JP2016212461A - Conception processing apparatus and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a conception processing apparatus and a program which can obtain an index (the degree of association) showing strength of association between conceptions on the basis of a conception map.SOLUTION: A conception weight calculation part reads conception map data showing whether or not there is relation between conceptions and calculates a weight value of conception which becomes larger with respect to the conception with the smaller number of links showing relations between one conception and another. A path weight calculation part calculates a weight value of a path in accordance with the weight value of the conception with respect to conception included on the path with respect to paths connecting one conception with another by link in conception map data. A degree-of-association calculation part calculates the degree of association between one conception and another in accordance with the weight value of path with respect to a path connecting the conception with the other.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a concept processing apparatus and program for processing a concept. In particular, the present invention relates to a concept processing apparatus and program for obtaining the degree of association between concepts.

  In information retrieval processing and knowledge processing using natural language, it is expected that more advanced processing can be performed by utilizing the relationship between concepts. In addition, it is expected that relationships between concepts can be collected efficiently without human intervention.

  As information representing the relationship between concepts, a relationship name between concepts, a numerical value representing the strength of the relationship between concepts, and the like can be considered. Among these, if the strength of association between concepts can be collected accurately and efficiently, it can be expected to contribute to the advancement of information retrieval processing and knowledge processing described above.

  Non-Patent Document 1 describes a technique for generating a large-scale similar word list using a result of large-scale stochastic clustering of dependency. This technology is based on the analysis of large-scale language resources. Based on the distribution hypothesis that semantically similar words appear in similar contexts, this technique defines contexts with words that appear around a word or words that have a dependency relationship. Analyzes the frequency of co-occurrence of

  Non-Patent Document 2 describes that a database in which the degree of association between words is calculated by analyzing a large amount of sentences and calculating the frequency of common occurrences in the same sentence is disclosed. As the co-occurrence score, three types of co-occurrence frequency, Dice coefficient, and discounting mutual information amount are used.

Kazuma Kazama, Stijn De Saeger, Kentaro Torizawa, Maki Murata, “Creating a Large Similar Word List Using Dependent Stochastic Clustering”, Linguistic Processing Society, 15th Annual Conference, Proceedings, March 2009 , P. 84-87, [online], [April 1, 2015 search], Internet <URL: http://www.anlp.jp/proceedings/annual_meeting/2009/pdf_dir/C1-6.pdf> ALAGIN Language Resource / Speech Resource Site, “(A-5) Word Co-occurrence Frequency Database”, [online], [April 10, 2015 search], Internet <URL: https: //alaginrc.nict.go. jp / resources / nict-resource / li-info / li-outline.html>

  In the method according to Non-Patent Document 1, for example, a high degree of association cannot be set for a pair having a relationship between a dish and an ingredient, such as a concept pair of “Paella” and “clam”. For this reason, even if the method of Non-Patent Document 1 is applied as it is to the degree of association in the concept map, good performance cannot be obtained.

  In addition, the method according to Non-Patent Document 2 is effective for the relationship between concept pairs such as cooking and ingredients, but similar usage, for example, the concept pair of “soccer” and “baseball”. A high value cannot be given to a word pair that is a pair that is rarely used in the same context. Therefore, it is difficult to apply this method as it is as the degree of association in the concept map.

  In addition, both the method according to Non-Patent Document 1 and the method according to Non-Patent Document 2 must analyze large-scale language resources.

  The present invention has been made on the basis of recognition of these problems, and provides a concept processing device and program capable of obtaining an index (degree of association) representing the strength of association between concepts based on a concept map. It is something to be offered.

  In order to solve the above problems, a concept processing apparatus according to an aspect of the present invention reads concept map data representing the presence or absence of a relationship between concepts, and the number of links representing a relationship between a concept and another concept is A concept weight calculation unit that calculates a weight value of the concept that becomes a larger value for the concept, and a path formed by connecting a concept and another concept with the link in the concept map data. The path weight calculation unit for calculating the weight value of the path according to the weight value of the concept for the concept included in the path, and the path for connecting a concept and another concept A degree-of-association calculating unit that calculates the degree of association between the concept and the other concept according to a path weight value.

  According to another aspect of the present invention, in the above concept processing device, the path weight calculation unit takes a harmonic average or a geometric average of the concept weight values for each concept included on the path. The path weight value is calculated.

  Further, according to one aspect of the present invention, in the concept processing device, the concept weight calculation unit calculates a logarithm of the inverse of the number of links between the concept and the other concept. The weight value is calculated.

  Further, according to one aspect of the present invention, in the above concept processing device, when the relevance calculation unit includes a plurality of paths in which the certain concept and the other concept are connected by the link, By multiplying the number of the plurality of paths by the value of the maximum weight of the plurality of path weights to calculate the degree of association between the concept and the other concept, Alternatively, the sum of the weights of the plurality of paths is calculated as the degree of association.

  According to another aspect of the present invention, the above concept processing device includes a similarity acquisition unit that acquires similarity data representing a similarity between concepts, and the path weight calculation unit includes the path for the path. The degree of similarity of the concept pair connected by the link included above is acquired from the similarity data, and the path weight value is calculated in accordance with the acquired degree of similarity.

  One embodiment of the present invention is a program for causing a computer to function as the conceptual processing device described above.

  According to the present invention, the degree of association between concepts can be obtained based on the concept map.

It is the block diagram which showed schematic function structure of the conceptual processing apparatus by 1st Embodiment of this invention. It is the schematic which shows an example of the concept map which the concept map memory | storage part by the same embodiment hold | maintains. It is the schematic for demonstrating the weight of the concept calculated in the embodiment, and has shown a part of concept map. It is the schematic which shows the structure of the data for holding | maintaining the weight of a concept, etc. which are put in the storage area in the working storage part by the embodiment. It is the schematic for demonstrating the weight of the path | pass calculated in the same embodiment, and has shown a part of concept map. It is the schematic which shows the means which stores the information which should be memorize | stored temporarily when the path weight calculation part searches a concept map, and is data put in the storage area in the working storage part by the embodiment. FIG. 5 is a schematic diagram showing means for storing information related to a path to each concept when the path weight calculation unit searches a concept map, which is data placed in a storage area in the work storage unit according to the embodiment. It is the 1st of the flowchart which shows the procedure of the process of the conceptual processing apparatus by the embodiment. It is 2nd of the flowchart which shows the procedure of the process of the conceptual processing apparatus by the embodiment. FIG. 5 is a schematic diagram showing an example of values used in the embodiment and held in the storage area shown in FIG. 4. FIG. 7 is a schematic diagram illustrating an example of a transition of a value which is data used in the embodiment and held in the storage area illustrated in FIG. 6. FIG. 8 is a schematic diagram illustrating an example (part 1) of transition of values used in the embodiment and held in the storage area illustrated in FIG. 7. FIG. 8 is a schematic diagram illustrating an example (part 2) of a transition of values held in the storage area illustrated in FIG. 7, which is data used in the embodiment. FIG. 8 is a schematic diagram showing an example (part 3) of a transition of values held in the storage area shown in FIG. 7, which is data used in the embodiment. It is a block diagram which shows schematic function structure of the conceptual processing apparatus by 2nd Embodiment of this invention. It is the schematic which shows the structure of the similarity data showing the similarity between the concepts which the similarity acquisition part in the embodiment acquires from the outside.

