JP2009230323A - Information analyzing device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information analyzing device capable of adaptively determining the representative vector of a group with respect to the characteristic of the distribution of components of a feature vector concerning an analysis object even if the characteristic is changed. <P>SOLUTION: The feature vectors are acquired which are classified into one of the plurality of groups. One of a plurality of kinds of representative vector determining methods is selected based on evaluation about the distribution of the components with values equal to or larger than prescribed component thresholds of the feature vectors. The representative vector of each group is determined by the selected representative vector determining method. The value to indicate the degree of holding each of the plurality of kinds of features by the corresponding analysis object is defined as the component in the feature vector. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an information analysis apparatus and a program.

  With the progress of the information society in recent years, a huge amount of computerized information has been accumulated in computers. For this reason, it is more difficult than ever to find valuable information from a large amount of accumulated information and to understand the overall structure of the information. In order to cope with these difficulties, it is required to classify these pieces of information by system and present them to the user.

  As a method of presenting such information, it is considered to visualize the classified information on a graph or map in a two-dimensional or three-dimensional space. This is effective in that the relationship between classified information can be intuitively understood. As one of the visualization methods described above, there is a conventional method in which each piece of information is represented as an analysis target by a multidimensional feature vector, the feature vector is analyzed, and the distribution of the analysis target is represented in a map. .

For example, Non-Patent Document 1 discloses the following method as an information analysis and visualization method when a large amount of documents are to be analyzed. First, feature vectors whose components are keywords included in each document are created for each document, and the analysis objects are classified into groups by clustering the feature vectors. Next, the center of gravity of each classified group is calculated as a representative vector of each group, and the representative vector is mapped to a two-dimensional space by principal component analysis. Furthermore, the feature vector corresponding to each document is mapped by the multidimensional scaling method so as to preserve the distances from the representative vectors of all the already mapped groups. By this method of mapping in two steps, each analysis object is plotted at a position close to the group to which it belongs, so the relationship between the analysis objects classified into groups is visualized.
James A. Wise, “The Ecological Approach to Text Visualization”, Journal of the American Society for Information Science (JOURNALOF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE), (USA), November 1999

  However, with the conventional method, it may be difficult to analyze the interrelationship between the above representative vectors due to a change in the distribution characteristics of the component of the feature vector to be analyzed.

  For example, when the representative vector is determined based on the center of gravity of the feature vector of each group, the feature of the group becomes inconspicuous when there are many overlapping feature vector components between groups.

  The present invention has been made in view of the above problems, and its purpose is to adaptively determine a representative vector according to the characteristics of the distribution of the feature vector components related to the analysis object even if the characteristics change. An object of the present invention is to provide an information analysis apparatus and a program that can be used.

  The invention of claim 1 is an information analysis apparatus, and for each of a plurality of analysis objects classified into any of a plurality of groups, a value indicating a degree of the analysis object having each of a plurality of types of features is defined as its component On the basis of evaluation relating to the distribution of components having a value equal to or greater than a predetermined component value threshold value of the feature vector acquired by the acquiring unit for all or part of the plurality of analysis objects. A selection unit that selects one of a plurality of types of representative vector determination methods, and a representative vector determination method selected by the selection unit, wherein the analysis unit classified into each group is acquired by the acquisition unit Determining means for determining a representative vector of the group based on the vector.

  According to a second aspect of the present invention, in the first aspect of the invention, the plurality of types of representative vector determining methods are limited in the amount of information used to determine the representative vector of the group out of the feature vector information related to each group. Including at least one limited representative vector determination method, wherein the selection unit has a value indicating the number of components having a value equal to or greater than the predetermined component value threshold of the feature vector acquired by the acquisition unit than a predetermined component number threshold If there are many, one of the limited representative vector determination methods is selected.

