JP6403850B1 - Information processing apparatus, information processing method, and program - Google Patents

Abstract

An object of the present invention is to generate information indicating what elements exist in an object.
A keyword indicating a feature of an object is extracted from a document data group including a plurality of document data related to the object, and the keyword indicating the feature of each element is determined by grouping the keyword for each element included in the object. Then, the relationship between the keywords indicating the characteristics of the object is determined, and feature information indicating the relationship between the characteristics of each element in the object is generated based on the determined keyword for each element and the relationship between the keywords. To do.
[Selection] Figure 2

Description

  The present invention relates to an information processing apparatus, an information processing method, and a program.

In recent years, a large amount of document data has been accumulated in a storage area of a company or the like, a storage area on the Internet, and the like and used for various purposes. The movement to use the large amount of document data for business and research is also becoming active.
As a technique for analyzing document data, there is Patent Document 1. Patent Document 1 discloses an information extraction system that analyzes a document described in a natural language, extracts information in a predetermined field, and outputs information including a relationship between words included in the document in a certain format. There is disclosed a technique for extracting more accurate information more accurately by performing syntax analysis while recognizing the relationship between keywords.

Japanese Patent Laid-Open No. 07-85071

It is desired to analyze document data relating to an object such as an organization (for example, a company, a school, a research institution, etc.) or an individual to understand what elements are included in the object. For example, it is desired to understand what kind of business is present in a company and what kind of research department exists in a research institution. Therefore, there has been a demand for generating information indicating what elements are present in the object. However, in Patent Literature 1, it is impossible to generate information indicating what elements exist in an object.
Accordingly, an object of the present invention is to generate information indicating what elements are present in an object.

  Therefore, the information processing apparatus according to the present invention provides a keyword indicating a feature of the object from the object document data group based on the appearance frequency of words in the object document data group, which is a document data group including a plurality of document data related to the object. Extraction means for extracting, the appearance frequency of the keyword extracted by the extraction means in the document data included in the object document data group, the rarity of the keyword extracted by the extraction means in the object document data group, Based on the first feature amount determining means for determining the feature amount of the document data included in the object document data group, and based on the feature amount determined by the first feature amount determining means, the object document Document data included in the data group is converted into the object Classification means for classifying into one or more sections corresponding to the included elements, and the extraction in the sections included in the one or more sections into which the document data included in the object document data group is classified by the classification means A keyword determining means for determining a keyword indicating a feature of an element corresponding to a section included in the one or more sections from the keywords extracted by the extracting means based on the appearance frequency of the keyword extracted by the means; A second feature amount determining unit that determines a feature amount of the keyword extracted by the extracting unit; and keywords extracted by the extracting unit based on the feature amount determined by the second feature amount determining unit Relevance determining means for determining the relevance of the keyword, the keyword determined by the keyword determining means, It has a relevance which is determined by the determining means, and a generation means for generating feature information indicating a relationship between features of each element in the object based on.

  According to the present invention, information indicating what elements are present in an object can be generated.

FIG. 1 is a diagram illustrating an example of a hardware configuration of the information processing apparatus. FIG. 2 is a flowchart illustrating an example of processing of the information processing apparatus. FIG. 3 is a diagram illustrating an example of the extracted keywords. FIG. 4 is a diagram for explaining an example of the feature amount of the document data. FIG. 5 is a diagram for explaining an example of the feature amount of document data. FIG. 6 is a diagram illustrating an example of a keyword for each cluster. FIG. 7 is a diagram for explaining an example of a keyword feature amount. FIG. 8 is a diagram illustrating an example of feature information. FIG. 9 is a diagram illustrating an example of feature information. FIG. 10 is a diagram illustrating an example of feature information. FIG. 11 is a diagram illustrating an example of an image indicating a keyword structured for each element.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
<Embodiment 1>
(Outline of processing)
For example, when an organization such as a company promotes business, it captures characteristics such as group information about a group, product / service information of the group, business structure and business structure between groups, and It is a feature of the business, and by grasping how each feature is related to each other, it is possible to more easily grasp the details of the organization, to grasp the details of the organization's business trends, to clarify the type of business and business type, Business matching is possible.
Therefore, in this embodiment, the information processing apparatus 100 extracts a keyword indicating the characteristics of a group from a document data group including a plurality of document data related to the group that is an object, and the keyword for each business that is an element included in the group. To determine the keywords that indicate the characteristics of each business, determine the relevance between the keywords that indicate the characteristics of the object, and based on the determined keywords for each business and the relationship between the keywords, A process for generating feature information indicating the relevance of features for each business in the group will be described.

(Hardware configuration of information processing device)
FIG. 1 shows a hardware configuration of the information processing apparatus 100. The information processing device 100 is an information processing device such as a personal computer, a server device, or a tablet device.
The information processing apparatus 100 includes a CPU 101, a main storage device 102, an auxiliary storage device 103, and a network I / F 104. Each element is connected to be communicable with each other via a system bus 105.
The CPU 101 is a central processing unit that controls the information processing apparatus 100. The main storage device 102 is a storage device such as a random access memory (RAM) that functions as a work area for the CPU 101 and a temporary storage location for data. The auxiliary storage device 103 is a storage device that stores various programs, various setting information, various document data, and the like. The auxiliary storage device 103 is mounted using a storage medium such as a read only memory (ROM), a hard disk drive (HDD), or a solid state drive (SSD). The network I / F 104 is an interface used for communication with an external device via a network such as the Internet or a LAN.
The CPU 101 executes processing based on a program stored in the auxiliary storage device 103, so that the functions of the information processing device 100, the processing of the flowchart described later in FIG. 2, (image output processing), and (search processing) are performed. The process etc. which are demonstrated by are implement | achieved.

(Feature information generation process)
FIG. 2 is a flowchart illustrating an example of processing of the information processing apparatus 100.
In step S <b> 201, the CPU 101 acquires a document data group including a plurality of document data related to a group for each of a plurality of predetermined groups. Hereinafter, a plurality of predetermined groups are referred to as a group of groups. In the following, a document data group related to a group is referred to as a group document data group. The group document data group is an example of an object document data group that is a document data group related to an object. For example, the CPU 101 acquires a group document data group relating to each group included in the group group stored in advance in the auxiliary storage device 103. Further, the CPU 101 may acquire a group document data group relating to each group included in the group group from an external storage device via the network I / F 104 and the Internet. In addition, the CPU 101 may acquire a group document data group regarding each group included in the group by searching for document data regarding each group on the Internet via a search engine.
Further, the CPU 101 may acquire a group document data group related to one group.
In S202, the CPU 101 extracts words from the plurality of group document data groups acquired in S201. For example, the CPU 101 extracts words by performing morphological analysis on each document data included in the group document data group.

In step S <b> 203, the CPU 101 determines, for each group included in the group, a score value indicating the importance of each word in the group document data group in the group. For example, the CPU 101 determines the score value of the word (1) included in the group document data group related to the group (1) included in the group group as follows.
First, the CPU 101 determines the appearance frequency of the word (1) in the group document data group related to the group (1) and the rarity of the word (1) in the plurality of group document data groups related to each group included in the group group. To do. The word appearance frequency in the document data group is an index indicating the degree of appearance of the word in the document data group. The word rarity in a plurality of document data groups is an index indicating the degree of rarity of the words in the plurality of document data groups.