Next, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
FIG. 1 is a block diagram showing a schematic functional configuration of the conceptual processing device according to the first embodiment. As illustrated, the concept processing device 1 includes a concept weight calculation unit 12, a path weight calculation unit 14, a relevance calculation unit 16, a parameter setting unit 18, a concept map storage unit 20, and a work storage unit 22. And the relevance degree storage unit 30. Each of these units is realized using, for example, an electronic circuit (including a computer). The concept map storage unit 20, the work storage unit 22, and the relevance level storage unit 30 are realized by using, for example, a magnetic hard disk device or a semiconductor memory as means for storing information.

The concept weight calculation unit 12 reads concept map data indicating the presence or absence of a relationship between concepts from the concept map storage unit 20, and the smaller the number of links indicating the relationship between a certain concept and another concept, the larger the concept. The weight value of the concept that becomes a value is calculated.
In addition, the concept weight calculation unit 12 calculates the weight value of the concept by calculating the logarithm of the inverse of the number of links between the concept and the other concept.

In the concept map data, the path weight calculation unit 14 determines, for a path formed by connecting a concept and another concept by the link, according to the concept weight value for the concept included in the path. The path weight value obtained is calculated.
In addition, the path weight calculation unit 14 calculates the weight value of the path by taking the harmonic average or the geometric mean of the concept weight values for each concept included in the path.
In addition, the path weight calculation unit 14 starts with the concept of the calculation target as a particularly specific procedure, and performs the search process for searching the concept map data while reaching the concept of the calculation target and the search process. The path weight value for the path between the concept and the path weight is calculated, and the path weight value calculated for the path is another link that connects the concept to be calculated and the reached concept. If the calculated path weight value is larger than the path weight value that has already been calculated for the path, the calculated path weight value is the weight for the maximum weight path between the concept to be calculated and the reached concept. Is written in the storage means as the value of.

The degree-of-association calculation unit 16 calculates the degree of association between the concept and the other concept according to the value of the path weight for a path connecting a certain concept and the other concept.
In addition, the relevance calculation unit 16 determines that the plurality of paths are included in the number of the plurality of paths, particularly when there are a plurality of paths formed by connecting the certain concept and the other concept by the link. The degree of association between the concept and the other concept is calculated by multiplying the maximum value of the path weights among the weight values of.

  The parameter setting unit 18 sets parameters stored in the work storage unit 22. Specifically, the parameter setting unit 18 appropriately sets a value of a parameter MAX_Distance described later.

  The concept map storage unit 20 stores concept map data indicating the presence or absence of a relationship between concepts. Details of the concept map data will be described later.

  The work storage unit 22 (work memory, which may be simply referred to as “memory” below) is a storage unit and has an area for storing information for work being processed. The work storage unit 22 has an area for storing parameters for processing.

  The association degree storage unit 30 stores the association degree between concepts calculated by the association degree calculation unit 16. The relevance data stored in the relevance storage unit 30 can be referred to from other devices as appropriate.

  Next, the concept map data stored in the concept map storage unit 20 will be described. The concept map is data representing whether or not there is a relationship between a concept and another concept. One form representing a concept map is a graph composed of nodes and links. Each concept corresponds to a node in this graph. The relationship between concepts corresponds to a link in the graph. There is either a relationship or not between one concept and another. Where there is a relationship between concepts, there is a link between those concepts. If there is no relationship between concepts, there is no link between those concepts. Note that the data format of the concept map may be other than that described here as long as it includes information indicating the presence or absence between concepts. It should be noted that as the concept map data used here, an already constructed concept map may be used as it is, or a new concept map may be constructed. One example of how to construct a concept map is to analyze concepts that appear in a vast amount of language resources. The language resource used here is, for example, web text that can be collected via the Internet. The concept analysis method that appears is, for example, analyzing the appearance of a plurality of concepts in a predetermined syntax, or analyzing a plurality of concepts that appear in the same context. In addition, when data that can be downloaded from a website directly represents a relationship between a plurality of concepts, the relationship obtained there may be directly used. For example, encyclopedia sites such as Wikipedia (URL https://en.wikipedia.org/) provide relationships between concepts on mobile phones that can be downloaded. An example of a concept map will be described next.

  FIG. 2 is a schematic diagram illustrating an example of a concept map held by the concept map storage unit 20. An actual concept map is very large-scale data, and the number of concepts included in the concept map is hundreds of thousands, millions, or more. However, in the same figure, for simplicity of explanation, a concept map including seven concepts is illustrated. As shown in the figure, the concept map is represented by an undirected graph. Each node of the graph corresponds to one concept. Each link in the graph represents a direct relationship between concepts. There are seven concepts included in the illustrated concept map: “disease”, “hypertension”, “adult disease”, “diabetes”, “cold”, “fever”, and “influenza”. In addition, when the links included in the illustrated concept map are represented by pairs of concepts connected by the links, “disease-hypertension”, “disease-adult disease”, “disease-diabetes”, “hypertension-adult disease” , “Diabetes-adult disease”, “disease-cold”, “disease-influenza”, “cold-fever”, and “influenza-fever”.

  The concept map storage unit 20 holds information on concept maps in a data format that is easy to process. As an example, the concept map storage unit 20 includes table data that associates a node identifier with the concept represented by the node, and table data that associates a link identifier and a pair of identifiers of two nodes connected by the link. Holds information about concept maps. Note that the present invention is not limited to this example, and a data format generally used for expressing a graph may be applied.

Next, the degree of association between concepts calculated by the concept processing device 1 will be described.
In the present embodiment, the degree of association between concepts is calculated according to how the concepts are connected in the concept map. In the processing by the concept processing device 1, the concept corresponds to the node on the graph as described above. In addition, there are cases where concepts are connected by links and cases where concepts are not connected. One link connects adjacent concepts and connects only those adjacent concepts. When one concept can reach another concept on the concept map, the concepts are connected to each other. The condition that the concepts are in a connected state is that one concept can reach the other concept by following one or more links. The route followed at that time is called a path here. A path consists of one or more links. When there are a plurality of links, the path is formed by connecting these links in series. When one concept and another concept are connected, there is not always a single path connecting the two concepts. Sometimes it can be reached through several different paths. One pass does not pass the same concept more than once. That is, the path does not form a closed circuit or go back. In this embodiment, based on the concept weight, the weight of the path including the concept is calculated. Further, the degree of association between concepts is calculated based on the path weight. The degree of association between concepts is also determined by the number of paths connecting the concepts.

