JP2008077403A - Evaluation device, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To determine a group of companies that helps to support decision making and to easily select target companies through scoring. <P>SOLUTION: A data conversion part 102 converts information stored in a company-to-company relationship DB 201 indicative of the transaction relationships between a plurality of companies into network data wherein the companies serve as nodes while the relationships between the companies serve as edges. In this case, for example, the edges may be weighted as the credit scores of the transaction relationships are taken into consideration. A community extracting part 103 extracts communities by applying eigenvalue expansion to the network data, clustering the companies, calculating the level of belonging of each company to its respective cluster, and comparing a threshold value τ to the levels of assignment. An output part 104 outputs company information and scores (levels of belonging) included in each community. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an evaluation apparatus, an evaluation method, and an evaluation program.

  Business relationships are used as an important source of information about management decisions. For example, when determining an investment destination or a loan amount limit, risk due to the business relationship or partnership relationship of the target company is considered.

2. Description of the Related Art Conventionally, a technique for grasping a business relationship between companies by searching a database such as a credit check result or a securities report from an analysis target company has been proposed (for example, see Patent Document 1).
JP 2003-30467 A

  However, in the above-described conventional technology, since the database is searched to grasp the relationship between companies, it is possible to know a company directly related to the company. However, there was a problem that the relationship between companies could not be grasped. In addition, when selecting target companies based on the relationship between companies, it is considered effective to give priority by scoring. However, in the conventional technology, the overall relationship is considered. There is a problem that a score cannot be given.

  The present invention has been made in consideration of such circumstances, and the purpose of the present invention is to comprehensively consider inter-company relations so that a group of companies useful for decision support can be grasped and scoring is performed. It is to provide an evaluation apparatus, an evaluation method, and an evaluation program that can easily select a target company.

  In order to solve the above-described problems, the present invention converts each of a plurality of data into a network data having a node and a degree of relevance between each data as an edge, and converting means for inputting the converted network data, and an input A degree of aggregation between the data is calculated from the network data, a classification means for classifying the plurality of data into a plurality of clusters based on the calculated degree of aggregation, and each of the plurality of clusters classified by the classification means Based on the degree-of-affiliation calculating means for calculating the degree of belonging to the corresponding cluster and the degree of belonging to the corresponding cluster for each data, a community composed of a plurality of data having a high degree of relevance for each cluster is extracted. And an extracting means.

  According to the present invention, in the above invention, the extraction unit is configured to extract the network data based on the degree of belonging to the corresponding cluster for each data until the number of communities to be extracted reaches a preset number of communities to be extracted. Is updated and input to the classification means.

  The present invention is characterized in that, in the above invention, the conversion means weights the degree of association between the plurality of data according to an evaluation purpose.

  In order to solve the above-mentioned problem, the present invention converts each of a plurality of data into a network data having each node as a node and the degree of association between each data as an edge, and converting the network data to be input. , Calculating the degree of aggregation of the data from the input network data, and classifying the plurality of data into a plurality of clusters based on the calculated degree of aggregation, and each of the classified clusters Extraction to extract a community composed of a plurality of data having a high degree of relevance for each cluster based on the degree of belonging calculation for calculating the degree of belonging to the corresponding cluster for each data and the degree of belonging to the corresponding cluster for each data And a step.

  Further, in order to solve the above-described problem, the present invention converts a plurality of pieces of data into network data having each node as a node and the degree of association between each data as an edge, and inputs the converted network data. A conversion step; a degree of aggregation between data is calculated from the input network data; and a step of classifying the plurality of data into a plurality of clusters based on the calculated degree of aggregation; and the plurality of classified clusters For each of the data, a membership degree calculating step for calculating the degree of belonging to the corresponding cluster, and a community composed of a plurality of data having a high degree of relevance for each cluster based on the degree of belonging to the corresponding cluster for each data. The extraction step of extracting is executed.

  According to this invention, each of a plurality of data is converted into network data with each node as a node and the degree of association between each data as an edge, the converted network data is input, and the degree of aggregation of the data from the input network data And classifying a plurality of data into a plurality of clusters based on the calculated degree of aggregation, calculating the degree of belonging to the corresponding cluster for each data of the classified plurality of clusters, Based on the degree of belonging to the cluster, a community composed of a plurality of data having a high degree of association for each cluster is extracted. Therefore, since the community according to the purpose which comprehensively considered the relationship between companies is obtained as an evaluation result, the advantage that the company group useful for decision support can be easily grasped | ascertained is acquired.