For example, when the CPU 101 extracts a total of A words from the group document data group related to the group (1) in S202 and B words (1) are extracted, the word (in the group document data group related to the group (1) ( The appearance frequency of 1) is determined as B / A. As described above, the ratio between the total number of words included in the group document data group related to the group (1) and the number of words (1) included in the group document data group is the word (1) in the group document data group. ) Is an example of the appearance frequency.
Further, for example, the CPU 101 has C group document data groups including document data including the word (1) in a plurality of group document data groups related to the group groups, and the group document data groups related to the group groups are grouped in the group document data groups. When there are D document data groups, the rarity of the word (1) in the plurality of group document data groups for each group included in the group is determined as Log (D / C). Further, the CPU 101 may determine the rarity of the word (1) in the plurality of group document data groups regarding each group included in the group as D / C. Thus, among the plurality of group document data groups related to each group group, the number of group document data groups including the word (1) and the number of group document data groups included in the plurality of group document data groups, A ratio, a logarithm based on an arbitrary positive number of this ratio, and the like are examples of the rarity of the word (1) in the plurality of group document data groups.

CPU 101 determines a score value of word (1) in group (1) based on the determined appearance frequency and rarity. For example, the CPU 101 determines the score value of the word (1) using a TF-IDF (Term Frequency-Inverse Document Frequency) method based on the determined appearance frequency and rarity. The higher the appearance frequency determined, the higher the score value, assuming that the CPU 101 is an important word that is frequently used in the group document data group related to the group (1).
Further, the higher the determined rarity, the higher the score value, assuming that the CPU 101 is an important word characteristic for the organization (1). Further, the lower the determined rarity, the lower the score value, assuming that the CPU 101 is an unimportant word that is not characteristic for each group in the group. In this way, the CPU 101 can determine a score value that indicates more precisely how important each word is in the group by using the determined rarity in consideration of the relationship with other groups. Thereby, CPU101 can extract the keyword which shows the characteristic of an organization in S204, considering the relation with other organizations (for example, the difference in the feature with other organizations, etc.).
The CPU 101 performs the above processing on all the words included in the group document data group related to the group (1), and determines the score value of each word in the group document data group related to the group (1).
In addition, the CPU 101 performs the same processing for other organizations other than the organization (1) included in the organization group, and calculates the score value of each word in each of the plurality of organization document data groups regarding each organization included in the organization group. decide.

In the present embodiment, the CPU 101 uses the plurality of group document data groups related to each of the plurality of groups included in the group for each group included in the group in S203, and uses each word included in the group document data group related to the group. The score value was determined. However, for example, when the CPU 101 acquires a group document data group related to one group in S201, the CPU 101 uses one group document data group related to the one group, and uses the group document data group related to the group. A score value may be determined.
In that case, the CPU 101 may determine the score value of each word included in the group document data group regarding the group as follows.
That is, the CPU 101 may determine the appearance frequency of each word in the group document data group and determine the score value of each word based on the determined appearance frequency. For example, the CPU 101 may determine the value of the appearance frequency of each word as the score value of each word.

In S <b> 204, the CPU 101 extracts a keyword indicating the characteristics of the organization for each organization included in the organization group. CPU101 extracts the keyword which shows the characteristic of organization (1) as follows, for example. That is, the CPU 101 indicates a feature of the organization (1) by selecting a word whose score value determined in S202 is equal to or greater than a predetermined threshold among the words extracted in S201 from the organization document data group related to the organization (1). Extract as keywords.
In addition, the CPU 101 performs the same processing for other organizations other than the organization (1), and extracts keywords for each organization included in the organization group. The process of S204 is an example of an extraction process for extracting a keyword.
FIG. 3 is a diagram illustrating an example of the keyword of the organization (1) extracted in S204. In the example of FIG. 3, the keywords of the organization (1) are “parts”, “clean”, “bed”, “engine”, “brake”, “furniture”, “energy”, “air conditioner”, “bedding”, It shows how ten words “transmission” have been determined.
In S205, the CPU 101 selects one organization from the organization group. In the following, the organization selected in S205 is set as the selected organization.

In S206, the CPU 101 determines, for each document data included in the group document data group related to the selected group, a score value indicating the importance of the selected group keyword document data extracted in S204. For example, the CPU 101 determines the score value of the keyword (1) of the selected group extracted in S204 in the document data (1) included in the group document data group regarding the selected group as follows. The process of S206 is an example of a first score determination process.
First, the CPU 101 determines the appearance frequency of the keyword (1) in the document data (1) and the rarity of the keyword (1) in the group document data group related to the selected group. The appearance frequency of a keyword in document data is an index indicating the degree of appearance of the keyword in the document data. The rarity of the keyword in the group document data group is an index indicating the degree of rarity of the keyword in a plurality of document data included in the group document data group.

For example, when the document data (1) includes E words in total and F keywords (1) are extracted in S204, the CPU 101 determines the appearance frequency of the keyword (1) in the document data (1) as F / E is determined. As described above, the ratio between the total number of words included in the document data (1) and the number of keywords (1) included in the document data (1) is the appearance frequency of the keyword (1) in the document data (1). It is an example. Further, for example, when the document data (1) includes E ′ keywords extracted as keywords of the selected organization in S204, and F keywords (1) are extracted in S204, the document data (1) The appearance frequency of the keyword (1) may be determined as F / E ′.
Further, for example, the CPU 101 has G document data including the keyword (1) in all the document data included in the group document data group related to the selected group, and all the document data is included in the group document data group related to the selected group. When there are such items, the rarity of the keyword (1) in the group document data group regarding the selected group is determined as Log (H / G). Further, the CPU 101 may determine the rarity of the keyword (1) in the group document data group regarding the selected group as H / G. Thus, the ratio between the number of document data including the keyword (1) and the number of document data included in the group document data group among all the document data included in the group document data group related to the selected group, A logarithm with an arbitrary positive number in this ratio as a base is an example of the rarity of the keyword (1) in the group document data group.

The CPU 101 determines the score value of the keyword (1) in the document data (1) based on the determined appearance frequency and rarity. For example, based on the determined appearance frequency and rarity, the CPU 101 determines the score value of the keyword (1) in the document data (1) using the TF-IDF method. As the determined appearance frequency is higher, the CPU 101 increases the score value as an important word that is used more frequently in the document data (1).
Further, the higher the rarity determined, the higher the score value, assuming that the CPU 101 is an important word characteristic for the document data (1) included in the group document data group related to the selected group. Further, the lower the determined rarity, the lower the score value, assuming that the CPU 101 is an unimportant word that is not characteristic for each document data included in the group document data group related to the selected group. In this way, the CPU 101 uses the determined rarity to determine which keyword is included in the document data in consideration of the relationship with other document data (for example, differences in features from other document data). A score value can be determined that more accurately indicates whether the degree is important.

The CPU 101 performs the above processing on all the keywords of the selected group extracted in S204 included in the document data (1), and determines the score value of each keyword in the document data (1). When the keyword (1) is not included in the document data (1), the CPU 101 may determine the score value of the keyword (1) in the document data (1) by the same process with the appearance frequency being 0, The score value of the keyword (1) in the document data (1) may be determined to a predetermined value (for example, 0).
Further, the CPU 101 performs the same processing for other document data other than the document data (1) included in the group document data group related to the selected group, and each document data included in the group document data group related to the selected group. Determine the keyword score value.

In S207, the CPU 101 determines the feature amount of each document data included in the group document data group related to the selected group based on the score value of each keyword determined in S206 for each document data included in the group document data group related to the selected group. decide. For example, the CPU 101 determines the feature amount of the document data included in the group document data group regarding the selected group as follows. That is, the CPU 101 determines a vector having each component of each keyword score value in the document data as a feature amount of the document data. The process of S207 is an example of a first feature amount determination process.
FIG. 4 is a diagram for explaining an example of the feature amount of the document data. A table 400 in FIG. 4 is a table showing the score value of each keyword in each document data included in the group document data group regarding the selected group. In the example of FIG. 4, the document data included in the group document data group regarding the selected group is n pieces of document data of document data (1) to document data (n). The keywords extracted as keywords of the selected organization in S204 are m keywords from keyword (1) to keyword (m). A column corresponding to the document data in the table 400 indicates the score value of each keyword in the document data. That is, the CPU 101 determines a vector having each score value included in a column corresponding to certain document data as a feature amount of the document data.