  The invention of claim 3 is the invention of claim 1 or 2, wherein the plurality of types of representative vector determining methods include a reference representative vector determining method for determining a centroid vector of feature vectors of each group, and a predetermined component value threshold value or more. A low overlap representative vector determination method that lowers the degree of overlap between components having values among representative vectors of different groups, and a representative of groups having components that have a value equal to or greater than a predetermined component value threshold. A high overlap representative vector determination method that makes a degree of overlap between vectors higher than that of the reference representative vector determination method, and the selection unit relates to a group having different components having a value equal to or greater than the predetermined component value threshold When the degree of overlap between feature vectors is evaluated and the degree of overlap is greater than or equal to a predetermined level, the method for determining the low overlap representative vector Select, if the extent of the overlap is less than a predetermined degree, selects the high overlapping representative vector determination method, characterized in that.

  According to a fourth aspect of the present invention, in the third aspect of the invention, the selecting means calculates a degree of proximity between feature vectors of different groups, and evaluates the degree of overlap according to the degree of proximity. Features.

  According to a fifth aspect of the present invention, in the invention of the third or fourth aspect, the low-overlapping representative vector determination method includes the reference representative vector among the feature vectors acquired by the acquisition unit for the analysis target classified into the groups. The method is characterized in that the representative vector of the group is determined by using only a part selected based on the representative vector of the group determined by the determination method.

  According to a sixth aspect of the present invention, in any one of the first to fifth aspects, coordinate information is generated by projecting the representative vector of each group determined by the determining means to the same coordinate system as a space in which a map is generated. Coordinate information generating means is further included.

  The invention according to claim 7 obtains a feature vector whose component is a value indicating a degree of the analysis target for each of the plurality of types of features for each of the plurality of analysis targets classified into any of a plurality of groups. Determining a plurality of types of representative vectors based on an evaluation relating to a distribution of components having a value equal to or greater than a predetermined component value threshold of a feature vector acquired by the acquiring unit for all or a part of the plurality of analysis objects; Based on the feature vector acquired by the acquisition unit for the analysis target classified into each group by the selection unit for selecting one of the methods, and the representative vector determination method selected by the selection unit, This is a program for causing a computer to function as a determination means for determining a representative vector.

  According to the first and seventh aspects of the present invention, the representative vector can be determined adaptively even when the distribution characteristics of the component of the feature vector related to the analysis object change.

  According to the second aspect of the present invention, when the value indicating the number of components having a value equal to or greater than the predetermined component value threshold of the feature vector to be analyzed is greater than the predetermined component number threshold, the relationship can be analyzed more appropriately. A representative vector can be determined. This is because the probability that the feature indicated by each component is information representative of the feature vector is low.

  According to the invention of claim 3, when the degree of overlapping between the feature vectors related to different groups with components having a value equal to or greater than a predetermined component value threshold is a predetermined degree or more, the overlap between representative vectors is reduced, When the overlapping degree is smaller than the predetermined degree, the representative vector whose relationship can be analyzed more appropriately can be determined by increasing the overlapping between the representative vectors.

  According to the fourth aspect of the present invention, the degree of proximity between feature vectors related to different groups can be used as an index for evaluating the degree of overlap between feature vectors related to different groups.

  According to the invention of claim 5, as a representative vector determining method for reducing the overlap between representative vectors, only a part selected based on the vector determined by the reference representative vector determining method is used. Can be used.

  According to the sixth aspect of the present invention, even if the characteristics of the distribution of the component of the feature vector related to the analysis object change, the coordinate information of the representative vector that is the original information of the map can be adaptively generated.

  Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is an example of a configuration diagram of an information analysis apparatus 1 according to an embodiment of the present invention. As shown in FIG. 1, the information analysis apparatus 1 includes a CPU 11, a memory 12, an input unit 13, and an output unit 14.

  The CPU 11 operates according to a program stored in the memory 12. The program may be provided by being stored in an information recording medium such as a CD-ROM or DVD-ROM, or may be provided via a network such as the Internet.

  The memory 12 includes a memory device such as a RAM or a ROM, or a recording device such as a hard disk. The memory 12 stores the program. The memory 12 stores information and calculation results input from each unit.

  The input unit 13 includes a means for communicating with an external computer, an external recording device such as a removable medium, and a keyboard and a mouse for receiving instructions from a user. The input unit 13 outputs information input from the outside to the CPU 11 and the memory 12 based on the control of the CPU 11.

