JP2008204190A - Classification evaluation device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a classification evaluation device for providing a user with an index in selecting target classification from classification obtained by automatic classification. <P>SOLUTION: This classification evaluation device for evaluating the result of the classification of a plurality of data on the basis of the characteristics of data is provided with: a similarity calculation part 104 for calculating the similarity of the characteristics of data with characteristics representing data belonging to the classification to which the data belong; and a classification evaluation part 106 for evaluating the classification on the basis of the similarity calculated about each data belonging to the classification. Thus, it is possible to automatically evaluate and display validity of classification in narrowing down the classification to be used for the retrieval of data from a plurality of classifications. Therefore, it is possible for a user to more quickly and easily find out his or her target classification. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a classification evaluation apparatus, and more particularly to a classification evaluation apparatus that evaluates a classification result obtained by classifying a data group including a plurality of data.

  Conventionally, various methods for automatically classifying data have been devised. In addition, a method for evaluating and presenting the usefulness of classification has been devised, which indicates whether a target object can be efficiently searched from automatically classified data.

  For example, Patent Document 1 discloses an automatic classification method that calculates word vectors of documents to be classified and automatically sets similar data to the same classification. In addition, a method is described in which one or more words / phrases with a large word vector value corresponding to the classification are displayed as keywords representing the classification. According to this method, the entire configuration of the document database can be grasped by the keyword. Furthermore, it is possible to easily narrow down the data by selecting a category suitable for the purpose by judging the keyword by looking at the keyword.

  However, the method of Patent Document 1 has the following two problems. One is that when the number of classifications is very large, it is difficult to visually confirm the keywords of each classification.

  The other is that when a document to be classified is not a paper or a patent bulletin, but a Web page, an e-mail, etc., there are many cases where there are few or no words in the document. In such a case, the keyword representing the classification may not necessarily be a word representing the content. In addition, when there are many documents that do not include a common keyword, a very common word that is often included in any document may be classified as a keyword.

JP-A-8-263514

  In this way, when using an automatically classified database, the difference in quality between appropriate and unsuitable classifications increases, but since both are displayed together, the target classification can be found efficiently. Becomes difficult.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a user with an index for selecting a target classification from classifications obtained by automatic classification. It is an object of the present invention to provide a new and improved classification and evaluation apparatus capable of satisfying the requirements.

  In order to solve the above-described problem, according to one aspect of the present invention, there is provided a classification evaluation apparatus for evaluating a result of classifying a plurality of data based on data characteristics. A classification comprising: a similarity calculation unit for calculating a similarity with a feature representing data belonging to the class to which the data belongs; and a classification evaluation unit for evaluating the classification based on the similarity calculated for each data belonging to the class An evaluation device is provided.

  With this configuration, it is possible to automatically evaluate and display the effectiveness of the classification in a case where the classification used for data search or the like is further narrowed down from a plurality of classifications, and the user can quickly and easily The target classification can be found.

  Further, the classification evaluation unit may further evaluate the classification based on the number of data belonging to the classification. For example, the classification may be evaluated based on a multiplier between the similarity value calculated by the similarity calculation unit and the number of data. As a result, a higher evaluation can be obtained for a classification that has a larger number of data and is easier to overlook the characteristics of the entire data.

  Further, the classification evaluation unit may evaluate the classification based on the average value of the similarity of the data belonging to the classification. As a result, it is possible to perform evaluation in consideration of a group of all data belonging to the classification.

  The classification may include a plurality of cells having different features, and the similarity calculation unit may calculate the similarity between the data features and the features of the cells to which the data belongs. Thereby, for example, when classification is performed using a self-organizing map including a plurality of cells, the characteristics of the classification method can be used for evaluation.

  The classification evaluation unit may further evaluate the classification based on the number of cells belonging to the classification. Alternatively, the classification evaluation unit may evaluate the classification based on the number of cells to which one or more data belong among the cells belonging to the classification. As a result, the classification can be evaluated based on whether or not the data belonging to the classification is dispersed throughout the classification. For example, if the ratio of the number of cells to which one or more data belongs among the cells belonging to the classification is large, it can be determined that the data is dispersed throughout the classification.

  The classification evaluation unit may evaluate the classification based on a value obtained by dividing the total number of similarities of all data belonging to the classification by the number of cells. As a result, a higher evaluation can be obtained for a classification having a higher degree of data similarity, and a higher evaluation can be obtained for a classification having a larger number of data. Alternatively, the data is not distributed over the entire classification, and a higher classification can be obtained for a classification that is partly collected.