  Further, according to the present invention, until the number of communities to be extracted reaches a preset number of communities to be extracted, the network data is updated based on the degree of belonging to the corresponding cluster for each data, and the classification means is updated. input. Therefore, it is possible to obtain an advantage that the classification of the cluster becomes more accurate and the understanding of the analysis result can be promoted.

  According to the present invention, the degree of association between a plurality of data is weighted according to the evaluation purpose. Therefore, since the degree of belonging to the community is obtained as a score (attribution degree), this score (attribution degree) can be used as the priority of the survey object selection criteria, and the target company can be easily selected. Benefits are gained.

Hereinafter, an evaluation apparatus according to an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of an evaluation apparatus 1 according to an embodiment of the present invention. In FIG. 1, an input unit 101 inputs, as parameters, the number of communities C to be extracted and a threshold value τ for determining companies in the communities to be extracted. The data conversion unit 102 converts information indicating business relationships between a plurality of companies stored in an inter-company relationship information DB (Data Base) 201 into network data, and outputs the data as network data 202.

  The community extraction unit 103 uses the network data 202 to classify into a plurality of clusters according to the degree of company aggregation, and extracts a community composed of a plurality of companies according to the degree of attribution of each company to the cluster. At this time, a score (degree of belonging) representing the degree of belonging to the community for each company is output to the output unit 104 and the degree of aggregation is stored in the degree of aggregation DB 203. The output unit 104 outputs company information and score (degree of belonging) included in each community.

Next, the operation of this embodiment will be described.
FIG. 2 is a flowchart for explaining the operation of the data conversion unit 102. The data conversion unit 102 inputs information indicating a business relationship between a plurality of companies stored in the inter-company relationship information DB 201 (step S1), and stores the information in the network data 202 with the company as a node and the inter-company relationship as an edge. Conversion is performed (step S2).

For example, when the business relationship is a business relationship, an adjacency matrix E having the relationship between the business i and the business j as an element e _ij when i and j are business ids (identification information) and the business i is a node. Can be configured. The adjacency matrix E is configured with e _ij = 1 (with a transaction) and e _ij = 0 (without a transaction), depending on whether there is a purchase / sale transaction relationship. This adjacency matrix E becomes the network data 202.

Moreover, you may make it give a weight to an edge according to the analysis objective. For example, when a credit score (score: credit score is also referred to as a score) is considered in the business relationship, i and j are company ids (identification information), score is a credit score of company i, and dim _i is an order of company i ( (the number of edges of i), the edge weight e _ij between the company i and the company j is determined as in the following equation (1), and the network data 202 is configured. dim is the number of branches of the node.

Here, the above-described aggregation degree, attribution degree, and cluster will be described. The degree of aggregation is the maximum eigenvalue, which is used for cluster classification. In adjacency matrix E described above, the number of companies subject to extraction community company is n, when the degree of belonging to a cluster of nodes i and the x _i, evaluating the degree of aggregation of the cluster by the following formula (2).

Here, T on the right shoulder of x represents transposition of x. E = e _ij (e _ii = 0) is a weighted adjacency matrix, and x = [x _i,. . . . . , X _n ]. However, | x | ² = 1. Here, the cluster having the highest degree of aggregation is determined by the eigenvector of the maximum eigenvalue obtained by solving the eigenvalue problem of the adjacency matrix E shown in the following equation (3), that is, by performing eigenvalue expansion. Details of eigenvalue expansion will be described later.

The element of the eigenvector x is given as a non-negative value because E is a non-negative symmetric matrix. The degree of aggregation is the maximum eigenvalue. Here, among the elements of x, the element x _{i *} of i _* having the maximum value is set as a normalization constant, and the maximum value is converted to 1 by the following equation (4) and defined as the degree of attribution.

  Next, FIG. 3 is a flowchart for explaining the operation of the community extraction unit 103. The community extraction unit 103 inputs the parameters (the number of communities C, the threshold τ) input from the input unit 101 and the network data 202 (n × n matrix) (step S10). Next, eigenvalue expansion is applied to the network data 202 (step S11). In eigenvalue expansion, n eigenvalues and n × n eigenvectors are obtained. Among these, only the maximum eigenvalue, that is, the eigenvector x with respect to the cohesion degree is used for the subsequent processing. The eigenvector x is composed of a matrix (vector) of one column for the maximum eigenvalue. That is, by applying eigenvalue expansion, a cluster composed of a plurality of companies (nodes) is configured. The maximum eigenvalue (aggregation degree) is recorded in the aggregation degree DB 203. Details of the eigenvalue expansion will be described later.