  The output unit 14 is a communication unit with an external computer or a display output unit to the user. The output unit 14 outputs the processing result of the CPU 11 to the outside based on the control of the CPU 11.

  FIG. 2 is a functional block diagram showing functions realized by the information analysis apparatus 1. The information analysis apparatus 1 functionally includes a feature vector acquisition unit 21, a determination method selection unit 22, a representative vector determination unit 23, a coordinate information generation unit 24, and a drawing unit 25. These functions are realized by the CPU 11 executing a program stored in the memory 12 and controlling the input unit 13 and the output unit 14.

  The feature vector acquisition unit 21 is realized centering on the CPU 11, the memory 12, and the input unit 13. The feature vector acquisition unit 21 inputs information about a plurality of feature vectors previously classified into a plurality of groups from the input unit 13 and stores the information in the memory 12.

  Here, the analysis target and the feature vector related to the input data will be described. In the present embodiment, the analysis target is an electronic document such as a patent document. One feature vector corresponds to one analysis target, and the component of the feature vector represents the degree of having each of a plurality of types of features of the analysis target. In this embodiment, the plural types of features of the analysis target are, for example, keywords extracted from bibliographic items such as keywords obtained by morphological analysis or F-terms in patent literature, and the degree of the feature is determined for each keyword. Appearance frequency. In this case, the value of the feature vector component is the appearance frequency of the keyword extracted from the electronic document.

  FIG. 3 is a diagram illustrating an example of a method for generating a feature vector from an analysis target. 3A, 3B, and 3C show feature vector generation when the analysis target is a sentence, and FIGS. 3D, 3E, and 3F show the case where the analysis target is a bibliographic item. This shows the generation of feature vectors.

  FIG. 3A shows sentences of analysis objects 1 and 2. The sentence in this example is a short sentence composed of a small number of sentences, but may be a long sentence composed of a plurality of chapters. For these sentences, morphological analysis and removal of keywords having no meaning such as particles are performed and the appearance frequency is calculated. FIG. 3B shows an example of the appearance frequency of the result. Using this, a feature vector in which a component value is determined based on the appearance frequency of each analysis target keyword (feature) is generated. FIG. 3C shows an example of the resulting feature vector. In FIG. 3C, feature vectors are expressed in a table in order to easily understand the correspondence between keywords (features) and component values. In this table, for example, the feature vector of the analysis target 1 is (1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 0, 0, 0, 0, 0). . Thereafter, the same expression is used for vectors. FIG. 3D shows bibliographic item keywords of the analysis objects 3 and 4. Here, F-term classification in patent literature is used as an example of a bibliographic item keyword. For bibliographic item keywords, direct keyword (feature) extraction based on the document structure and appearance frequency are calculated (an example of the resulting appearance frequency is shown in FIG. 3E). A vector is generated (an example of the resulting feature vector is FIG. 3 (f)). Note that the feature vector shown in FIG. 3 is not normalized so as to have a length of 1 for easy understanding, but may be normalized. Note that feature vectors can be generated by the same method even when the number of analysis objects is large.

  Further, the feature vectors are previously classified into a plurality of groups. The classification method is not particularly limited, and the classification may be performed by a known clustering method such as a K-means method, or the classification may be performed artificially.

  Note that the feature vector acquisition unit 21 not only inputs information from the outside via the input unit 13 but also internally acquires information on the feature vector recorded in advance on the memory 12 by another program or the like. Also good.

  The determination method selection unit 22 is realized centering on the CPU 11. The determination method selection unit 22 selects one of a plurality of types of representative vector determination methods by evaluating the distribution of components having a value equal to or greater than a predetermined component value threshold of the plurality of feature vectors stored in the memory 12. . A processing flow of the determination method selection unit 22 in the present embodiment is shown in FIG.

  S31 is a step of evaluating the number of components having a value equal to or larger than a predetermined component value threshold of the feature vector as the above-described component distribution. Specifically, processing is performed according to the following procedure.

  The determination method selection unit 22 counts the number of components whose component value of each acquired feature vector is greater than zero. Further, an average value of the number of components of all the acquired feature vectors is taken, and it is determined whether or not it is equal to or greater than a predetermined component number threshold.

