesimal positive constant, and Λ (t) uv With the estimated value of the precision matrix in each iteration, the terms corresponding to each element of the penalty term are The program according to claim 2, characterized in that it is given in the form of:

8. The output step outputs a list of variable pairs that have differences in the relationship. The program according to feature 1.

9. The output step outputs the relationships or differences between the group of variables as a graph structure. The program according to feature 1.

10. The first and second datasets are generated based on journal entry data. The program according to feature 1.

11. Acquisition means for acquiring a first dataset and a second dataset different from the first dataset, Estimation means for estimating the first and second precision matrices so as to minimize an information criterion defined based on the first and second datasets acquired by the acquisition means, which is a likelihood term based on the first likelihood for the first dataset and the second likelihood for the second dataset, and a penalty term that evaluates sparsity indicated by the number of non-zero elements in the difference between the first and second precision matrices, which is the first precision matrix in the first dataset and the second precision matrix in the second dataset. An output means that outputs information showing the difference in the relationship between the variable groups between the first dataset and the second dataset, based on the first and second precision matrices estimated by the estimation means, An information processing device having

12. Obtain a first dataset and a second dataset that is different from the first dataset. Based on the acquired first and second datasets, the first and second precision matrices are estimated to minimize an information criterion defined by a likelihood term based on the first likelihood for the first dataset and the second likelihood for the second dataset, and a penalty term that evaluates sparsity indicated by the number of non-zero elements in the difference between the first and second precision matrices, the first precision matrix in the first dataset and the second precision matrix in the second dataset. Based on the estimated first and second precision matrices, information is output showing the difference in the relationships between the variable groups in the first and second datasets. Data analysis methods.

Citation Information

Patent Citations

  • Accounting information processing device, accounting information processing method, and accounting information processing program

    JP6667865B1