  First, the concept processing device 1 calculates the weight of each concept included in the concept map. The concept weight is a numerical value representing the importance of the concept in calculating the degree of association. The concept weight is calculated according to the number of links from the concept to other concepts. The smaller the number of links, the greater the value of the concept weight. This is based on the premise that the concept is precious when the number of links connected from a concept is small. More specifically, the concept weight is calculated by the logarithm of the reciprocal of the number of links from the concept to other concepts connected. However, in order to normalize the value of the concept weight, the logarithm of the numerical value obtained by dividing the total number of concepts on the concept map by the number of links is used as the concept weight. The base of the logarithm may be determined as appropriate.

  Next, the conceptual processing device 1 calculates a path weight. According to the above definition, there are multiple concepts on the path. Both ends of the path are also concepts. The path weight is calculated by the weight of each concept included in the path. More specifically, the path weight is calculated by some average of the weights of the concepts included in the path. Here, the certain average is, for example, a simple arithmetic average, harmonic average, or geometric average. More specifically, the concept processing device 1 calculates the harmonic average value of the weights of each concept as the value of the weight of the path. This is because it has been found from experiments using an actual concept map that the most suitable path weight can be calculated by using the harmonic average.

  Next, the concept processing device 1 calculates the degree of association between concepts based on the path weight value. The degree of association between concepts according to the present embodiment is calculated as the product of the maximum weight value of the path weights between the concepts and the number of paths existing between the concepts.

  A procedure for calculating the degree of association between concepts with a realistic calculation amount will be described later. In the procedure, in order to search the concept map with a reasonable amount of calculation, an upper limit is set on the search distance (= the number of serial links included in one path). This upper limit value can be set as a parameter. Note that the degree of association between concepts when using this procedure is directed. That is, the relevance value from concept A to concept B may be different from the relevance value from concept B to concept A.

The following is a summary.
Thus, the degree of association between concepts calculated by the concept processing device 1 is obtained according to the number of paths existing between the concepts. That is, the greater the number of paths, the greater the relevance value. This is consistent with the assumption that concepts connected by various paths are highly related to each other.
Further, the degree of association between concepts calculated by the concept processing device 1 is obtained in accordance with the path weight (its maximum value). That is, the greater the maximum value of the path weight, the greater the relevance value. This is consistent with the assumption that concepts connected by highly important paths are highly related to each other.
In addition, the path weight calculated by the concept processing device 1 is obtained according to the concept weight included in the path. The greater the concept weight, the greater the path weight. As a path weight, it is preferable to use a harmonic average of concept weights, but a geometric average of concept weights may be used.

Here, the concept weight calculated by the concept weight calculator 12 will be described.
FIG. 3 is a schematic diagram for explaining the weight of the concept, and shows a part of the concept map. In the figure, circles labeled “A”, “C”, “D”, and “E” are nodes corresponding to concepts. A line extending from the node is a link between the nodes and represents a direct relationship between the concepts. In the figure, only a part of the concept map is shown, but the actual concept map extends beyond the range shown in the figure.

  Although only a part of the concept map is shown in the figure, the number of nodes (concept number) of the entire concept map is N. N is a positive integer, for example, N = 10000. In order to explain the weight of the concept, as an example, attention is paid to a node labeled “C” shown in FIG. From this node “C”, direct links to nodes “A”, “D”, and “E” are connected. Note that a direct link from the node “C” to a node other than the nodes “A”, “D”, and “E” is not connected. That is, the number of links from the node “C” to other nodes directly connected is three.

The concept weight in this embodiment is calculated by the following equation.
Concept weight = log (N / n)
Here, N is the total number of concepts included in the entire concept map as described above. N is the number of concepts directly connected to the concept. The base of the logarithm may be determined as appropriate, but here the base is 10. In the illustrated example, the number of other nodes directly connected to the node “C” is 3 (n = 3), and the total number of nodes included in the concept map is 10,000 (N = 10000). Therefore, the weight of the concept corresponding to the node “C” is calculated as follows using the above formula.
Conceptual weight of node “C” = log ₁₀ (10000/3) = 3.52
However, the third decimal place is rounded off.

  FIG. 4 is a schematic diagram showing a configuration of data for holding concept weights. This data is provided in a predetermined area of the work storage unit 22. The illustrated data is tabular data and includes items of node ID, concept (concept corresponding to the node), connected concept, and weight. One row of this table corresponds to one concept, ie one node. The node ID is an ID that uniquely identifies a node in the concept map. In the illustrated example, an integer value is used as the node ID. In the concept item, a word representing the concept is stored. In the connected concept item, a list of node IDs of concepts connected by a direct link from the concept is stored. In the weight item, a weight value calculated for the concept is stored.

  In the figure, some values of the entire table are shown. The node ID ranges from 98 to 102. The concept of each node is A, B, C, D, E. When attention is paid to the concept C (node ID is 100), there are three concept node IDs connected to the concept C: 98, 101, and 102. That is, the concept C is connected to the concepts A, D, and E by a direct link (as illustrated in FIG. 3). Further, the value 3.52 is stored as the weight of the concept C.

Here, the path weight calculated by the path weight calculation unit 14 will be described.
FIG. 5 is a schematic diagram for explaining path weights, and shows a part of the concept map. In the figure, the circles labeled “1”, “2”, “3”, “4”, “5” are nodes corresponding to concepts, respectively. A line extending from the node is a link between the nodes and represents a direct relationship between the concepts. In the figure, only a part of the concept map is shown, but the actual concept map extends beyond the range shown in the figure.

  A path is a route from one concept to another. For example, in the figure, there are at least two paths from the node “1” to the node “5”. The first path is represented as “1-2-3-5” as a column of nodes. Similarly, the second path is represented as “1-2-4-5” as a node column. In this way, a path can be represented by arranging nodes on the path as a column including nodes at both ends. In this embodiment, the path weight is defined as a harmonic average of the weights of each node (including nodes at both ends) on the path.

In general, the harmonic average of _M numerical values w ₁ , w ₂ ,..., W _M is expressed by the following equation. That is, M / (1 / w ₁ + 1 / w ₂ +... + 1 / w _M ).

Therefore, when the weights from the nodes “1” to “5” (that is, the concept weights) have been calculated, and each of them is _denoted by w _n1 , w _n2 ,..., W _nM , The weight of the path and the weight of the second path are each expressed by the following equations.
(Weight of the first path) = 4 / (1 / w _n1 + 1 / w _n2 + 1 / w _n3 + 1 / w _n5 )
(Weight of the second path) = 4 / (1 / w _n1 + 1 / w _n2 + 1 / w _n4 + 1 / w _n5 )

  FIG. 6 is a schematic diagram illustrating a part of the data configuration held by the work storage unit 22. The data shown in the figure is mainly means for storing information to be temporarily stored when the path weight calculation unit 14 searches for a concept map while referring to the concept map storage unit 20. The meaning of each data item included in the figure is as follows.