  For example, the case where the predetermined component number threshold is set to 5 will be described with reference to FIG. In the case of the bibliographic item of FIG. 3 (f), the number of components whose feature vector components of analysis objects 3 and 4 are greater than 0 is 3 and 4, respectively. The average value of the number of components is 3.5, and is determined to be smaller than the predetermined component number threshold. On the other hand, in the case of the sentence in FIG. 3C, the number of components satisfying the above conditions in the feature vectors of the analysis targets 1 and 2 is 10 and 6, respectively. The average value of the number of components is 8, which is determined to be larger than the predetermined component number threshold.

  Here, the reason why the component whose component value is 0 or more is counted is because, in this embodiment, whether or not a keyword appears is used as a criterion. In this case, the predetermined component value threshold is zero.

  If the average value of the number of components described above is equal to or less than a predetermined component number threshold, a method that uses all the information of the feature vectors of each group is selected as the representative vector determination method for each group. This is because it is highly likely that each component represents a feature vector. An example of such information is information that has been artificially indexed, such as the bibliographic items of FIG. The representative vector determination method selected in the present embodiment is a method of calculating the center of gravity of the feature vectors classified into each group (S32).

  On the other hand, if the average value of the number of components described above is larger than a predetermined component number threshold value, the distance between the classifications is further determined (S33). In this case, it is highly likely that each component does not represent a feature vector, and it is highly likely that unnecessary features are mixed. In S33, the distribution of other types of components is evaluated, and a representative vector generation method (restricted representative vector determination method) that reduces unnecessary features is selected (S34, S35). Details will be described later in the description of S33.

  The predetermined component number threshold value may be a predetermined fixed value, or may be a value determined by calculation such as (the number of dimensions of the feature vector ÷ 3) using the feature vector acquired by the feature vector acquisition unit 21. . In this embodiment, the average value of the number of feature vector components is used as a determination condition. However, the number of feature vector components only needs to be representative, for example, using a median value or a minimum value. Also good.

  S33 is a step of evaluating the degree of overlapping of components having a value equal to or greater than a predetermined component value threshold between feature vectors in different groups as the distribution of components in the feature vector. Specifically, processing is performed according to the following procedure.

  First, the reference representative vector of each group is calculated, and the Euclidean distance between the reference representative vectors is obtained. Here, the center of gravity of the feature vector of each group is used as the reference representative vector. The average distance between the reference representative vectors for all combinations indicates the degree of overlap.

  Here, the Euclidean distance indicates the degree of overlap, for example, for the following reason. In this embodiment, each component of the feature vector is the appearance frequency of the keyword, and since the value of each component of the reference representative vector is 0 or more, if there are many overlapping components, the difference in the values of the overlapping components is There is a tendency that the Euclidean distance expressed by the square of the sum of squares becomes smaller and closer. Note that the Euclidean distance between vectors is maximum when the respective components are orthogonal to each other. In this case, the inner product between the vectors is 0, that is, all the components do not overlap. Here, the distance is not limited to the Euclidean distance, and may be a distance calculated by a known distance calculation method such as a cosine distance. The distance between groups may not necessarily be calculated using a reference representative vector, and a feature vector closest to the other group may be selected from each group and the distance between them may be calculated.

  If the average distance is greater than or equal to a predetermined distance, it is determined that the degree of overlap of components having a value greater than or equal to a predetermined component value threshold between feature vectors in different groups is greater than or equal to a predetermined level. A representative vector determination method (low overlap vector determination method) that reduces the degree of overlap between components that are equal to or greater than a predetermined component value threshold is selected. The method selected in this embodiment is a method in which a feature vector closest to the center of gravity of each group is used as a representative vector (S34).

  FIG. 5 is a diagram showing an example in which the feature vector closest to the center of gravity of the group is determined as the representative vector. In this figure, group 1 has four feature vectors V11, V12, V13, and V14, and group 2 has four feature vectors V21, V22, V23, and V24. The representative vectors are feature vectors V14 and V22 that are closest to the group centroid. It can be seen that the representative vector has fewer non-zero components than the group centroid, and the features are more clarified. Furthermore, by selecting the feature vector closest to the center of gravity as the representative vector, the degree of overlap between the representative vectors in group 1 and group 2 is made smaller (for example, the distance is increased) than the degree of overlap between the center of gravity vectors. Can do. Note that the representative vector determination method is not necessarily limited to the method of selecting the feature vector closest to the centroid of each group. Select several vectors close to the centroid of each group, and further determine the centroid as the representative vector. May be.