Further, the CPU 101 may determine, as a feature amount of the document data, a vector obtained by dimension-compressing a vector having each keyword score value of each keyword in the document data as a component.
For example, the CPU 101 generates a vector having each score value included in a column corresponding to the document data in the table 400 as an element for each document data included in the group document data group regarding the selected group. Then, the CPU 101 dimensionally compresses each generated vector using a method such as principal component analysis, singular value decomposition, and auto encoder, and determines each dimensionally compressed vector as a feature amount of each document data.
FIG. 5 is a diagram for explaining an example of the feature amount of document data. A table 500 in FIG. 5 is a table showing the value of each element (base) of the feature amount in each document data included in the group document data group regarding the selected group. In the example of FIG. 5, as in FIG. 4, the document data included in the group document data group regarding the selected group is n pieces of document data (1) to document data (n). A column corresponding to the document data in the table 500 indicates the value of each element of the vector of feature amounts of the document data. That is, a column corresponding to certain document data in the table 500 indicates a vector of feature amounts of the document data. The vector indicated by the column corresponding to the document data in the table 500 is a vector in which the vector indicated by the column corresponding to the document data in the table 400 of FIG. 4 is dimensionally compressed. As described above, the CPU 101 can reduce the burden of the document data clustering process in S208 by determining the dimension-compressed vector as the feature amount of the document data.

In S208, the CPU 101 clusters each document data included in the group document data group related to the selected group based on the feature amount determined in S207 for each document data included in the group document data group related to the selected group. For example, the CPU 101 clusters each piece of document data included in the group document data group related to the selected group using a clustering technique such as the k-means method, the shortest distance method, or the Ward method. The CPU 101 can cluster document data having similar features into one cluster by clustering a plurality of document data included in the group document data group based on the feature amount of the document data. A plurality of similar document data included in the group document data group can be assumed to be document data indicating characteristics of some element included in the group. Therefore, the CPU 101 clusters a plurality of document data included in the group document data group related to the selected group, so that each of the plurality of document data corresponds to one or more classifications (clusters) corresponding to elements included in the selected group. Can be classified.
In S209, the CPU 101 determines a cluster indicating a business in the selected group from each cluster clustered in S208 based on a predetermined criterion. For example, it can be assumed that the document data related to the business included in the group has a certain number or more when the business activity is performed. Therefore, it can be assumed that there are more than a predetermined number of document data relating to the business. Therefore, for example, when the number of document data included in the cluster is equal to or greater than a predetermined threshold (for example, 10), the CPU 101 determines the cluster as a cluster indicating a business. Hereinafter, a cluster determined as a cluster indicating a business is referred to as a business cluster.

In S210, the CPU 101 determines, for each cluster determined as a business cluster in S209, a score value indicating the importance of the selected organization keyword extracted in S204 in each business cluster. For example, the CPU 101 determines the score value of the keyword (1) of the selected organization extracted in S204 in the cluster (1) that is the cluster determined as the business cluster in S209 as follows. The process of S210 is an example of a second score determination process.
First, the CPU 101 determines the appearance frequency of the keyword (1) in the cluster (1) and the rarity of the keyword (1) in all the clusters determined as the business cluster. The appearance frequency of a keyword in a cluster is an index indicating the degree of appearance of the keyword in document data included in the cluster. The rarity of keywords in all clusters determined as business clusters is an index indicating the degree of rarity of the keywords in all clusters.

The CPU 101, for example, includes all I words in all document data included in cluster (1), and J keywords (1) are extracted from all document data included in cluster (1) in S204. In this case, the appearance frequency of the keyword (1) in the cluster (1) is determined as J / I. Thus, the ratio between the total number of words included in all document data included in cluster (1) and the number of keywords (1) included in all document data included in cluster (1) is the cluster. It is an example of the appearance frequency of the keyword (1) in (1). Further, the CPU 101 includes, for example, all I ′ keywords extracted as keywords of the selected organization in S204 in all document data included in the cluster (1), and all of the keywords included in the cluster (1) in S204. When J keywords (1) are extracted from the document data, the appearance frequency of the keyword (1) in the cluster (1) may be determined as J / I ′.
For example, the CPU 101 includes K clusters including document data including the keyword (1) in the clusters included in all clusters determined in S209, and the clusters determined as the business clusters in S209. When there are L in total, the rarity of the keyword (1) in all clusters determined as business clusters in S209 is determined as Log (L / K). Further, the CPU 101 may determine the rarity of the keyword (1) in all clusters determined as the business cluster in S209 as L / K. As described above, the ratio between the number of clusters including the document data including the keyword (1) included in all the clusters determined as the business cluster in S209 and the number of clusters included in all the clusters, A logarithm based on an arbitrary positive number of the ratio is an example of the rarity of the keyword (1) in all the clusters.

The CPU 101 determines the score value of the keyword (1) in the cluster (1) based on the determined appearance frequency and rarity. For example, the CPU 101 determines the score value of the keyword (1) in the cluster (1) using the TF-IDF method based on the determined appearance frequency and rarity. As the determined appearance frequency is higher, the CPU 101 increases the score value as an important word that is used more frequently in the cluster (1).
Further, the higher the rarity determined, the higher the score value, assuming that the CPU 101 is a characteristic important word in the cluster (1) included in the group document data group related to the selected group. Further, the lower the determined rarity, the lower the score value, assuming that the CPU 101 is an unimportant word that is not characteristic for each cluster included in the group document data group related to the selected group. In this way, the CPU 101 uses the determined rarity to determine how important each keyword is in the cluster in consideration of the relationship with other clusters (for example, differences in features from other clusters). It is possible to determine a score value that indicates whether or not.

The CPU 101 performs the above processing on all the keywords of the selected group extracted in S204 included in the cluster (1), and determines the score value of each keyword in the cluster (1). If the keyword (1) is not included in any of the document data included in the cluster (1), the CPU 101 sets the appearance frequency to 0 and determines the score value of the keyword (1) in the cluster (1) by the same processing. Alternatively, the score value of the keyword (1) in the cluster (1) may be determined to a predetermined value (for example, 0).
Further, the CPU 101 performs the same processing for other clusters other than the cluster (1) determined as the business cluster in S209, and determines the score value of each keyword in each of the clusters determined as the business cluster.

If there is only one cluster (for example, if there is one cluster determined as a business cluster in S209, or if the number of clusters clustered in S208 is one), the CPU 101 determines in S210. You may determine the score value of the keyword in a cluster as follows.
That is, the CPU 101 may determine the appearance frequency of each keyword in the cluster and determine the score value of each keyword based on the determined appearance frequency. For example, the CPU 101 may determine the appearance frequency value of each keyword as the score value of each keyword.