The data item “concept of calculation target” is a concept selected at that time in the concept map, and is a concept for which the degree of association is calculated. For example, when the concept to be calculated is “disease” in the concept map shown in FIG. 2, the concept processing device 1 calculates the degree of association from this concept “disease” to each other concept. In addition, the concept of calculation object may be called "concept A" for convenience below.
The data item “starting point concept array” is an array having, as elements, a concept that becomes a starting point when searching for a path of a predetermined distance (a distance of a value held by the variable D) from the concept to be calculated.
The data item “number of concepts traced” is data that holds the distance from the concept to be calculated at that time, and corresponds to the variable D described above.

The data item “concept of origin” is a concept selected as the origin at that time. In the data item “concept of starting point”, one concept selected from the array of starting point concepts is stored. The concept of the starting point may be referred to as “concept C” for convenience in the following.
The data item “arrangement of adjacent concepts” is an array having, as elements, concepts adjacent to the concept of the starting point. Here, the concept “adjacent” means that one concept and another concept are connected via a single link. For example, in the example of the concept map shown in FIG. 2, the concept “disease” and the concept “hypertension” are adjacent to each other. The concept “fever” and the concept “cold” are concepts adjacent to each other. Further, since the concept “diabetes” and the concept “cold” are not directly connected by a link, they are not adjacent to each other. Note that the adjacent concept (that is, one element of the array of adjacent concepts) may be referred to as “concept B” for convenience in the following.

  A usage example of each data area shown in FIG. 6 will be described later as a processing example.

  FIG. 7 is a schematic diagram illustrating a part of the data configuration held by the work storage unit 22. The data shown in the figure mainly stores numerical values for calculating the relevance of each concept when the path weight calculation unit 14 searches the concept map while referring to the concept map storage unit 20. is there. Specifically, the path weight calculation unit 14 uses a processing procedure described later to calculate a path between a concept to be calculated (concept A) and each concept (especially when there are a plurality of paths, the weight is calculated). The information about the maximum path is updated sequentially. This data is tabular data, and each row of this table corresponds to one concept. The meaning of each data item included in the figure is as follows.

  The data item “concept” stores data of words representing the concept. For example, in the example of the concept map shown in FIG. 2, this data item stores information that uniquely identifies the concept, such as “hypertension” or “adult disease”. Further, in terms of the processing procedure of the path weight calculation unit 14, the concept represented by this data item corresponds to the “concept B”.

The data item “MAX_Weight” holds the value of the maximum weight among the weights of the path from the concept to be calculated (concept A) to the concept (concept B). In general, when the concept A and the concept B are connected within a predetermined distance (MAX_Distance), the number of paths (routes that can be traced) between the two concepts is one or more. The path weight calculation unit 14 updates the data item “MAX_Weight” with the maximum value among the already calculated path weights among these paths.
The data item “MAX_Path” stores information on a path corresponding to the above “MAX_Weight”. That is, the data item “MAX_Path” is for holding information on the path having the maximum weight already calculated among the paths between the concept A and the concept B. Specifically, the data item “MAX_Path” holds a column of concepts (including concepts at both ends of the path) included in the path as information on the path. Such information on the concept column included in the path uniquely identifies the path on the concept map.

The data item “NUM_Path” stores the number of paths from concept A to concept B. Note that when the path weight calculation unit 14 searches the concept map, the value of NUM_Path is calculated as follows. That is, the number of paths from concept A to concept B when D = n (where 0 ≦ n <MAX_Distance−1) is NUM_Path _{D = n} (A, B). The number of paths from concept A to concept C when _{D = n} is NUM_Path _{D = n} (A, C). Here, NUM_Path _{D = n} (A, B) does not include the number of paths passing through the concept C. When a path by a direct link from concept C to concept B is searched for at D = n + 1 (that is, a path from concept A to concept B is searched via concept C at D = n + 1). The following formula holds:

NUM_Path _{D = n + 1} (A, B) = (NUM_Path _{D = n} (A, B)) + (NUM_Path _{D = n} (A, C))

Next, a procedure of processing by the conceptual processing device 1 will be described.
FIG. 8 and FIG. 9 are flowcharts showing a procedure of processing by the conceptual processing device 1. These two figures are connected to each other by a connector of the flowchart. Hereinafter, the procedure will be described with reference to this flowchart.

  First, in step S10 (FIG. 8), the parameter setting unit 18 assigns the maximum distance (MAX_Distance) as a parameter value. Specifically, the parameter setting unit 18 writes the MAX_Distance setting value in a predetermined area in the work storage unit 22. The value of MAX_Distance assigned is a positive integer. The value of MAX_Distance defines the maximum distance when the concept is followed from concept to concept in the concept map. For example, when a concept and a concept are connected by a direct link, the distance between the concepts by the route is 1. In addition, when a concept and a concept are indirectly connected by a link through only one intermediate concept in the middle, the distance between the concepts by the route is 2. That is, when a concept and a concept are indirectly connected by a link through one or more intermediate concepts, the number of links when following the route is the distance between the concepts.

  Next, in step S20, the concept weight calculation unit 12 creates a concept list in which each concept included in the concept map is enumerated by the concepts linked by the direct link. The concept weight calculator 12 writes the concept list in the connected concept column in the table shown in FIG. The concept weight calculator 12 calculates the weight of each concept based on the connected concept list. The calculation method is based on the log (N / n) formula as described above. The concept weight calculation unit 12 writes the value calculated for each concept into the weight column in the table shown in FIG.

  The processes from the next step S30 to S90 and S100 are processes in which the path weight calculation unit 14 calculates and updates the path weight while following the concept map.

  In step S30, the path weight calculation unit 14 stores the concept to be calculated (referred to as concept A) in the starting concept array. That is, the concept to be calculated is the initial value of the starting point concept. Further, the path weight calculation unit 14 initializes the value of D, which is a variable for counting the distance traveled, to 0. This variable D stores the value of the path length (number of links included in the path) at the time when the concept map is followed from concept to concept. The concept processing apparatus 1 determines a termination condition for searching the concept map based on the relationship between the variable D and the value of the above-described parameter MAX_Distance.

  In step S40, the path weight calculation unit 14 sets the concept selected from the starting point concept array as the starting point concept (referred to as concept C). The path weight calculation unit 14 writes the selected concept in an area for storing the concept of the starting point in the memory. Then, the path weight calculation unit 14 writes all of the previous concepts connected by a direct link from the concept of the starting point (concept C) to the array of adjacent concepts. At this time, the path weight calculation unit 14 may obtain a concept connected to the concept C with reference to the concept map storage unit 20 or obtain a concept connected to the concept C by referring to the data of FIG. 4 already created. May be.