  By using the representative vector determination method as described above, the influence of components having a value equal to or greater than a predetermined component value threshold (indicating characteristics) in common among a plurality of groups is excluded, and components having characteristics only in groups are displayed. The ratio can be increased, the difference between the representative vectors of each group can be clarified, and a representative vector that is easier to analyze can be determined.

  On the other hand, if the average distance is equal to or greater than the predetermined distance, it is determined that the degree of overlapping of components having a value equal to or greater than the predetermined component value threshold between feature vectors in different groups is smaller than the predetermined level. A representative vector determining method (high overlapping representative vector determining method) that increases the degree of overlap between components equal to or greater than a predetermined component value threshold is selected. The representative vector determination method selected in the present embodiment creates a representative vector in which the top N frequently occurring components (N is an integer of 1 or more) are extracted and left in each group, and other components are set to 0. Method (S35).

  FIG. 6 is a diagram showing an example in which a representative vector is determined by extracting a frequent component from a group feature vector. In the example of this figure, N is 3 and a representative vector is created. Group 1 includes four feature vectors V11, V12, V13, and V14, and group 2 includes four feature vectors V21, V22, V23, and V24. The number of overlapping components among the centroid vectors of each group is four, while the number of overlapping components among the vectors generated by taking the top three frequently occurring is three, and the number of overlapping components is reduced. However, in the normalized representative vector, the value of the component that overlaps with another representative vector is large, so that the degree of overlap can be increased (for example, the distance can be reduced). The omitted component is a component that does not appear (does not overlap) in common in a plurality of groups. In the example of this figure, N is set to 3, but other numbers smaller than the dimension number of the feature vector such as 5 and 10 may be used.

  In this way, by extracting frequent components, multiple groups are excluded except for the influence of components that have a value equal to or greater than a predetermined component value threshold only in each group (indicating characteristics) and that do not exhibit characteristics in other groups. It is possible to determine a representative vector that is easier to analyze by increasing the proportion of components that exhibit features in common, that is, the proportion of overlapping components.

  In the present embodiment, the distance between the reference representative vectors of each group is used to evaluate the degree of overlap between feature vectors of different groups with components having a value equal to or greater than a predetermined component value threshold. However, other indicators may be used. For example, the determination method selection unit 22 determines whether a feature vector that satisfies the condition that the value of a certain component is equal to or greater than a predetermined component value threshold exists in both groups for the two groups that determine the degree of overlap. Check every time. As a result of the confirmation, it is possible to count the number of components that have been confirmed as having feature vectors that satisfy the condition in both groups, and use the counted value to determine the degree of overlap between the two groups.

  In the present embodiment, the determination method selection unit 22 performs two determinations of S31 and S33, but only one of the determinations may be performed. Further, when only S33 is determined, instead of selecting one of the two representative vector determination methods depending on whether the degree of overlap is greater than or less than a predetermined degree, the degree of overlap is determined. Depending on, one of more types of representative vector determination methods may be selected. For example, as the degree of overlap increases, any one of the representative vector determination methods is performed in the order of the low overlap representative vector determination method, the reference representative vector as a representative vector, and the high overlap representative vector determination method. You may make it select.

  The determination method selection unit 22 selects a representative vector determination method based on a plurality of feature vectors stored in the memory 12 in the above-described embodiment. Instead, the input unit 13 from a person who operates the information analysis apparatus. The representative vector determination method that receives the instruction may be selected.

  The representative vector determination unit 23 is realized with the CPU 11 as the center. The representative vector determination unit 23 determines a representative vector for each of the plurality of feature vectors stored in the memory 12 using the representative vector determination method selected by the determination method selection unit 22, and the determined representative vector Is stored in the memory 12. The method of determining the representative vector is as described in the description of the determination method selection unit 22.