In S <b> 211, the CPU 101 determines a keyword indicating a cluster characteristic for each cluster determined as a business cluster in S <b> 209. For example, the CPU 101 determines a keyword indicating the characteristics of the cluster (1) as follows. That is, among the keywords extracted for the selected group in S204, the CPU 101 selects a keyword having a score value determined for the cluster (1) in S210 that is equal to or greater than a predetermined threshold, and a keyword indicating the characteristics of the cluster (1) Determine as.
The CPU 101 performs the same processing for other clusters determined as business clusters in S209 other than the cluster (1), and determines a keyword for each cluster. The process of S211 is an example of a keyword determination process.
FIG. 6 is a diagram illustrating an example of keywords determined in S211 for each cluster determined as a business cluster in S209. The example of FIG. 6 shows an example of the keyword determined in S211 for each cluster determined as a business cluster in S209 when the selected group is the group (1). Each of the three frames in FIG. 6 indicates a cluster determined as a business cluster in S209. The word in each frame is a keyword indicating the characteristics of the cluster corresponding to each frame. In the example of FIG. 6, a cluster having three keywords “bed”, “bedding”, and “furniture”, a cluster having three keywords “air conditioner”, “clean”, and “energy”, “parts”, It is shown that a cluster having four keywords “brake”, “transmission”, and “engine” is included in the group (1). In this way, the CPU 101 can group the keywords extracted in S204 for each business by the processing in S211.

In S212, the CPU 101 determines the feature amount of each keyword in the selected group extracted in S204. For example, the CPU 101 determines the keyword feature amount based on the score value of each keyword determined for each cluster in S210 as follows. That is, for the keyword, the CPU 101 determines, as the feature amount of the keyword, a vector having each keyword score value determined in S210 for each cluster. This feature amount can be regarded as a feature amount indicating how important the corresponding keyword is for each cluster. The process of S212 is an example of a second feature amount determination process.
FIG. 7 is a diagram for explaining an example of a keyword feature amount. A table 700 in FIG. 7 is a table showing the score value of each keyword in each cluster determined as a business cluster in S209. In the example of FIG. 7, the clusters determined as the business clusters are k clusters from cluster (1) to cluster (k). In the example of FIG. 7, the keywords extracted as the keywords of the selected organization in S204 are m keywords from keyword (1) to keyword (m). A column corresponding to a certain cluster in the table 700 indicates a score value of each keyword in the cluster. The row corresponding to the keyword in the table 700 indicates the score value of the keyword in each cluster. The CPU 101 determines, as the feature amount of each keyword, a vector having each score value indicated by the row corresponding to each keyword in the table 700 as an element, as the feature amount of the keyword. Further, the CPU 101 may determine, as the feature amount of the keyword, a vector obtained by dimensionally compressing this vector using a method such as principal component analysis, singular value decomposition, or auto encoder.
For example, the CPU 101 may determine the keyword feature amount based on the score value of each keyword determined for each document data in S206 as follows. That is, for the keyword, the CPU 101 may determine, as the feature amount of the keyword, a vector having each keyword score value determined in S206 for each document data as each component. This feature amount can be regarded as a feature amount indicating how important the corresponding keyword is for each document data. Further, the CPU 101 may determine, as the feature amount of each keyword, a vector obtained by dimensionally compressing this vector using a method such as principal component analysis, singular value decomposition, or auto encoder.

In S213, the CPU 101 determines, as the relevance between the keywords in the selected organization, the relevance indicating the degree of relevance between the keywords determined for each cluster in S213 based on the feature amount determined in S212. The relevance between keywords may be different for different organizations. For example, assume that there are Company A and Company B that manufacture and sell automobiles. Company A preferentially manufactures and sells high-end cars, and Company B preferentially manufactures and sells inexpensive mini cars. In this case, the relevance between the keyword “automobile” and the keyword “luxury” in the company A is greater than the relevance between the keyword “automobile” and the keyword “luxury” in the company B. The CPU 101 determines the relevance between the keywords in the selected organization in S213 in consideration of the difference in relevance between the keywords in each organization. The process of S213 is an example of an association determination process.
For related keywords, the higher the score value in one cluster (or document data) of one keyword in a cluster (or document data), the higher the score in that cluster (or document data) of the other keyword It can be assumed that the value will also be high. That is, it can be assumed that the feature amounts determined in S212 between the related keywords are such that the greater the value of an element of a certain dimension in one feature quantity, the greater the value of the element in that dimension of the other feature quantity. Therefore, it can be assumed that the inner product between the feature quantities of the keywords increases as the relevance between the keywords increases. Therefore, for example, the CPU 101 calculates the inner product of the vector that is the feature amount of the keyword (1) determined in S212 and the vector that is the feature amount of the keyword (2) determined in S212 as the keyword (1). The degree of association with the keyword (2) is determined.

In S214, the CPU 101 determines the importance level indicating the importance of the selected group for each keyword determined for each cluster in S211. For example, based on the score value determined in S203, the CPU 101 determines the importance indicating the importance in the selected group for each keyword determined for each cluster in S211. For example, the CPU 101 determines, as the importance level, the score value determined in S203 for each keyword determined for each cluster in S211. The process of S214 is an example of an importance determination process.
For each cluster determined as a business cluster in S209, the CPU 101 determines a nuclear keyword that is a keyword representing the cluster based on the importance of the keyword determined for each cluster. For example, the CPU 101 determines a keyword having the highest importance among keywords determined for a certain cluster as a core keyword of the cluster.
Further, for example, based on the score value determined for each cluster in S210, the CPU 101 may determine the importance indicating the importance in the selected group for each keyword determined for each cluster in S211. For example, the CPU 101 may determine the score value determined in S211 as the importance level for each keyword determined for each cluster in S211.
In the present embodiment, the CPU 101 performs the processes of S210 to S214 using the cluster determined as the cluster corresponding to the business in S209. As a result, the CPU 101 can reduce the processing load for determining a keyword indicating a feature for a cluster other than the business. However, the CPU 101 may perform the processes of S210 to S214 using all the clusters clustered in S208. In that case, the CPU 101 may not perform the process of S209.

In S215, the CPU 101 determines whether or not the processes in S205 to S214 have been executed for all groups included in the group of groups. If the CPU 101 determines that the processes in S205 to S214 have been executed for all the groups included in the group, the process proceeds to S216. If the CPU 101 determines that there is an organization that does not execute the processing of S205 to S214 among the organizations included in the organization group, the CPU 101 proceeds to the processing of S205.
In the present embodiment, the CPU 101 executes the processes of S205 to S214 for all groups included in the group of groups. However, the CPU 101 may execute the processes of S205 to S214 only for a predetermined group among the groups included in the group of groups. In that case, the CPU 101 determines in S215 whether or not the processes in S205 to S214 have been executed for all of the predetermined groups. If it is determined that the processes have been executed, the CPU 101 proceeds to the process in S216 and does not execute them. If it is determined that there is a group, the process proceeds to S205.
Further, the CPU 101 may accept designation of an organization to be processed in S205 to S214 from an external information processing apparatus or the like. In that case, CPU101 will perform the process of S205-S214 about only the designated group among the groups included in the group of groups. In that case, the CPU 101 determines in S215 whether or not the processes of S205 to S214 have been executed for all the specified organizations. If it is determined that the processes have been executed, the CPU 101 proceeds to the process of S216, If it is determined that there is, the process proceeds to S205.
In S216, the CPU 101 determines the meaning of each keyword extracted in S204 for all groups included in the group. For example, the CPU 101 pays attention to the co-occurrence of a word and determines the potential meaning of the keyword using a method such as word2vec or a topic model that is a method for determining the meaning of the word. Co-occurrence is a property that indicates that certain words and other words tend to appear at the same time in a document or sentence. The potential meaning that the keyword has is the meaning that the keyword potentially has, and is an example of the meaning of the keyword. In the present embodiment, the CPU 101 acquires a vector corresponding to each keyword in a semantic space expressing the potential meaning of the word as information indicating the potential meaning of each keyword.
In addition, the CPU 101 may determine the meaning of each keyword using a dictionary that stores the meanings of a plurality of words. The process of S216 is an example of a semantic determination process.