Next, in step S50, the path weight calculation unit 14 calculates the weight of the path from concept A to concept B via concept C for each concept (referred to as concept B) in the array of adjacent concepts. When the weight calculated for the path is larger than the weight of the path from the concept A to the concept B that has already been calculated (the value of the weight calculated as MAX_Weight (A, B)), the MAX_Weight ( A, B) is updated, and the path information is recorded as the value of MAX_Path (A, B). The path is obtained by extending the path already recorded in MAX_Path (A, C) to concept B. In other words, the path is obtained by connecting the concept B to the path already recorded in MAX_Path (A, C).
If there is no value already calculated as MAX_Weight (A, B), the value calculated this time is stored as the path weight value MAX_Weight (A, B). The path corresponding to the weight is stored as MAX_Path (A, B).

  When the concept (concept B) extracted from the adjacent concept array is the calculation target concept (concept A) itself in the process of step S50 above, the path weight calculation unit 14 determines the weight of the path. Do not calculate or record the path as MAX_Path.

  Next, in step S60, the number of paths from concept A to concept B is updated for each concept B processed in step S50 above. That is, the value of NUM_Path (A, C) is added to the value of NUM_Path (A, B) so far, and the result of the addition is set as a new value of NUM_Path (A, B). This corresponds to a new search for a path from concept A to concept B via concept C.

  Next, in step S70, the path weight calculation unit 14 determines whether or not an uncalculated concept remains in the starting concept array. If an uncalculated concept remains in the starting concept array (step S70: YES), the path weight calculation unit 14 returns to step S40 to perform calculation for the next starting concept. If no uncalculated concepts remain in the starting concept array (step S70: NO), the path weight calculation unit 14 proceeds to the next step S80.

  Moving to FIG. 9, in step S80, the path weight calculation unit 14 increments the value of D, which is a variable for counting the distance.

  In step S90, the path weight calculation unit 14 determines whether the value of the variable D has reached the parameter value MAX_Distance. Specifically, it is determined whether or not the inequality D <MAX_Distance is satisfied. When D <MAX_Distance, that is, when the distance of the searched path has not yet reached the set maximum value MAX_Distance (step S90: YES), the process proceeds to step S100 and the processing by the path weight calculation unit 14 is performed. to continue. That is, in this case, the path weight calculation unit 14 continues the process of searching for a path for the new value of D incremented in step S80 and calculating the path weight. On the other hand, when D <MAX_Distance is not satisfied, that is, when the distance of the searched path reaches the set maximum value MAX_Distance (step S90: NO), the process proceeds to processing by the relevance calculation unit 16 in step S110.

  When the process proceeds to step S100, the path weight calculation unit 14 writes each concept (concept B) processed in steps S50 and S60 to the starting concept array (see FIG. 6). Then, control is returned to step S40 (FIG. 8), and the process is repeated.

When the process proceeds to step S110, in the same step, the relevance calculation unit 16 calculates the relevance between the concepts based on the path weight and the number of paths calculated in the processing so far. Specifically, the degree-of-relevance calculation unit 16 describes the concept (concept B) having two or more routes that can be reached within the range set in the parameter MAX_Distance from the original concept (concept A) to be calculated. Calculate relevance by formula. Ie;
(1) The degree of association from concept A to concept B is obtained by multiplying the value of MAX_Weight from concept A to concept B by the value of the number of paths NUM_Path from concept A to concept B. That is, the degree of association from concept A to concept B is MAX_Weight (A, B) × NUM_Path (A, B).
(2) However, the relevance calculated in the above (1) is normalized as follows. That is, the degree of association from concept A to concept B after normalization is the degree of association from concept A to concept B calculated by the equation (1) above, with the sum of the degrees of association from concept A to each concept. Shall be divided. That is, the degree of association after normalization is {MAX_Weight (A, B) × NUM_Path (A, B)} / (total sum of degrees of association from concept A to each concept on the concept map). By normalizing in this way, a value independent of the size of the concept map can be calculated.

  It should be noted that the degree of relevance is calculated by limiting to the concept (concept B) that has two or more routes that can be reached within the range set in the parameter MAX_Distance from the original calculation target concept (concept A) in the process of step S110 above. The reason for this is as follows. That is, the concept B that has only one route from the concept A to the concept B has a sufficiently small degree of association with the concept A and can be regarded as having no relation.

  Next, in step S120, the concept processing device 1 determines whether or not there is a concept to be calculated and an uncalculated concept. If there is an uncalculated concept (step S120: YES), the process returns to step S30 (FIG. 8) in order to move to the process for the next concept to be calculated. If there is an uncalculated concept (step S120: NO), the process of the entire flowchart is terminated.

[Example of processing]
Next, an example of processing will be described. In the following description, reference is made to FIG. 11, FIG. 12, FIG. 13, and FIG. These drawings are schematic views showing the state of the memory (work storage unit 22) in the course of the processing examples described below. FIG. 11 shows transition of values held in the storage area described with reference to FIG. 12, 13, and 14 show transitions of values held in the storage area described with reference to FIG. 7.

  The concept map targeted by the processing of this example is the one shown in FIG. The example of the concept map shown in the figure includes seven concepts of “disease”, “hypertension”, “diabetes”, “adult disease”, “cold”, “influenza”, and “fever”. Each of these concepts corresponds to a node in the concept map graph. And the link which connects between nodes respond | corresponds to the relationship between concepts. In the graph illustrated in the figure, there are links of disease-high blood pressure, disease-adult disease, disease-diabetes, disease-cold, disease-flu, adult disease-hypertension, adult disease-diabetes, cold-fever, influenza-fever. include.

A processing example will be described below according to the processing procedure. Note that the step numbers mentioned in the following description of the processing examples are the step numbers in the procedure shown in the flowcharts of FIGS.
Step S10: First, the parameter setting unit 18 of the conceptual processing device 1 sets the value of the parameter MAX_Distance to 2 in this processing example.

  Step S20: Next, the concept weight calculation unit 12 creates a list that enumerates directly connected concepts for each concept included in the concept map, and calculates a weight.

  FIG. 10 is a schematic diagram illustrating an example of a table that holds information about each concept created in step S20. As shown in the figure, this table has a tabular structure, and each row corresponds to a concept (node) included in the concept map. Each digit corresponds to a data item (attribute), and there are each digit of node ID, concept (node), connected concept (node), and weight.