  The coordinate information generation unit 24 converts coordinates from the representative vector of each group and the feature vector classified into each group into a coordinate system of a space for generating a map. The representative vector and the feature vector are high-dimensional vectors having keyword (feature) types of dimensions, and a lower-dimensional (two-dimensional, three-dimensional, etc.) coordinate system (hereinafter referred to as a map coordinate system) that is a map generation target. ) To project to.

  The coordinate information generation unit 24 performs a two-step process of first projecting the representative vector onto the map coordinate system and then projecting the feature vector onto the map coordinate system so as to preserve the distance from the representative vector. Thereby, not only between the representative vectors of each group but also the analysis target of each group is mapped while storing the relationship. Hereinafter, a case where the map coordinate system is a two-dimensional space will be described as an example for the two-step mapping.

  An example of the projection of the representative vector, which is the mapping at the first stage, is as follows. FIG. 7 is a diagram illustrating an example of a distribution image of feature vectors of each group existing in a multidimensional space. This figure schematically represents multidimensional data in three dimensions. The feature vectors are classified into four groups of A, B, C, and D. The region within the two-dot chain line indicates the region where the feature vector exists, and the circle indicates the coordinates of the representative vector in the multidimensional space. FIG. 8 is a diagram showing an example of an image obtained by projecting the representative vector of each group shown in FIG. 7 onto the map coordinate system. This projection can be performed by a known method such as principal coordinate analysis or principal component analysis. Hereinafter, the mapped space is referred to as a map target space.

  Next, an example of the projection of each feature vector, which is the second-stage mapping, is as follows. The coordinate information generation unit 24 performs the following mapping process for each group. First, two representative vectors different from the representative vector of the group to be subjected to the mapping process (hereinafter referred to as “target group”) are selected. The criterion for selecting the two representative vectors may be, for example, in the order of the distance from the representative vector of the target group. For example, if group A in FIG. 8 is the target group, the representative vectors of group B and group D are selected. Then, for the feature vector M classified into the target group, the representative vector of the group, and the total (M + 3) of the two selected representative vectors, the same 2 as the map coordinate system using principal coordinate analysis or the like 2 Project into the dimensional space. The information projected here is called temporary map information. Next, the coordinate information of each feature vector included in the temporary map information created for each group is subjected to affine transformation, and this is used as coordinates on the map target space corresponding to the feature vector. When each temporary map is affine transformed, for each of the three representative vectors included in each temporary map information, the coordinates after the affine transformation and the coordinates projected in the first-stage mapping are matched. Also, instead of the second-stage mapping method described so far, the distance between each feature vector and the representative vectors of all groups already mapped, which is a known method described in Non-Patent Document 1, is stored. As described above, a mapping method using a multidimensional scaling method may be used.

  The drawing unit 25 generates an image drawn so as to be recognized by the user from the coordinates in the map target space corresponding to the representative vector and the feature vector generated by the coordinate information generation unit 24, and outputs the generated image via the output unit 14. . However, instead of outputting to the output unit 14, it may be converted into a predetermined image data format such as JPEG, stored in the memory 12, and provided to the user.

  FIG. 9 is a diagram illustrating an example of an image in which the drawing unit 25 projects the feature vectors of each group onto the map coordinate system. In FIG. 9, the coordinates corresponding to the feature vector itself are not plotted, and the density at which the analysis target exists within a certain range is represented by shading.

  In the embodiments described so far, the patent document is mainly described as the analysis target to be analyzed. However, the analysis target is not limited to that as long as it has a plurality of types of features. For example, it may be an HTML document or XML document existing on the Web, audio information recorded on an electronic medium, or the like.

  In addition, the appearance frequency of the keyword (feature) appearing in the analysis target is used as it is as the component of the feature vector. For example, if the appearance frequency is 0, the value of the component may be −1, and if it is more, it may be expressed as +1. . In this case, the predetermined component value threshold may be set to -1.