In S217, the CPU 101 includes the keywords for each cluster determined in S211 for each group included in the group, the relevance between the keywords determined in S213 for each group included in the group, and the group included in the group. A feature indicating the relationship between characteristics of each business in each organization included in the organization group based on the importance of the keyword for each cluster determined in S214 for each organization and the meaning of each keyword determined in S216 Information is generated, and the generated feature information is stored in the auxiliary storage device 103 or the like.
The CPU 101 determines, for example, the keywords for each cluster determined in S211 for each group included in the group, the degree of association between the keywords determined in S213 for each group included in the group, and the groups included in the group. Information indicating the importance of the keyword for each cluster determined in S214 and the meaning of each keyword determined in S216 is generated as feature information. For example, the CPU 101 generates table information as illustrated in FIGS. 8 to 10 as feature information.

8 to 10 are diagrams illustrating an example of feature information.
A table 800 in FIG. 8 is a keyword management table for managing information on the keywords for each cluster and the importance of the keywords for each cluster for the organization A included in the organization group. The table 800 includes items of “keyword”, “group keyword score”, “business category”, and “business keyword score”. The item “keyword” indicates each keyword extracted in S204 for the organization A.
The item “group keyword score” indicates the score value determined in S203 for the corresponding keyword. “Business category” indicates information for identifying the category determined as the business category in S209. The item “business keyword score” indicates the score value determined in S210 for the corresponding keyword in the corresponding business.

Each “keyword” corresponding to the same “business category” is a keyword indicating the characteristics of the business indicated by the corresponding “business category”. In the example of FIG. 8, the keywords “word1” and “word2” indicate the characteristics of the same business indicated by “category1”.
Further, like the keyword “word15” in the table 800, the same keyword may be determined as a keyword indicating the characteristics of each of a plurality of businesses.
The CPU 101 determines the correspondence between the keyword and the business for each group included in the group based on the keywords for each cluster determined in S211 for each group included in the group, and the “keyword” in the table 800 is determined. And the value of the item “business category”. The CPU 101 generates a keyword management table similar to the table 800 for each group included in the group and stores it in the auxiliary storage device 103.

Either “group keyword score” or “business keyword score” indicates the importance of each keyword in the business determined in S214. When the score value determined in S203 is determined as the importance level of each keyword in S214, the item “group keyword score” indicates the importance level of the corresponding keyword in the corresponding business. When the score value determined in S210 is determined as the importance level of each keyword in S214, the item “business keyword score” indicates the importance level of the corresponding keyword in the corresponding business.
The CPU 101 determines the value of the item “group keyword score” based on the score value determined in S203, and determines the value of the item “business keyword score” based on the score value determined in S210. .
Further, the CPU 101 may not include items other than the items indicating the importance of each keyword in the business in the “group keyword score” and “business keyword score” in the table 800.

A table 900 in FIG. 9 is a relationship management table for managing information indicating the relationship between keywords determined in S213 for the organization A included in the organization group.
In the table 900, information on the feature amount determined for each keyword in S212 for the organization A is stored. As described in S213, the inner product of these feature amounts indicates the degree of association between corresponding keywords. Therefore, the CPU 101 can determine the degree of association indicating the degree of association between keywords by obtaining the inner product of vectors indicated by columns corresponding to each keyword in the table 900. Therefore, in the present embodiment, the CPU 101 generates a table 900 that stores information on the feature amount determined in S <b> 212 as information indicating the relevance between keywords and stores it in the auxiliary storage device 103. The CPU 101 generates an association management table for each organization included in the organization group in the same manner as the table 900 and stores it in the auxiliary storage device 103.
Further, the CPU 101 calculates the inner product of the feature amounts determined in S212 between the keywords, generates the calculated degree of association information as information indicating the relationship between the keywords, and stores the information in the auxiliary storage device 103. Also good.
In addition, the CPU 101 calculates the inner product of the feature amounts determined in S212 between the keywords, and when the calculated relevance is equal to or greater than a predetermined threshold, the relevance calculated is determined to be related to the keywords. Is less than a predetermined threshold value, it is assumed that the keywords are not related to each other, and information indicating the presence / absence of the relationship between the keywords is generated as information indicating the relationship between the keywords and stored in the auxiliary storage device 103. May be.

A table 1000 in FIG. 10 is a semantic management table that manages information on the potential meaning of each keyword extracted in S204 for all groups included in the group.
The column corresponding to each keyword in the table 1000 indicates the value of each base of a vector indicating the potential meaning of the corresponding keyword. The CPU 101 determines the value of each element of the table 1000 based on the vector indicating the potential meaning extracted for each keyword in S216.
As the potential meanings of the keywords are similar, the vectors indicating the potential meanings of the keywords are also similar. For this reason, the inner product of the vectors indicating the potential meaning of the keyword is an index indicating the degree of similarity of the potential meaning between the keywords. Therefore, the CPU 101 can determine the inner product of the vectors indicated by the columns corresponding to the keywords in the table 1000 as the similarity indicating the degree of similarity between the keywords.

Further, the CPU 101 determines the group of groups based on the keyword for each cluster determined in S211 for each group included in the group of groups and the degree of association between the keywords determined in S213 for each group included in the group of groups. Feature information indicating a relationship between features for each business in each of the organizations included may be generated.
In that case, the CPU 101 generates, as a keyword management table, a table including the items “keyword” and “business category” among the items of the table 800 described with reference to FIG. 8 for each group included in the group of groups. Further, the CPU 101 generates the relationship management table described with reference to FIG. 9 for each group included in the group.
Then, the CPU 101 stores the generated keyword management table and relevance management table in the auxiliary storage device 103 as feature information.

(Image output processing)
The CPU 101 can generate and output an image in which keywords are structured for each business of an organization based on the feature information generated by the processing of FIG.
Assume that the CPU 101 receives a request for an image in which a keyword is structured for each business for each organization included in the organization group from an external information processing apparatus via the network I / F 104, for example.
In that case, the CPU 101 acquires the feature information generated by the processing of FIG. 2 from the auxiliary storage device 103, and based on the acquired feature information, the keywords are structured for each business for each group included in the group. An image is generated and output by transmitting the generated image to the request source. The requesting information processing apparatus displays the received image on the display unit, thereby presenting the user with an image in which keywords are structured for each business for each group included in the group.

FIG. 11 is a diagram illustrating an example of an image generated based on the feature information, in which keywords are structured for each business for a certain organization. With reference to FIG. 11, processing for generating an image in which keywords are structured for each business for the group (1) described in FIGS. 3 and 6 will be described.
CPU101 acquires the keyword management table which stores the information of the keyword for every cluster, and the importance of the keyword for every cluster about organization (1) from characteristic information. And CPU101 determines the business contained in organization (1) from the item of "business category" of the acquired table, for example. The CPU 101 determines a keyword indicating the characteristics of the business for each business included in the group (1) from the item of “keyword” corresponding to each determined business.
As shown in FIG. 6, the organization (1) has a business corresponding to the keywords “bed”, “bedding”, “furniture”, a business corresponding to the keywords “clean”, “air conditioner”, “energy”, Businesses corresponding to the keywords “parts”, “brake”, “transmission”, and “engine” are included. Therefore, the CPU 101 determines three businesses, and keywords corresponding to each of the determined three businesses are “bed”, “bedding”, “furniture”, “clean”, “air conditioner”, “energy”, “Parts”, “Brake”, “Transmission” and “Engine” are determined. In addition, the CPU 101 determines “furniture”, “energy”, and “parts” as core keywords representing clusters corresponding to each business.