Next, by dividing the eigenvector x by the maximum value x ^* of the elements of x, the degree of belonging to the cluster of each company is calculated (step S12), the threshold τ and the degree of belonging are compared, A company is determined (step S13). Next, the input data is converted (step S14). That is, the element e _ij of the network data 202 (E) is updated by the following formula (5) using the degree of attribution m. Thereby, network data 202 (E) from which the degree of attribution obtained up to the t-th is removed is obtained.

  Here, t corresponds to the number of repetitions. Next, it is determined whether or not only the community number C to be extracted has been extracted (step S15). If the community number C has not been reached, the process returns to step S11 and the above-described processing is repeated. When the community number C is reached, the process is terminated.

  In this way, steps S11 to S14 are repeated, and when calculating the next cluster, as a method of excluding the influence of the already extracted cluster, the t + 1-th degree is obtained by removing the degree of membership obtained from the adjacency matrix up to the t-th. By obtaining the second cluster, it becomes possible to classify the company into a plurality of clusters. Here, the number of clusters included in the network is determined by a change in the degree of aggregation.

  Here, FIG. 5A is a conceptual diagram illustrating an example of the inter-company relationship information DB 201. In the illustrated example, company A has transactions with companies B, C, E, and F, company B has transactions with companies A, C, and G, and company C has transactions with companies A and B. ..., company K indicates that there are transactions with companies E, H, I, and J. The network data 202 converted from the inter-company relationship information DB 201 is, for example, as shown in FIG. Each node (circle) corresponds to each company AG, and transactions between companies are indicated by edges. Further, the network data 202 is represented by a matrix shown in FIG. 5C, and 1 and 0 are assigned to the elements of the matrix according to the presence or absence of transactions on the network.

  When the network data 202 is shown in FIG. 5C, it is classified into two clusters 1 and 2 as shown in FIG. Note that, as shown in FIG. 6, the degree of attribution indicating how much the companies belonging to the respective clusters 1 and 2 belong to the clusters 1 and 2 is calculated. Then, according to the threshold τ (0.3 in the illustrated example), the communities 1 and 2 are extracted from the clusters 1 and 2 as shown in FIG.

Next, FIG. 4 is a flowchart showing the procedure of the Jacobian method, which is a typical algorithm for eigenvalue expansion described above. In the Jacobian method, first, the data (processing target matrix) is set to E (n × n) = e _ij , the convergence determination parameter eps is set to a value close to 0 (such as 0.001), and X (N × n) = x _ij is prepared (step S20). Next, the element a _IJ having the maximum value among the non-diagonal elements (i ≠ j) of A and the position information I and J of the element are searched (step S21).

Next, it is determined whether or not it has converged (e _IJ <eps) (step S22). If it has not converged, θ is calculated (step S23), and a _Ij and a _Jj (j = 1, 2) ,..., N) are updated (step S24), e _iI and e _iJ (i = 1, 2,..., N) are updated (step S25), and x _iI and x _iJ (i = 1, 2) are updated. ,..., N) are updated (step S26), the process returns to step S21, and is repeated until convergence. When converged, V is output as an eigenvector, and the diagonal portion of A is output as an eigenvalue (step S27), and the process ends.

  Next, the result of community extraction using actual data by the above-described evaluation method according to the present embodiment will be described. It is examined whether or not a community corresponding to the configured relationship is extracted by using the purchase and sales transaction information of the company. Here, a case where a business relationship is used and a case where a credit score is reflected in the business relationship will be described. In the following, (1) Nikkei NEEDS: purchase and sales transaction information of 2903 listed companies in 2003, (2) Teikoku Databank COSMOS1: Credit score included in the summary file was used.

  First, when the business relationship is a business relationship, that is, when the credit score is not considered in the processing of FIG. 2, the result shown in FIG. 8 is obtained. In cluster 1, a general trading company community was extracted, and in cluster 2 an automobile industry community was extracted. It can be seen that these have a close relationship between sales destinations and suppliers and are appropriate as an extracted community. Thereby, it becomes possible to easily grasp to which community the partner company belongs.