The node ID is identification information given for convenience to each concept. In the illustrated example, numerical values from 1 to 7 are assigned as node IDs.
A concept (node) is a word representing the concept. In this example, words such as “disease” and “adult disease” are stored in this column.
The connected concept (node) is a list of previous concepts (nodes) connected by a direct link from the concept. For example, there are five nodes connected by a direct link from the node “disease”, which are “high blood pressure”, “adult disease”, “diabetes”, “cold”, “influenza”. In the connected concept column for the concept “disease”, a list in which these five linked concepts are listed is stored.
The weight stores the value of the weight for the concept (node). Since the total number of concepts appearing in this concept map is 7, for example, the weight for the concept “disease” (the number of concepts directly connected to is 5) is log (7 / 5), that is, 0.15 (rounded off to the third decimal place). Also, for example, the weight for the concept “high blood pressure” (the number of concepts directly connected is 5) is similarly log (7/2), that is, 0.54. The same applies to the weights for other concepts.

[D = 0, concept of starting point is “disease”]
Step S30: Next, the path weight calculation unit 14 selects one concept to be calculated, and stores the concept in the area of the concept to be calculated in the memory. The concept processing device 1 initializes the value of the number of concepts traced as the distance from the starting point. That is, the value D = 0 is written in the area of the number of concepts traced in the memory. Further, the concept processing apparatus 1 stores only the concept of the calculation target selected here as an element of the array of the starting concept in the memory.
Step S40: Next, the path weight calculation unit 14 sets the concept selected from the starting concept array in the memory as the starting concept. Then, the concept processing device 1 writes the selected concept in the concept area of the starting point in the memory. At present, only the “illness” element is written in the starting concept array, so the concept processing apparatus 1 writes “disease” in the starting concept area in the memory. Furthermore, the concept processing device 1 writes all the concepts directly connected from the concept of the starting point to the adjacent concept array. Specifically, there are five concepts that are directly connected to “disease”: “high blood pressure”, “adult disease”, “diabetes”, “cold”, and “influenza”. Is written in the area of the array of adjacent concepts.
The state in the memory at the time when the processes of steps S30 and S40 are completed is as shown in the row a of FIG.

  Next, the path weight calculation unit 14 sequentially performs the processes of step S50 and step S60 as follows for each concept stored in the array of adjacent concepts. That is, the concept processing apparatus 1 stores each concept stored in the starting concept array in FIG. 11 as the starting concept, and stores it in the concept to be calculated (in the current situation, “disease”) and the adjacent concept array. Calculate the relevance between concepts that are elements. The relevance is calculated as a harmonic average of the weights of the concepts (including the concepts at both ends of the path) on the currently traced path. Also, if the relevance calculated here is greater than the relevance already calculated, the information (table shown in FIG. 7) stored in the work storage unit 22 is updated. Since this is the first time for calculating the degree, the path weight (MAX_Weight for each concept) is updated for each concept. Further, when updating the weight of this path, the path when the degree of association is calculated is also recorded (MAX_Path for each concept is updated).

(1) First, “hypertension” is extracted as concept B from the array of adjacent concepts.
Step S50: The degree of association between the concept “disease” to be calculated and the concept (concept B) “hypertension” which is an element stored in the array of adjacent concepts is 0.15 which is the weight of “disease”. (Refer to FIG. 10, and the same applies to the following) and the harmonic average of 0.54 which is the weight of “high blood pressure”. The value is 0.23 (rounded off to the third decimal place. The same applies hereinafter). Further, the path corresponding to this degree of association is a path of disease-hypertension. That is, regarding the concept “high blood pressure”, MAX_Weight = 0.23 and MAX_Path = “disease, high blood pressure”.
Step S60: Here, the number of paths (NUM_Path) is updated. It is special only when D = 0, and here, NUM_Path (A, C) = NUM_Path (disease, illness) = 1 is calculated. That is, for the concept “hypertension”, the value of NUM_Path is updated to 1.

(2) Next, “adult disease” is extracted as concept B from the sequence of adjacent concepts.
Step S50: The degree of association between the calculation target concept “disease” and “adult disease”, which is concept B selected from the array of adjacent concepts, is 0.15 which is the weight of “disease”, and “adult disease” As a harmonic average of 0.37, which is a weight of "." The value is 0.21. The path corresponding to this degree of association is a path of disease-adult disease. That is, regarding the concept “adult disease”, MAX_Weight = 0.21 and MAX_Path = “disease, adult disease”.
Step S60: As in the case of “hypertension”, only D = 0 is special, and NUM_Path (A, C) = NUM_Path (disease, illness) = 1 is calculated. That is, the value of NUM_Path is also updated to 1 for “adult disease”.

(3) Next, “diabetes” is extracted as concept B from the array of adjacent concepts.
Step S50: The degree of association between the concept “disease” to be calculated and “diabetes”, which is the concept B selected from the array of adjacent concepts, is 0.15 which is the weight of “disease”, and “diabetes”. It is obtained as a harmonic average with a weight of 0.54. The value is 0.23. The path corresponding to the degree of association is a path of disease-diabetes. That is, for the concept “diabetes”, MAX_Weight = 0.23 and MAX_Path = “disease, diabetes”.
Step S60: As in the case of “hypertension”, only D = 0 is special, and NUM_Path (A, C) = NUM_Path (disease, illness) = 1 is calculated. That is, for “diabetes”, the value of NUM_Path is updated to 1.

(4) Next, “cold” is extracted as concept B from the array of adjacent concepts.
Step S50: The degree of association between the concept “disease” to be calculated and “cold”, which is the concept B selected from the array of adjacent concepts, is 0.15 which is the weight of “disease” and “cold” It is obtained as a harmonic average with a weight of 0.54. The value is 0.23. Further, the path corresponding to this degree of association is a path of disease-cold. That is, regarding “cold”, MAX_Weight = 0.23 and MAX_Path = “disease, cold”.
Step S60: As in the case of “hypertension”, only D = 0 is special, and NUM_Path (A, C) = NUM_Path (disease, illness) = 1 is calculated. That is, the value of NUM_Path is also updated to 1 for “cold”.

(5) Next, “influenza” is extracted as concept B from the array of adjacent concepts.
Step S50: The degree of association between the calculation target concept “disease” and “influenza” that is the concept B selected from the array of adjacent concepts is 0.15 which is the weight of “disease”, and “influenza” It is obtained as a harmonic average with a weight of 0.54. The value is 0.23. A path corresponding to this degree of association is a path of disease-influenza. That is, regarding “influenza”, MAX_Weight = 0.23 and MAX_Path = “disease, influenza”.
Step S60: As in the case of “hypertension”, only D = 0 is special, and NUM_Path (A, C) = NUM_Path (disease, illness) = 1 is calculated. That is, the value of NUM_Path is also updated to 1 for “influenza”.