The CPU 101 determines the importance of each keyword from the item “group keyword score” or “business keyword score” in the acquired table.
In addition, the CPU 101 obtains an association management table indicating the association between keywords for the group (1) from the feature information. And CPU101 determines the feature-value of each keyword from the acquired table, calculates | requires the inner product of the determined feature-value, and determines the relevance degree of each keyword. When the determined degree of association is equal to or greater than a predetermined threshold, the CPU 101 determines that two keywords corresponding to the degree of association are related. Further, when the determined degree of association is less than a predetermined threshold, the CPU 101 determines that the two keywords corresponding to the degree of association are not related.
Further, the CPU 101 acquires the table 1000 described with reference to FIG. 10 from the feature information. And CPU101 acquires the vector which shows the potential meaning of the keyword of each business of organization (1) from the table 1000. FIG.

Based on the keywords for each business determined for the group (1), whether or not each keyword is related, the importance of each keyword, and the potential meaning of each keyword, the CPU 101 creates an image in which the keyword is structured for each business. Generate.
The CPU 101 arranges an elliptical block indicating each keyword (hereinafter referred to as a keyword block) in the image so that the corresponding business can be visually recognized. For example, the CPU 101 arranges keyword blocks corresponding to the same business so that the distance between them is equal to or less than a predetermined value, and sets the keyword blocks corresponding to different businesses to a distance equal to or greater than a predetermined value. By arranging so that each of the keyword blocks can be visually recognized. In addition, for example, the CPU 101 may describe a frame corresponding to each business in the image and arrange a keyword block corresponding to the business indicated by the frame in each frame. In the example of FIG. 11, a set of “bed”, “bedding”, “furniture”, a set of “clean”, “air conditioner”, “energy”, “parts”, “brake”, “transmission”, “engine” The group of "" is shown densely.

Further, the CPU 101 determines the size of each keyword block based on the importance of each keyword. For example, the CPU 101 determines the size of each keyword block so as to increase as the importance of the corresponding keyword increases. As a result, the CPU 101 can more easily visually recognize which keyword is important. In the example of FIG. 11, since the importance of “component” is the highest, the keyword block of “component” has the largest size.
Moreover, CPU101 describes the line which connects the keyword blocks corresponding to the keyword which has a relationship based on the presence or absence of the relationship between each keyword. In the example of FIG. 11, a state in which lines connecting related keyword blocks are described. Thereby, CPU101 can structure the keyword blocks corresponding to the keyword which is related, and can grasp | ascertain by the visual recognition of the characteristic of each business more easily.
In the example of FIG. 11, the CPU 101 describes a solid line as a line connecting keyword blocks corresponding to related keywords belonging to the same cluster. In addition, the CPU 101 describes a line different from a solid line such as a broken line or a double line as a line connecting keyword blocks corresponding to core keywords belonging to different clusters. In this way, the CPU 101 connects and structures the keyword blocks corresponding to the nuclear keywords belonging to each cluster by using different lines from the lines connecting the keyword blocks corresponding to the keywords belonging to the same cluster. A structure representing the characteristics of the organization can be displayed. FIG. 11 shows a state in which the keyword blocks in the cluster are connected by a solid line, and “energy”, “furniture”, and “parts” that are the core keywords of each cluster are connected by a broken line. . Moreover, CPU101 is good also as describing the line which connects the keyword blocks corresponding to a related keyword only about the keyword block corresponding to the same business. In that case, the lines connecting the “energy”, “furniture”, and “component” in the image of FIG. 11 are not described.
Further, the CPU 101 determines the display mode (color, pattern, etc.) of the background of each keyword block according to the potential meaning of each keyword. The potential meaning of a keyword is determined regardless of which organization the keyword shows and which business features it shows. For this reason, the background of the keyword block of the keyword having a similar potential meaning becomes a similar display mode regardless of whether it is a group or a business. Therefore, the CPU 101 can more easily grasp the similar characteristics between different organizations and different businesses by visual recognition.

Further, the CPU 101 may generate an image in which the keyword is structured for each business based on the keyword for each business determined for the group (1) and the presence / absence of association between the keywords.
Further, when the CPU 101 receives a request for an image in which a keyword is structured for each business for an organization (one or more) included in the organization group from an external information processing apparatus, ) For groups only, an image in which keywords are structured for each business may be generated and output.

The user can grasp the following by visually recognizing the image in which the keyword is structured for each business for the group.
For example, it is assumed that the information processing apparatus 100 generates feature information using a company to which the user belongs and competitors of the company as a group. Then, the information processing apparatus 100 generates an image in which an image in which keywords are structured for each business is generated for the company to which the user belongs and its competitors. The user can grasp the following by visually recognizing the image. That is, the user can grasp the characteristics of the company by visually recognizing the image in which the keyword is structured for each business. In addition, users can compare the images with keywords for each of their businesses with the images with keywords for each of their competitors' business. Can understand the strengths and weaknesses of each other and the strengths and weaknesses of competitors.
In addition, for a company that operates multiple businesses, the user can understand the structure of the keywords of the company and other companies that have similar keyword structures, so that the business that the company operates Accurately grasp.

By grasping such a situation, the user can find a more appropriate sales destination, business alliance partner, capital alliance partner, purchase sales destination, and the like.
The processing of this embodiment can be used in business operations (such as grasping own company / competition / other companies and partner consideration) at business companies. It can also be used in brokerage companies such as introducing M & A destinations at securities companies and introducing business matching destinations at banks. It can also be used in research and research institutions, institutional investors, etc. in surveys of industry and individual company trends in research institutions and institutional investors.

(Search process)
The CPU 101 can search for an organization related to the designated word based on the feature information generated by the processing of FIG.
Assume that the CPU 101 receives a request to search for an organization related to a certain word (for example, “helmet”) from an external information processing apparatus via the network I / F 104, for example.
In that case, the CPU 101 acquires the feature information generated by the processing of FIG. 2 from the auxiliary storage device 103, searches the group of organizations related to the specified word based on the acquired feature information, and searches Information indicating the selected organization is transmitted to the request source.
For example, the CPU 101 searches for a designated word from the “keyword” item in the keyword management table generated for each group included in the group. When the importance corresponding to the searched “keyword” is equal to or greater than a predetermined threshold, the CPU 101 transmits information indicating the organization including the searched “keyword” to the request source as a search result.

Further, the CPU 101 may perform the following processing.
The CPU 101 obtains a keyword related to the designated word and a keyword similar to the designated word based on the association management table generated for each group of groups included in the feature information and the table 1000. Search for. Then, the CPU 101 transmits the searched keyword list to the request source. The requesting information processing apparatus displays the received keyword list on the display unit and presents it to the user. The user additionally selects the word (s) used for the organization search from the displayed keyword list. The requesting information processing apparatus transmits the word selected by the user to the information processing apparatus 100 and requests to search for an organization related to the first transmitted word and the newly transmitted word.
For example, it is assumed that the requesting information processing apparatus first requests the information processing apparatus 100 to search for a group using the word “helmet”. In this case, the information processing apparatus 100 transmits a keyword related to “helmet” and a list of similar keywords to the request source. The user selects the word “breathability” from the received keyword list when he / she wants to find a company developing a helmet intended for use in summer. Then, the requesting information processing apparatus transmits the word “breathability” to the information processing apparatus 100 and requests to search for an organization related to “helmet” and “breathability”. In the following, the first designated word (“helmet”) is the original search word, and the additionally designated word (“breathability”) is the additional related word.