  As described above, according to the present invention, it is possible to provide a user with an index for selecting a target classification from classifications obtained by automatic classification.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

(First embodiment)
First, a classification evaluation apparatus according to a first embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram illustrating a schematic configuration of a classification evaluation apparatus 100 according to the present embodiment.

  Next, details of the classification evaluation apparatus 100 according to the present embodiment will be described. As shown in FIG. 1, the classification evaluation apparatus 100 according to the present embodiment includes a feature vector extraction unit 101, an automatic classification unit 102, a reference vector acquisition unit 103, a similarity calculation unit 104, and a data number calculation unit 105. And a classification evaluation unit 106 and a display unit 107. The classification evaluation apparatus 100 receives a database 110 in which a plurality of data is stored as an input, evaluates a classification result obtained by classifying the input data in the data database 110, and presents it to the user. Hereinafter, each part of the classification evaluation apparatus 100 will be described.

(Feature vector extraction unit 101)
The feature vector extraction unit 101 is a functional unit for vectorizing features of data in the database 110 that is a classification target. The feature vector extraction unit 101 extracts one or more feature elements from the data in the database 110, and obtains a vector composed of the feature elements extracted for each data. Hereinafter, a vector obtained for each data is referred to as a feature vector.

(Automatic classification unit 102)
The automatic classification unit 102 is a functional unit for classifying each data in the database 110. The automatic classification unit 102 uses the feature vector obtained by the feature vector extraction unit 101 to calculate the similarity between the data in the database 110 and classifies the similar data as the same classification. As a classification method used in the automatic classification unit 102, for example, a classification method such as a k-means method can be used.

(Reference vector acquisition unit 103)
The reference vector acquisition unit 103 is a functional unit for acquiring a feature vector representing the classification of each classification obtained by the automatic classification unit 102. Here, a feature vector representing each acquired category is referred to as a reference vector. The reference vector can be obtained by, for example, a method of calculating an average of feature vectors of all data included in the classification.

(Similarity calculation unit 104)
The similarity calculation unit 104 is a functional unit for calculating the similarity between the feature vector of included data and the reference vector of the classification for each classification. The similarity is obtained, for example, by calculating the Euclidean distance, the inner product, or the like between the feature vector of each data and the classification reference vector. The calculated similarity value is given from the similarity calculation unit 104 to the classification evaluation unit 106.

(Data count calculation unit 105)
The data number calculation unit 105 is a functional unit for obtaining the number of data included in each classification. The calculated number of data of each classification is given from the data number calculation unit 105 to the classification evaluation unit 106.

(Classification evaluation unit 106)
The classification evaluation unit 106 is a functional unit for calculating an evaluation value indicating the effectiveness of data included in each classification and evaluating the classification. For the evaluation value calculation, the similarity calculated by the similarity calculation unit 104 and the number of data of each classification calculated by the data number calculation unit 105 are used.

  For example, the classification evaluation unit 106 may calculate the evaluation value by obtaining an average value or a maximum value of the similarity of the data included in each classification. Thereby, a higher evaluation value can be obtained as the similarity of the data included in the classification is higher (data variation is smaller). Alternatively, the evaluation value may be calculated by multiplying the average value of the degrees of similarity of the data included in the classification by the number of data (summing the degrees of similarity). Thereby, a higher evaluation value can be obtained as the number of data increases.

  Alternatively, when the number of data is large, the evaluation value may be calculated using the logarithm of the number of data instead of the number of data. As a result, it is possible to calculate the evaluation value such that importance is attached to the similarity of data when the number of data is large, and importance is attached to the number of data when the number of data is small.

  Note that the evaluation value calculation method shown here is merely an example, and the evaluation value can be calculated by various calculation methods using the similarity of data and the number of data.

(Display unit 107)
The display unit 107 is a functional unit for displaying according to the evaluation value obtained by each classification and classification evaluation unit 106 obtained by the automatic classification unit 102. For example, the classifications may be displayed in a list in descending order of evaluation values, or the classifications may be displayed in different colors according to the evaluation value levels. In addition, the classification with a high evaluation value may be displayed in a dark color so that the user can easily find the classification with a high evaluation value.

  The configuration of the classification evaluation apparatus 100 has been described above. Each unit of the classification evaluation apparatus 100 may be configured by software in which a program module that can execute each function described above is installed in an information processing apparatus such as a computer, or a processor that can execute each function described above. Or the like.