As described above, as D = 0, the degree of association between the “disease” that is the concept to be calculated and the concept being traced (each concept that was an element stored in the array of adjacent concepts) is calculated. It was.
FIG. 12 is a schematic diagram showing the state of a table (the table shown in FIG. 7) that stores calculation results for each concept at this stage. As shown in the figure, MAX_Weight for “high blood pressure” in the first row, “adult disease” in the second row, “diabetes” in the third row, “cold” in the fourth row, “flu” in the fifth row And MAX_Path and NUM_Path values are stored.

Step S80: Next, D is incremented and D = 1.
Step S90: Since MAX_Distance = 2 is set, the condition of D <MAX_Distance is satisfied. Therefore, the process proceeds to step S100.

  Step S100: The path weight calculation unit 14 stores each concept calculated when the loop of steps S50, S60, and S70 is repeated in an array of starting concepts. That is, as shown in FIG. 11 b, the elements of “high blood pressure”, “adult disease”, “diabetes”, “cold”, and “influenza” are stored in the starting point concept array. Next, the process returns to step S40.

  Step S40: The path weight calculation unit 14 first selects the concept “high blood pressure” from the starting point concept array, and sets this “high blood pressure” as the starting point concept (concept C). Then, all the previous concepts connected to the concept C (concepts adjacent to the concept C) are stored in the adjacent concept array. Specifically, each element of “adult disease” and “disease” is stored in the array of adjacent concepts. This is the state shown in FIG.

Steps S50 and S60 are performed for “adult disease” in the adjacent concept array.
Step S50: The path weight calculation unit 14 calculates a weight for the path “disease-hypertension-adult disease” that has been followed. It is a harmonic average of 0.15 (weight of concept “disease”), 0.54 (weight of concept “hypertension”) and 0.37 (weight of concept “adult disease”), and its value is 0.27. is there.
Since this value of 0.27 is larger (heavy) than 0.21 (refer to the second row in FIG. 12), which is a weight from “illness” to “adult disease” calculated so far, MAX_Weight ( A, B) are updated. In addition, MAX_Path is updated with a path corresponding to this new weight of 0.27. That is, MAX_Weight = 0.27 and MAX_Path = “disease, hypertension, adult disease”.
FIG. 13 is a schematic diagram showing the state of the table (the table shown in FIG. 7) at this time. As shown in the figure, MAX_Weight and MAX_Path are updated for the concept “adult disease”.

Step S60: The path weight calculation unit 14 updates the value of NUM_Path (number of paths) so that NUM_Path = 2. That is, NUM_Path (disease → adult disease) at D = 1 is the sum of the value of NUM_Path (disease → adult disease) at D = 0 and the value of NUM_Path (disease → hypertension) at D = 0. That is, NUM_Path (A, C) at D = 0 is added to NUM_Path (A, B) at D = 0.
FIG. 14 is a schematic diagram showing the state of the table (the table shown in FIG. 7) at this time. As shown in the figure, NUM_Path is also updated for the concept “adult disease”.

  Since the concept “disease” is the concept of the calculation target itself, the calculation is not performed for the path “disease-hypertension-disease”.

  Hereinafter, although detailed description is omitted, the calculation procedure of steps S50 and S60 is repeated for the concept stored in the starting concept array and the weight calculation is not yet completed. That is, the calculation procedure of steps S50 and S60 is repeated for each concept of “adult disease”, “diabetes”, “cold”, and “flu”. In this process, MAX_Path and NUM_Path at the time of D = 1 from the concept “disease” to each of the above concepts are obtained.

Step S80: Next, D is incremented and D = 2.
Step S90: Since MAX_Distance = 2 is set, the condition of D <MAX_Distance is not satisfied. Therefore, the process proceeds to step S110.

Step S110: The relevance calculation unit 16 calculates the relevance from the concept “disease” to each concept by MAX_Weight × NUM_Path and normalizes the value. However, this calculation is limited to a concept having two or more routes from the concept “disease”.
Then, the concept processing device 1 outputs the degree of association (normalized) between the concept “disease” and each concept calculated by the above multiplication MAX_Weight × NUM_Path.
This completes the processing of the concept “disease” to be calculated.
The relevance calculation unit 16 writes the calculated relevance information into the relevance storage unit 30.

  Step S120: It is determined whether there is an uncalculated calculation target concept (that is, a calculation target concept other than the above-mentioned “disease”). If there is, the process moves to step S30 and continues. If not, the processing of the entire flowchart shown in FIGS. 8 and 9 is terminated.

  The relevance calculated by the method described above is directed. In other words, the calculated relevance value depends on the direction from concept to concept. For example, the value of the degree of association from the concept “disease” to the concept “cold” is different from the value of the degree of association from the concept “cold” to the concept “disease”.

As described above, according to the concept processing device 1 of the first embodiment, the degree of association from one concept to another concept can be calculated from the data of the concept map. At this time, the concept processing device can calculate the above-mentioned relevance from only the data of the concept map without requiring data such as language resources made up of actual sentence examples.
Moreover, according to the conceptual processing apparatus 1 of 1st Embodiment, said relevance degree can be calculated with the limited calculation amount realistically appropriate. One of the factors is that an upper limit (MAX_Distance) of the distance when searching the concept map is provided.
Further, the degree of association between concepts calculated by the concept processing device 1 of the first embodiment is calculated as a good value when applied to an actual concept map. This is because the processing procedure and calculation method for calculation are particularly devised.

  Further, according to the technique described as the first embodiment, for a wide concept (for example, the concept “Japanese sports”, etc.), the number of concepts that are adjacent to each other is large, so the concept weight is reduced. Therefore, the relevance value related to such a broad concept becomes small. In addition, for concepts that are too narrow (for example, “NHK Sunday Sports” representing a specific program by a specific broadcasting station), the total number of paths that include such a concept is reduced. The value is still small. In other words, the two extreme concepts have the effect of reducing the degree of association. That is, a low relevance level can be calculated for noise. That is, processing such as information retrieval can be performed while reducing the influence of noise in the concept.

[Second Embodiment]
Next, a second embodiment of the present invention will be described. Note that description of matters common to the above-described embodiment may be omitted, and description will be made focusing on matters specific to the present embodiment.
FIG. 15 is a block diagram showing a schematic functional configuration of the conceptual processing apparatus according to the present embodiment. As illustrated, the concept processing device 2 includes a concept weight calculation unit 112, a path weight calculation unit 114, a relevance calculation unit 116, a parameter setting unit 18, a concept map storage unit 20, and a work storage unit 22. The relevance storage unit 30 and the similarity acquisition unit 140 are included.
The feature of the present embodiment is that the similarity acquisition unit 140 is provided, and the weight calculation method and the relevance calculation method are different from the first embodiment.