The CPU 101 searches for a group related to the original search word and the additional related word from the group of groups. The CPU 101 selects an organization using the original search word and an additional related word (or a word similar to the additional related word) as keywords from a table similar to the table described in FIG. 8 generated for each organization included in the organization group. Search for. Further, the CPU 101 may further search for an organization using the original search word and a word similar to the additional related word as keywords. Based on the table 1000, the CPU 101 can determine a word similar to the additional related word.
The CPU 101 determines a match score indicating how much the search group matches the search condition. For each retrieved group, the CPU 101 determines a match score based on the importance of the original search word and the additional related word (or a word similar to the additional related word) in the group. For example, the CPU 101 determines the match score so that the value increases as the importance of the original search word and the additional related word (or a word similar to the additional related word) in the group increases. In addition, for example, the CPU 101 makes the match score of the group related to the original search word and the additional related word (or a word similar to the additional related word) in the group larger than the match score of the non-related group. As such, the match score may be determined. In addition, for example, the CPU 101 sets the match score so that the match score of the group in which the original search word and the additional related word in the group are keywords of the same business cluster is larger than the match score of the group that is not. You may decide.

The CPU 101 transmits the retrieved group information and the match score information of each group to the requesting information processing apparatus. The requesting information processing apparatus arranges the retrieved group information in the descending order of match score and displays the information on the display unit. The user can visually recognize the information of the retrieved organizations and grasp the organizations that are likely to be related. Further, the requesting information processing apparatus may display a keyword for each business of each group on the display unit for a predetermined number (for example, three) of groups having the highest match score. Then, the user selects a keyword to be an additional related word to be further added from the displayed keywords. Then, the requesting information processing apparatus transmits information on the selected keyword to the information processing apparatus 100. The information processing apparatus 100 adds the received keyword to the additional related word.
Then, the CPU 101 repeats the same processing based on the original search word and the additional related word to which the received keyword is added (or a word similar to the additional related word). Thereby, the information processing apparatus 100 can search for an organization that matches the condition desired by the user.

Further, the CPU 101 can search for a group similar to the designated group from the group by performing the following process based on the feature information generated by the process of FIG.
For example, the CPU 101 receives designation of a certain group from an external information processing apparatus or the like, and receives a search request for a group similar to the group. In the following, the designated group is the original group. For example, for each of the keywords included in the original organization (hereinafter referred to as keyword (A)), the CPU 101 determines the importance in the original organization and the importance in other organizations (hereinafter referred to as search destination organizations). The sum of products is obtained as a similarity score indicating the degree of similarity between the original organization and the search destination organization. Then, the CPU 101 generates a list in which organizations included in the search destination organization are arranged in descending order of the obtained similarity score, and transmits the generated list to the search request source. Further, the CPU 101 may determine a group having a similarity score equal to or higher than a predetermined threshold as a group similar to the original group, and transmit information indicating the determined group to the search request source.

Further, the CPU 101 may obtain a similarity score in consideration of a keyword similar to the keyword (A). For example, for each keyword (A), the CPU 101 calculates the keyword (A) and the product of the importance of the keyword (A) in the original organization and the importance of the keyword similar to the keyword (A) in the search destination organization. A value obtained by multiplying the similarity with the keyword may be added to the similarity score to obtain the similarity score.
Further, the CPU 101 may obtain a similarity score in consideration of the business structure (which business cluster each keyword belongs to). For example, among the keywords (A), the CPU 101 belongs to a single business cluster (hereinafter referred to as keyword (B)) and belongs to the single business cluster. If so, you may do the following. That is, the CPU 101 multiplies the product of the importance of the keyword (B) in the original organization and the importance of the keyword (B) in the search destination organization by a coefficient (hereinafter referred to as the first coefficient). The similarity score may be obtained by adding to the similarity score. The first coefficient is, for example, 2 or the like. In this way, when the keyword (B) is in the same business cluster even at the search destination, the CPU 101 can determine that the business structure is similar and increase the similarity score.
Further, the CPU 101 may obtain a similarity score in consideration of the relevance between keywords. Among the keywords (A), a keyword belonging to a single business cluster in the original organization and having a relevance exceeding a threshold (hereinafter referred to as a keyword (C)) also in the search destination organization When belonging to a single business cluster and the relevance exceeds the threshold, the following may be performed. That is, the CPU 101 calculates a similarity by multiplying the product of the importance of the keyword (C) in the original organization and the importance of the keyword (C) in the search destination organization by a coefficient (hereinafter, the second coefficient). The similarity score may be obtained by adding to the degree score. The second coefficient may be a value larger than the first coefficient such as 3, for example. In this way, the CPU 101 determines that the business structure is similar when the keyword (C) is in the same business cluster at the search destination and the relevance exceeds the threshold, and the similarity score is determined. , Can be higher.

Further, the CPU 101 can also search for a company having a business similar to the designated business from the group by performing the following processing based on the feature information.
The CPU 101 receives, for example, designation of a certain business in a certain organization from an external information processing apparatus or the like, and accepts a search request for a group having a business similar to that business. In the following, the designated business is the original business. For example, for each of the keywords included in the original business (hereinafter referred to as keyword (A ′)), the CPU 101 determines the importance in the original business and the business included in other organizations (hereinafter referred to as search destination business). The sum of the products of the importance levels is obtained as a similarity score indicating the degree of similarity between the original business and the search destination business. Then, the CPU 101 generates a list in which organizations including the search destination business are arranged in descending order of the obtained similarity score, and transmits the generated list to the search request source. In addition, the CPU 101 determines an organization including a business whose similarity score is equal to or higher than a predetermined threshold as an organization including a business similar to the original business, and transmits information indicating the determined organization to a search request source. May be.

The CPU 101 may obtain a similarity score in consideration of a keyword similar to the keyword (A ′). For example, for each keyword (A ′), the CPU 101 calculates the keyword (A ′) by the product of the importance of the keyword (A ′) in the original business and the importance of a keyword similar to the keyword (A ′) in the search destination business. A value obtained by multiplying the similarity between A ′) and the keyword may be added to the similarity score to obtain the similarity score.
Further, the CPU 101 may obtain a similarity score in consideration of the business structure (which business cluster each keyword belongs to). For example, the CPU 101 assigns a keyword (A ′) that belongs to a single business cluster (hereinafter referred to as a keyword (B ′)) to a single business cluster corresponding to the search destination business. If you belong, you may do the following. That is, the CPU 101 adds a value obtained by multiplying the product of the importance of the keyword (B ′) in the original business and the importance of the keyword (B ′) in the search destination business by the first coefficient to the similarity score. Then, a similarity score may be obtained. In this way, when the keyword (B ′) is in the same business cluster even at the search destination, the CPU 101 can make the similarity score higher by assuming that the business structures are similar.
Further, the CPU 101 may obtain a similarity score in consideration of the relevance between keywords. Among the keywords (A ′), the search destination business for a keyword (hereinafter referred to as a keyword (C ′)) that belongs to a single business cluster in the original business and whose relevance exceeds a threshold value. If it belongs to a single business cluster corresponding to, and the relevance exceeds the threshold, the following may be performed. That is, the CPU 101 adds, to the similarity score, a value obtained by multiplying the product of the importance of the keyword (C ′) in the original business and the importance of the keyword (C ′) in the search destination business by the second coefficient. Then, a similarity score may be obtained. By doing this, the CPU 101 determines that the business structure is similar when the keyword (C ′) is in the same business cluster even at the search destination and the relevance exceeds the threshold, and the similarity score Can be made higher.

(Summary)
As described above, in the present embodiment, the information processing apparatus 100 extracts a keyword indicating the characteristics of a group from a document data group including a plurality of document data related to the group, and groups the keywords for each business that is an element included in the group. Thus, the keywords indicating the characteristics of each business are determined, the relevance between the keywords indicating the characteristics of the organization is determined, and the business in the organization is determined based on the determined keywords for each business and the relationship between the keywords. Feature information indicating the relationship between each feature was generated. Thereby, the information processing apparatus 100 can generate information indicating what kind of business is in the organization.