  Next, an example of the classification evaluation process executed by the classification evaluation apparatus 100 according to the present embodiment will be described based on FIGS. Here, FIG. 2 is a flowchart of the classification evaluation process executed by the classification evaluation apparatus 100. FIG. 3 is a diagram schematically illustrating an example of data classified by the automatic classification unit 102 of the classification evaluation apparatus 100. FIG. 4 is a diagram showing the similarity to the data of each classification calculated from the classification example shown in FIG. 3, the average value of the similarity, the number of data, and the evaluation value.

  Referring to FIG. 2, first, in step S <b> 120, the feature vector extraction unit 101 extracts one or more feature elements from the data in the database 110. Next, in step S122, a feature vector of each data is obtained based on the extracted feature elements. Further, in step S124, the automatic classification unit 102 classifies the data in the database 110 using the feature vector.

  It is assumed that the data in the database 110 has been classified as shown in FIG. 3, for example, through the above processing. Although FIG. 3 is shown in two dimensions for the sake of explanation, it may generally be multidimensional. In FIG. 3, white circles indicate the feature vectors of the data D1 to D12 in the database 110 plotted in the space. A solid straight line represents a classification boundary obtained by the operation of the automatic classification unit 102 in step S124. Here, it is assumed that a classification including four classifications A11, A12, A13, and A14 is obtained by the classification processing in step S124. Here, for explanation, only the above four classifications will be described as an example. However, the present invention is not limited to this, and can be applied even when a larger number of classifications are obtained.

  Hereinafter, the classification evaluation process illustrated in FIG. 2 will be described with reference to the classification result example of FIG. In step S126, the reference vector acquisition unit 103 acquires the reference vector of each classification obtained in step S124. In FIG. 3, the reference vector of each classification obtained here is represented by x. For example, in the case of classification A11, since the included data is only one of D1, the feature vector of the data becomes the reference vector C11. In the case of the classification A12, since the four data D2 to D5 are included, a vector serving as the center of gravity is calculated to obtain C12 which takes the reference phrase shown in FIG. Similarly, for the classifications A13 and A14, reference vectors C13 and C14 are obtained by calculating the center of gravity of the data.

  Next, in step S128, the similarity between the data belonging to each category and the category is calculated. Specifically, the distance d between the feature vector of each data and the classification reference vector is calculated, and the value obtained by the calculation of 1 / (1 + d) is set as the similarity. For example, taking the classification A12 shown in FIG. 3 as an example, first, the distance between the reference vector C12 of the classification A12 and D2 to D5 is calculated. Here, it is assumed that the distances between C12 and D2 to D5 are 1.0, 0.2, 1.0, and 1.1, respectively. Next, 1 / (1 + d) is calculated with respect to the obtained distance value to obtain the similarity of 0.5, 0.83, 0.5, and 0.48. Similar calculations are performed for the classifications A11, A12, and A14, and the similarity between each data and the classification is calculated.

  FIG. 4 shows the similarity value between each of the data D1 to D12 calculated in step S128 and the categories A11 to A14, and the average value of the similarities of each category. From the average value of the similarities, the degree of variation of data in each classification can be found. That is, it can be seen that the higher the average value, the less data variation, and the classification composed of similar data. Referring to FIG. 4, it can be seen that, among the categories A11 to A14, the category A12 has the most data, and the category A13 has the most variation in data.

  Next, in step S130, the number of data of each classification is calculated. In the example of FIG. 3, the calculated number of data of the classifications A11 to A14 is shown in FIG. As shown in FIG. 4, the numbers of data of the classifications A11 to A14 are 1, 4, 5, and 2, respectively.

  Next, in step S132, an evaluation value for each classification is calculated. Here, the total value of the similarities with the data belonging to each classification obtained in step S128 is used as the evaluation value. In the example of FIG. 3, the calculated evaluation values of the classifications A11 to A14 are shown in FIG. As shown in FIG. 4, the evaluation values of the classifications A11 to A14 are 1.0, 2.31, 1.46, and 1.06, respectively.

  In step S134, the classification is displayed in the order of the evaluation values obtained in step S132. Therefore, the display order of the classification is A12 → A13 → A14 → A11. The classification may be displayed by increasing the visibility by displaying a large element of the reference vector or by displaying data that is closest to the reference vector.

  As described above, the classification composed of similar data, such as A12, has a high evaluation value, and data including a large number of data such as A13. Evaluation values are lower for classifications with large variations. Therefore, it can be expected that classifications composed of similar data are gathered as information, and it can be determined that the effectiveness of the data is high.