  FIG. 9 shows a network for 336 companies having a high degree of cluster membership. In the figure, black circles, parallelograms, squares, triangles, and rhombus dots indicate nodes that are companies, respectively, and parallelograms, squares, triangles, and rhombus dots are the first, 2, 3 and 4 belong to the cluster. In the case of belonging to a plurality of clusters, it is indicated by the previously extracted cluster. From the figure, it can be confirmed that a dense part of the business relationship is extracted as a community on the network data of the business relationship.

  Next, in the process of FIG. 2, when a credit score (score) is considered in addition to the business relationship, the result shown in FIG. 10 is obtained. As shown by the bold line, it can be seen that the credit score is low and related companies are extracted as communities. For example, when the partner company is a high-ranking company in the community, it is possible to reduce the risk by performing a more detailed analysis.

  According to the above-described embodiment, as a result of the evaluation, the relationship between companies is comprehensively considered and a community according to the purpose is obtained. Therefore, it is possible to easily grasp a company group that is useful for decision support. Moreover, since the degree which belongs to a community is obtained as a score (attribution degree), this score (attribution degree) can be utilized as a priority of a survey object selection criterion. In addition, since the number of essential communities included in the data is determined based on the change in the degree of aggregation, it is possible to promote understanding of the analysis results.

  In the embodiment described above, company data is used as an example, but the present invention is not limited to this, and various data may be targeted.

  The conversion unit of the present invention corresponds to the data conversion unit 102, and the classification unit, the belonging degree calculation unit, and the extraction unit correspond to the community extraction unit 103.

It is a block diagram which shows the structure of the evaluation apparatus by embodiment of this invention. 4 is a flowchart for explaining the operation of a data conversion unit 102. 5 is a flowchart for explaining the operation of a community extraction unit 103. It is a flowchart which shows the procedure of the Jacobian method which is a typical algorithm of eigenvalue expansion. It is a conceptual diagram for demonstrating conversion to network data. It is a conceptual diagram for demonstrating a cluster and a degree of attribution. It is a conceptual diagram for demonstrating the cluster in network data. It is a figure which shows the result at the time of making business relationship into business relationship. It is a conceptual diagram which shows a network and a cluster at the time of making business relationship into a business relationship. It is a figure which shows a result at the time of reflecting a credit score in business relationship.

Explanation of symbols

101 Input unit 102 Data conversion unit 103 Community extraction unit 104 Output unit 201 Inter-company relationship information DB
202 Network data 203 Cohesion degree DB

Claims (5)

Conversion means for converting each of the plurality of data into a network data having a node and a degree of association between each data as an edge, and inputting the converted network data;
Classification means for calculating the degree of aggregation between the data from the input network data, and classifying the plurality of data into a plurality of clusters based on the calculated degree of aggregation;
For each data of the plurality of clusters classified by the classification means, an attribution degree calculating means for calculating an attribution degree for the corresponding cluster;
An evaluation apparatus comprising: an extraction unit that extracts a community composed of a plurality of data having a high degree of association for each cluster based on the degree of belonging to the corresponding cluster for each data. The extraction means includes
Until the number of communities to be extracted reaches a preset number of communities to be extracted, the network data is updated based on the degree of belonging to the corresponding cluster for each data and is input to the classification means. The evaluation apparatus according to claim 1.   The evaluation apparatus according to claim 1, wherein the conversion unit weights the degree of association between the plurality of data according to an evaluation purpose. A conversion step of converting each of a plurality of data into a network data having a node and a degree of association between each data as an edge, and inputting the converted network data;
A step of calculating the degree of aggregation of the data from the input network data, and classifying the plurality of data into a plurality of clusters based on the calculated degree of aggregation;
For each data of the plurality of classified clusters, an attribution calculation step for calculating the attribution to the corresponding cluster,
An extraction method comprising: extracting a community composed of a plurality of data having a high degree of association for each cluster based on the degree of belonging to the corresponding cluster for each data. On the computer,
A conversion step of converting each of a plurality of data into a network data having a node and a degree of association between each data as an edge, and inputting the converted network data;
A step of calculating the degree of aggregation of the data from the input network data, and classifying the plurality of data into a plurality of clusters based on the calculated degree of aggregation;
For each data of the plurality of classified clusters, an attribution calculation step for calculating the attribution to the corresponding cluster,
An extraction program for extracting a community composed of a plurality of data having a high degree of association for each cluster based on the degree of belonging to the corresponding cluster for each data.

Images