<Other embodiments>
In the first embodiment, the information processing apparatus 100 is a single information processing apparatus. However, the information processing apparatus 100 may be configured as a system including a plurality of information processing apparatuses connected to be communicable with each other via a network (LAN or the Internet). In that case, the CPU of each of the plurality of information processing devices included in the information processing device 100 executes the processing in cooperation with each other based on the program stored in the auxiliary storage device of each information processing device, so that the information processing device 100 And the processing described in the flowchart of FIG. 2, (image output processing), (search processing), and the like are realized.
In the first embodiment, the information processing apparatus 100 generates feature information indicating the relationship between features for each business that is an element included in the organization. However, the information processing apparatus 100 may generate feature information indicating the relationship between features for each element included in an object other than a group. For example, the information processing apparatus 100 extracts a keyword indicating an individual characteristic from a document data group including a plurality of document data posted by an individual to a document posting site or the like, and clusters each document data included in the document data group. , Obtaining each cluster indicating the elements (personality, hobbies, etc.) included in the individual, determining keywords indicating the characteristics of each cluster, determining the relevance between the keywords, and keywords indicating the characteristics of each determined cluster; Information indicating the relevance between keywords may be generated as feature information.

As mentioned above, although preferable embodiment of this invention was explained in full detail, this invention is not limited to the specific embodiment which concerns.
For example, part or all of the functional configuration of the information processing apparatus 100 described above may be implemented in the information processing apparatus 100 as hardware.

100 Information processing apparatus 101 CPU

Claims (14)

Extraction means for extracting a keyword indicating the feature of the object from the object document data group based on the appearance frequency of words in the object document data group, which is a document data group including a plurality of document data related to the object;
Based on the appearance frequency of the keyword extracted by the extracting unit in the document data included in the object document data group and the rarity of the keyword extracted by the extracting unit in the object document data group, the object document First feature amount determining means for determining a feature amount of document data included in the data group;
Classifying means for classifying document data included in the object document data group into one or more sections respectively corresponding to elements included in the object based on the feature amount determined by the first feature amount determining means. When,
Document data included in the object document data group by the classification unit is extracted by the extraction unit based on the appearance frequency of the keyword extracted by the extraction unit in the classification included in the one or more classifications. A keyword determining means for determining a keyword indicating a feature of an element corresponding to the category included in the one or more categories from the keywords,
Second feature value determining means for determining the feature value of the keyword extracted by the extracting means;
Relevance determination means for determining relevance between keywords extracted by the extraction means based on the feature quantity determined by the second feature quantity determination means;
Generating means for generating feature information indicating a relationship between features of each element in the object based on the keyword determined by the keyword determining means and the relevance determined by the relevance determining means;
An information processing apparatus. First score determination means for determining a score value indicating the importance of the word in the object based on the appearance frequency of the word in the object document data group;
Document data included in the object document data group by the classification unit is extracted by the extraction unit based on the appearance frequency of the keyword extracted by the extraction unit in the classification included in the one or more classifications. Second score determining means for determining a score value indicating importance in a section included in the one or more sections of the keyword,
Further comprising
The extraction unit extracts a keyword indicating the feature of the object from the object document data group based on the score value determined by the first score determination unit,
The keyword determining unit is configured to obtain a feature of an element corresponding to a category included in the one or more categories from the keyword extracted by the extracting unit based on the score value determined by the second score determining unit. The information processing apparatus according to claim 1, wherein a keyword to be displayed is determined. Further comprising importance determining means for determining importance indicating the importance of the keyword extracted by the extracting means in the object based on the score value determined by the first score determining means;
The generating means calculates the feature information based on the keyword determined by the keyword determining means, the relevance determined by the relevance determining means, and the importance determined by the importance determining means. The information processing apparatus according to claim 2 to be generated. Based on the score value determined by the second score determining means, further comprising an importance determining means for determining an importance indicating the importance of the keyword extracted by the extracting means in the object,
The generating means calculates the feature information based on the keyword determined by the keyword determining means, the relevance determined by the relevance determining means, and the importance determined by the importance determining means. The information processing apparatus according to claim 2 to be generated.   The said 2nd feature-value determination means determines the feature-value of the keyword extracted by the said extraction means based on the score value determined by the said 2nd score determination means. The information processing apparatus described.   The extraction means, based on the appearance frequency of words in the object document data group and the rarity of words in a plurality of document data groups related to each of a plurality of objects including the object, from the object document data group, The information processing apparatus according to claim 1, wherein a keyword indicating a feature of the object is extracted.   The first feature amount determining means includes the appearance frequency of the keyword extracted by the extracting means in the document data included in the object document data group, and the rareness of the keyword extracted by the extracting means in the object document data group. And determining a score value indicating the importance of the keyword extracted by the extraction means in the document data included in the object document data group, and determining the object document data group based on the determined score value. 7. A vector indicating the feature of the document data included in the document is determined, the determined vector is dimensionally compressed, and the dimension-compressed vector is determined as a feature amount of the document data included in the object document data group. The information processing apparatus according to claim 1.   The keyword determination means includes the appearance frequency of the keyword extracted by the extraction means in the category included in the one or more categories, the rarity of the keyword extracted by the extraction means in the one or more categories, 8. The information processing apparatus according to claim 1, wherein a keyword indicating a feature of an element corresponding to a category included in the one or more categories is determined from the keywords extracted by the extraction unit based on . Further comprising meaning determining means for determining the meaning of the keyword extracted by the extracting means;
The generation unit generates the feature information based on the keyword determined by the keyword determination unit, the relevance determined by the relevance determination unit, and the meaning determined by the meaning determination unit. The information processing apparatus according to claim 1.   The information processing apparatus according to claim 9, wherein the meaning determination unit extracts a potential meaning of the keyword extracted by the extraction unit, and determines the extracted potential meaning as a meaning of the keyword extracted by the extraction unit. .   11. The information processing apparatus according to claim 1, further comprising an output unit that outputs an image indicating a relationship between features of each element in the object based on the feature information generated by the generation unit. The extraction means extracts a keyword indicating the feature of the object from the object document data group based on the appearance frequency of words in the object document data group including a plurality of document data related to the object that is a company,
The classifying unit converts document data included in the object document data group based on the feature amount determined by the first feature amount determining unit to one corresponding to a business that is an element included in the object. The information processing apparatus according to claim 1, which is classified into the above categories. An information processing method executed by an information processing apparatus,
An extraction step of extracting a keyword indicating the feature of the object from the object document data group based on the appearance frequency of words in the object document data group, which is a document data group including a plurality of document data related to the object;
Based on the appearance frequency of the keyword extracted in the extraction step in the document data included in the object document data group and the rarity of the keyword extracted in the extraction step in the object document data group, the object document A first feature amount determining step for determining a feature amount of document data included in the data group;
A classification step of classifying document data included in the object document data group into one or more sections corresponding to elements included in the object, based on the feature amounts determined in the first feature amount determination step. When,
The document data included in the object document data group in the classification step is extracted in the extraction step based on the appearance frequency of the keyword extracted in the extraction step in the classification included in the one or more classifications. A keyword determination step for determining a keyword indicating a feature of an element corresponding to the category included in the one or more categories from the keywords,
A second feature amount determination step for determining a feature amount of the keyword extracted in the extraction step;
A relevance determination step for determining relevance between keywords extracted in the extraction step based on the feature amount determined in the second feature amount determination step;
A generation step of generating feature information indicating a relationship between features of each element in the object based on the keyword determined in the keyword determination step and the relationship determined in the relevance determination step;
An information processing method including:   A program for causing a computer to function as each unit of the information processing apparatus according to any one of claims 1 to 12.

Images