  In addition, in the case of a classification where the number of data is as small as 2 even when the distance between the data is close to each other as in A14, the evaluation value is lower than in the classification where the variation between data is large and the number of data is large as in A13. Thereby, it becomes easy to confirm the classification | category which contains more data and it is easy to overlook the characteristics of the whole database first.

  In this way, it is possible to discriminate classifications that are estimated to have high data validity and quality, and provide information that enables the overall picture of the data to be confirmed more efficiently than a map of all classifications. Can do.

(Second Embodiment)
Next, a classification evaluation apparatus according to the second embodiment of the present invention will be described with reference to FIG. FIG. 5 is a block diagram illustrating a schematic configuration of the classification evaluation apparatus 200 according to the present embodiment.

  Next, details of the classification evaluation apparatus 200 according to the present embodiment will be described. As shown in FIG. 5, the classification evaluation apparatus 200 according to the present embodiment includes a feature vector extraction unit 201, a self-organizing map creation unit 202, a cell classification unit 203, a similarity calculation unit 204, and a data number calculation. A unit 205, a cell number calculation unit 206, a classification evaluation unit 207, and a display unit 208 are configured. The classification evaluation apparatus 200 according to the present embodiment evaluates classification when data in the database 210 is classified by a self-organizing map, and performs evaluation according to the characteristics of the self-organizing map. Hereinafter, each part of the classification evaluation apparatus 200 will be described.

(Feature vector extraction unit 201)
The feature vector extraction unit 201 has substantially the same function as the feature vector extraction unit 101 according to the first embodiment described above. The feature vector extraction unit 201 extracts one or more feature elements from the data in the database 210 and obtains a feature vector of each data.

(Self-organizing map creation unit 202)
The self-organizing map creating unit 202 is a functional unit for creating a self-organizing map with the feature vector of each data in the database 210 as an input. As a method for creating the self-organizing map used here, a generally used method can be used. The created self-organizing map is composed of a plurality of cells, and each cell has a different feature vector. The feature vector that each cell has is called a reference vector.

(Cell classification unit 203)
The cell classification unit 203 is a functional unit for classifying each cell of the self-organizing map. The cell classification unit 203 classifies cells based on the reference vector of each cell. In the present embodiment, a feature element that represents the reference vector of each cell is used as a representative element, and cells having the same representative element are classified as the same classification. As the representative element of each cell, for example, an element having the largest value among the characteristic elements constituting the reference vector of each cell may be used as the representative element of the cell.

(Similarity calculation unit 204)
The similarity calculation unit 204 is a functional unit for calculating the similarity between each data in the database 210 and each cell of the self-organizing map and classifying the data into any cell. The similarity between the data and the cell is obtained, for example, by calculating the Euclidean distance, the inner product, etc. between the feature vector of each data and the reference vector of each cell. Further, the similarity calculation unit 204 searches for a cell having the highest similarity with the data, that is, a cell most similar to the data, and classifies the data into the cell. Further, the similarity (maximum value of similarity) with the classified cell is set as the similarity between the data and the classification, and is given to the classification evaluation unit 207.

(Data number calculation unit 205)
The data number calculation unit 205 is a functional unit for calculating the number of data included in each classification. The data number calculation unit 205 calculates the number of data by summing the number of data belonging to the cells classified into the same classification in the cell classification unit. The calculated number of data of each classification is given from the data number calculation unit 205 to the classification evaluation unit 207.

(Cell number calculation unit 206)
The cell number calculation unit 206 is a functional unit for calculating the number of cells included in each classification. The cell number calculation unit 206 may calculate the number of only cells into which one or more data are classified by the similarity calculation unit 204 in addition to the number of all cells included in the classification. The calculated number of cells of each classification (or the number of cells to which the data belongs) is given from the cell number calculation unit 206 to the classification evaluation unit 207.

(Classification evaluation unit 207)
The classification evaluation unit 207 calculates an evaluation value indicating the effectiveness of data included in each classification and evaluates the classification. For the calculation, the similarity with each data classification obtained by the similarity calculation unit 204, the number of data calculated by the data number calculation unit 205, the number of cells calculated by the cell number calculation unit 206, and the like are used.

  Similar to the classification evaluation unit 106 according to the first embodiment described above, the classification evaluation unit 207 includes an average value and a maximum value of similarity of data included in each classification, a sum of similarities, a similarity and the number of data. The evaluation value may be calculated by obtaining a multiplier of. Further, a logarithm of the number of data may be used instead of the number of data.