It is a block diagram showing a configuration example of an information analysis apparatus according to an embodiment of the present invention. It is a figure which shows the functional block of the information analysis apparatus which concerns on embodiment of this invention. It is a figure which shows an example of the method of producing | generating a feature vector from an analysis object. It is a figure which shows an example of the processing flow of a determination method selection part. It is a figure which shows the example in which the feature vector nearest to the center of gravity of a group is determined as a representative vector. It is a figure which shows the example which extracts a frequent component from the feature vector of a group, and determines a representative vector. It is a figure which shows an example of the distribution image of the feature vector of each group which exists on multidimensional space. It is a figure which shows an example of the image which projected the representative vector of each group on the map coordinate system. It is a figure which shows an example of the image which projected the feature vector of each group on the map coordinate system.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Information analysis apparatus, 11 CPU, 12 Memory, 13 Input part, 14 Output part, 21 Feature vector acquisition part, 22 Determination method selection part, 23 Representative vector determination part, 24 Coordinate information generation part, 25 Drawing part

Claims (7)

For each of a plurality of analysis objects classified into any of a plurality of groups, an acquisition means for acquiring a feature vector having as its component a value indicating the degree of the analysis object having each of a plurality of types of features;
Of the plurality of types of representative vector determination methods, based on the evaluation on the distribution of components having a value equal to or greater than a predetermined component value threshold value of the feature vector acquired by the acquisition unit for all or a part of the plurality of analysis targets. A selection means for selecting one;
A determining unit that determines a representative vector of the group based on a feature vector acquired by the acquiring unit with respect to an analysis target classified into each group by a representative vector determining method selected by the selecting unit;
An information analysis apparatus comprising: The plurality of types of representative vector determination methods include at least one limited representative vector determination method in which an amount of information used for determining a representative vector of the group among information on feature vectors related to each group is limited;
The selection unit is configured to determine whether the feature vector acquired by the acquisition unit has a value indicating the number of components having a value equal to or greater than the predetermined component value threshold value when the number of components is greater than a predetermined component number threshold value. Select one of them,
The information analysis apparatus according to claim 1. The plurality of types of representative vector determination methods include a reference representative vector determination method for determining a centroid vector of a feature vector of each group, and a degree in which components having a value equal to or greater than a predetermined component value threshold overlap between representative vectors of different groups. Lower overlap representative vector determination method that lowers the reference representative vector determination method, and the degree of overlap of components having a value equal to or greater than a predetermined component value threshold between different groups of representative vectors than the reference representative vector determination method A high overlapping representative vector determination method for increasing,
The selection means evaluates the degree of overlapping of components having a value equal to or greater than the predetermined component value threshold between feature vectors related to different groups, and when the overlapping degree is equal to or higher than the predetermined level, the low overlap representative Selecting a vector determination method, and selecting the high overlap representative vector determination method when the overlapping degree is smaller than a predetermined level;
The information analysis apparatus according to claim 1, wherein the information analysis apparatus is an information analysis apparatus. The selection means calculates a degree of proximity between feature vectors relating to different groups, and evaluates the degree of overlap according to the degree of proximity;
The information analysis apparatus according to claim 3. The low overlap representative vector determination method is selected based on a representative vector of the group determined by the reference representative vector determination method among the feature vectors acquired by the acquisition unit for the analysis target classified into each group. A method for determining a representative vector of the group using only a part of
The information analysis apparatus according to claim 3 or 4, characterized by the above. Coordinate information generating means for generating coordinate information obtained by projecting the representative vector of each group determined by the determining means to the same coordinate system as the space in which the map is generated;
The information analysis apparatus according to claim 1, further comprising: Acquisition means for acquiring, for each of a plurality of analysis objects classified into any of a plurality of groups, a feature vector whose component is a value indicating the degree of the analysis object having a plurality of types of features;
Of the plurality of types of representative vector determination methods, based on the evaluation on the distribution of components having a value equal to or greater than a predetermined component value threshold value of the feature vector acquired by the acquisition unit for all or a part of the plurality of analysis targets. A selection means for selecting one,
A determining unit that determines a representative vector of the group based on a feature vector acquired by the acquiring unit with respect to an analysis target classified into each group by a representative vector determining method selected by the selecting unit;
As a program to make the computer function.

Images