  The evaluation value may be calculated by dividing the evaluation value obtained from the similarity and the number of data by the same method as the classification evaluation unit 106 according to the first embodiment by the number of cells belonging to the classification. Alternatively, instead of the total number of cells included in the classification, the number of cells to which one or more data belongs may be used. As a result, a higher evaluation value is obtained as the number of cells included in the classification or the number of cells to which one or more data belongs is smaller. That is, it can be determined that there is little variation in data within the classification.

(Display unit 208)
The display unit 208 has substantially the same function as the display unit 107 according to the first embodiment described above.

  The configuration of the classification evaluation apparatus 200 has been described above. Each unit of the classification evaluation apparatus 200 may be configured by software in which a program module that can execute each function described above is installed in an information processing apparatus such as a computer, or a processor that can execute each function described above. Or the like.

  Next, based on FIG.6 and FIG.7, an example of the classification evaluation process performed by the classification evaluation apparatus 200 concerning this embodiment is demonstrated. Here, FIG. 6 is a flowchart of the classification evaluation process executed by the classification evaluation apparatus 200. FIG. 7 is an explanatory diagram showing an example of a self-organizing map created by the self-organizing map creating unit 202 of the classification evaluation apparatus 200. FIG. 8 is an explanatory diagram showing the number of data, the number of cells, and the evaluation value of each classification calculated from the example of the self-organizing map shown in FIG.

  Referring to FIG. 6, first, in step S <b> 220, the feature vector extraction unit 201 extracts one or more feature elements from the data in the database 210. Next, in step S222, a feature vector of each data is obtained based on the extracted feature elements.

  Next, in step S224, the self-organizing map creating unit 202 creates a map as shown in FIG. 7 based on the obtained feature vector. The map shown in FIG. 7 is composed of 6 × 6 36 cells A1 to F6, and each cell has a different feature vector. The feature vector that the cell has is called a reference vector.

  Next, in step S226, the cell classification unit 203 acquires, as a representative element, an element having the largest value among the feature elements included in the reference vector of each cell. For example, in the example of FIG. 7, each cell has a 7-dimensional or higher feature vector (referred to as a reference vector) including feature elements X1 to X7, and the value is the highest among the feature elements including X1 to X7. Get the big element. In FIG. 7, the representative element of each cell acquired in step S226 is shown in the upper stage of each cell.

  Next, in step S228, the cell classification unit 203 classifies the cells with the same representative element as the same classification element. In FIG. 7, the boundaries of the classification of the cells classified by the process of step S228 are indicated by bold lines. Therefore, a group of cells delimited by the thick lines in FIG. 7 corresponds to one classification, and data classified into cells included in the classification is data belonging to the classification.

  In general, the visualization by the self-organizing map is up to this point, and it is difficult to select an appropriate classification when the number of classifications is large and the quality varies greatly. In the present embodiment, an index for easily selecting an appropriate classification can be presented to the user through the processing of the following steps.

  Next, in step S230, the similarity calculation unit 204 calculates the similarity between each data and each of the cells A1 to F6, and classifies the data into the cell having the highest similarity. The similarity calculation unit 204 calculates the distance between the feature vector of the data and the reference vectors of all the cells for each data, and calculates the similarity. Further, the data is classified into the cell having the highest similarity. As a result of such classification, the number of data belonging to each cell is shown in the lower part of each cell in FIG.

  Next, in step S232, the data number calculation unit 205 calculates the number of data belonging to each classification. This is calculated by summing up the number of data classified into the cells belonging to each classification as a result of the classification in step S230. For example, in the example of FIG. 7, in the case of the classification whose representative element is X2, the cells belonging to the classification are D1, D2, E1, E2, E3, F1, F2, and F3, so the number of data belonging to those cells (The lower value of each cell in FIG. 7) is summed to obtain 0 + 0 + 2 + 0 + 0 + 6 + 3 + 0 = 11. For the other classifications, the number of data is calculated by the same calculation. FIG. 8 shows the calculated number of data for each classification.

  Next, in step S234, the cell number calculation unit 206 calculates the number of cells belonging to each classification. For example, in the example of FIG. 7, in the case of a classification whose representative element is X2, there are eight cells belonging to the classification, D1, D2, E1, E2, E3, F1, F2, and F3. At this time, the number of cells may be calculated by excluding those in which the number of data belonging to the cell is zero. In that case, in the case of classification of X2, the number of cells is three, E1, F1, and F2. For other classifications, the number of cells is calculated by the same calculation. FIG. 8 shows the calculated number of data for each classification.

  Next, in step S236, the classification evaluation unit 207 calculates an evaluation value for each classification. For example, if the average value of the similarity of the classification whose representative element is X2 is 0.8, the evaluation value for this classification is (average value of similarity) × (number of data) ÷ (data of 1 or more Cell number) = 0.8 × 11 ÷ 3 = 2.93.

  For comparison, in the example of FIG. 7, when the average value of the similarities of all the classifications is 0.8, the evaluation value calculated by the above formula is shown in FIG. For example, the classification of X2 and X4 is 11 that has the largest number of data and seems to be a classification that represents the entire data, but the evaluation value of X2 is higher. This means that in X4 classification, the data is distributed in many cells and is not necessarily data gathered with the same characteristics, but data that can be divided into finer categories within the classification. Yes. On the other hand, in the classification of X2, it can be seen that the data is concentrated in a part of the cells in the classification, and the relevance between the data is high and the effectiveness is high.

  Finally, in step S238, the classification is displayed based on the evaluation value. For example, the evaluation values are displayed in descending order, that is, in the order of X7 → X2 → X3 → X5 → X6 → X4 → X1. As a result, X7 is displayed with the highest emphasis, and the user can easily confirm the classification with high effectiveness.

  As described above, when data is classified into a plurality of classifications, and the classification used for data retrieval or the like is further narrowed down from among the classifications, the user can be quickly evaluated by automatically evaluating and displaying the classifications. You can easily find the target classification.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to the example which concerns. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are naturally within the technical scope of the present invention. Understood.

It is a block diagram which shows schematic structure of the classification | category evaluation apparatus concerning the 1st Embodiment of this invention. It is a flowchart of the classification evaluation process performed by the classification evaluation apparatus concerning the embodiment. It is explanatory drawing which shows an example of the data classified by the automatic classification | category part concerning the embodiment. It is explanatory drawing which shows the similarity with the data of each classification | category calculated from the classification example shown in FIG. 3, the average value of similarity, the number of data, and an evaluation value. It is a block diagram which shows schematic structure of the classification | category evaluation apparatus concerning the 2nd Embodiment of this invention. It is a flowchart of the classification evaluation process performed by the classification evaluation apparatus concerning the embodiment. It is explanatory drawing which shows the example of the self-organization map created by the self-organization map creation part concerning the embodiment. It is explanatory drawing which shows the data number of each classification | category calculated from the example of the self-organization map shown in FIG. 7, the number of cells, and an evaluation value.

Explanation of symbols

100, 200 Classification evaluation apparatus 101 Feature vector extraction unit 102 Automatic classification unit 103 Reference vector acquisition unit 104 Similarity calculation unit 105 Data number calculation unit 106 Classification evaluation unit 107 Display unit 110, 210 Database 201 Feature vector extraction unit 202 Self-organization Map creation unit 203 Cell classification unit 204 Similarity calculation unit 205 Data number calculation unit 206 Cell number calculation unit 207 Classification evaluation unit 208 Display unit

Claims (7)

A classification evaluation apparatus for evaluating a result of classifying a plurality of data based on data characteristics,
For each data, a similarity calculation unit that calculates the similarity between the characteristics of the data and the characteristics representing the data belonging to the classification to which the data belongs;
A classification evaluation unit that evaluates the classification based on the similarity calculated for each data belonging to the classification;
A classification evaluation apparatus comprising:   The classification evaluation apparatus according to claim 1, wherein the classification evaluation unit further evaluates the classification based on the number of the data belonging to the classification.   The classification evaluation apparatus according to claim 1, wherein the classification evaluation unit evaluates the classification based on an average value of the similarities of the data belonging to the classification. The classification is composed of a plurality of cells having different characteristics,
The classification evaluation apparatus according to claim 1, wherein the similarity calculation unit calculates a similarity between a feature of the data and a feature of a cell to which the data belongs.   The classification evaluation apparatus according to claim 4, wherein the classification evaluation unit further evaluates the classification based on the number of the cells belonging to the classification.   6. The classification evaluation unit according to claim 4, wherein the classification evaluation unit evaluates the classification based on the number of the cells to which one or more of the data belong among the cells belonging to the classification. Classification evaluation device.   The classification evaluation unit evaluates the classification based on a value obtained by dividing the total number of the similarities of all the data belonging to the classification by the number of the cells. Classification evaluation apparatus in any one of.

Images