  In the concept processing device 2 according to the present embodiment, the similarity acquisition unit 140 acquires similarity data representing the similarity between concepts from the outside. Then, the path weight calculation unit 114 calculates the path weight while using the similarity between concepts by referring to the similarity data acquired by the similarity acquisition unit 140. That is, the path weight calculation unit 114 uses a similarity value given from the outside as an initial value when obtaining the weight of a path that traces the concept map stored in the concept map storage unit 20. In other words, the path weight calculation unit 114 acquires, for the path, a similarity degree regarding a concept pair connected by a link included in the path from the similarity data, and the path weight according to the acquired similarity degree. Is calculated.

  FIG. 16 is a schematic diagram illustrating a configuration of similarity data acquired by the similarity acquisition unit 140. As shown in the figure, the similarity data is tabular data, and has items of concept 1, concept 2, and similarity as items. Concept 1 and concept 2 correspond to the concepts included in the concept map, respectively. The data item “similarity” holds a value of similarity between concepts 1 and 2. As long as the similarity data includes the value of similarity between concepts as information, the data format and the like may be different from the illustrated data.

  The similarity value is supplied from the outside, and the similarity between concepts can be given by using various methods. An example of similarity between concepts is context similarity. The method for obtaining the context similarity is an existing technology. In order to obtain the context similarity, a language resource consisting of a large amount of sentences is acquired, a syntax analysis of each sentence included in the language resource is performed, and a statistical analysis is performed on the context in which the word appears. Here, a word corresponds to a concept. In addition, the context here refers to the statistical characteristics of concepts (words) around a concept (word) when a concept (word) appears in a sentence included in a language resource, or the appearance of a concept (word). Such as positioning in the syntax. A statistical feature of surrounding concepts (words) is represented by a word vector based on the surrounding concepts (words) as an example. This word vector is a multi-dimensional vector (for example, tens of thousands or hundreds of thousands of dimensions) in which each element corresponds to a word, and each element represents whether or not the word appears in the context. The concepts (words) that tend to appear in similar contexts have higher context similarities as the tendency is stronger. This is based on the premise that a certain concept (word) is characterized by the surrounding concept or syntax.

The method for obtaining the weight according to the present embodiment is as follows.
The concept weight calculator 112 calculates the concept weight according to the following equation.
(Concept weight) = log (N / n)
Here, n is the number of links directly connected from the concept to another concept. N is the total number of concepts included in the concept map.

Then, the path weight calculation unit 114 calculates the weight of the path between the adjacent concepts A and B by any of the following formulas (A1), (A2), and (A3).
Formula (A1): (weight of path between adjacent concepts) = sim (A, B) × (weight of concept A + weight of concept B)
Formula (A2): (weight of path between adjacent concepts) = sim (A, B) × (weight of concept A × weight of concept B)
Expression (A3): (weight of path between adjacent concepts) = sim (A, B) × min (weight of concept A, weight of concept B)
Note that the path weight calculation unit 114 uses any one of the above formulas (A1), (A2), and (A3), and uses one of these formulas consistently. That is, when calculating the degree of association between concepts, the formula for calculating the path weight between concepts is not changed midway.

Here, sim (A, B) is the similarity between the concept A and the concept B, and the value can be referred to by the similarity data acquired by the similarity acquisition unit 140.
Further, min (weight of concept A, weight of concept B) is obtained by applying a minimum value function that returns a minimum value among arguments to each weight. That is, min (concept A weight, concept B weight) returns the smaller one of the concept A weight and the concept B weight.

  When calculating the degree of association from one concept (A) to another concept (B), the path weight calculation unit 114 determines that the above formulas (A1) and (A2) ) Or (A3), the value obtained as it is is used as the weight.

Further, when calculating the weight from a certain concept (A) to another concept (B) that is not directly connected (not adjacent), the path weight calculation unit 114 performs the following calculation. That is, the harmonic average (B1) or geometric average (B2) of the weights (any one of the formulas (A1), (A2), and (A3)) for each link included in the path connecting these two concepts is calculated. .
For example, the weight between concepts having a distance of 2 is calculated by either of the following formulas (B1) and (B2).

Formula (B1): (weight from concept A to B) = (2 × weight from concept A to C × weight from concept C to B) / (weight from concept A to C + concept from concept C to B weight)
Formula (B2): (weight from concept A to B) = SQRT (weight from concept A to C × weight from concept C to B)
The path weight calculation unit 114 uses any one of the above formulas (B1) and (B2), but uses one of these formulas consistently. That is, when calculating the degree of association between concepts, the weight calculation formula (either harmonic average or geometric average) is not changed midway.

Here, concepts A to C are connected by a direct link, and concepts C to B are connected by a direct link. That is, the distance between the concepts A and B is 2, and the concept C is interposed therebetween.
Further, SQRT () in the above formula (B2) is a function that returns the square root of the argument.
Even when the distance from the concept A to the concept B is larger than 2, the path weight calculation unit 114 extends the above formula (B1) or (B2), and calculates the path weight by the harmonic mean or the geometric mean. To do.

The degree-of-association calculation unit 116 obtains the degree of association from the concept A to B by either of the following methods (C1) or (C2).
(C1) The path weight from the concept A to B calculated by the path weight calculation unit 114 is output as it is as the degree of association from the concept A to B (written in the degree of association storage unit 30). When there are a plurality of paths from concept A to B, a representative path is selected by one of the following methods.
(C1-1) The shortest path is adopted.
(C1-2) A path having the largest weight (path weight) is adopted.

(C2) Multiplying the maximum weight (path weight) of the path weights from concept A to B by the number of paths from concept A to B, and the result is the degree of association from concept A to B Is output (written in the association degree storage unit 30). When the method (C2) is adopted, the processing procedure (procedure for searching a concept map) described in the flowchart in the first embodiment can also be used.
The relevance calculation unit 116 uses any one of the above methods (C1) and (C2) (in the case of (C1), any one of the methods (C1-1) and (C1-2). However, the selected method is used consistently. That is, when calculating the degree of association between concepts, the method for obtaining the degree of association is not changed midway.

  According to the concept processing device 2 of the second embodiment, data on the degree of similarity between concepts acquired from the outside is used, and between the concepts according to the concept weight, the path weight, or the like (or both). Relevance can be calculated. In the case where data of similarity is already provided from the outside, it is possible to obtain a degree of relevance that well reflects the data.

[Third Embodiment]
Next, a third embodiment of the present invention will be described. Note that description of matters common to the above-described embodiment may be omitted, and description will be made focusing on matters specific to the present embodiment.
The block diagram showing the functional configuration of the conceptual processing apparatus according to this embodiment is the same as the block diagram shown in FIG.
The concept processing device of the first embodiment multiplies the maximum value of path weights between concepts (the weight of the path having the maximum weight among a plurality of paths) and the number of paths between the concepts. Was used to calculate the degree of association between the concepts. On the other hand, in this embodiment, when there are a plurality of paths from concept A to concept B, the relevance is obtained based on the sum of the weights of the paths.

Specifically, it is as follows.
The concept weight calculator 12 of this embodiment calculates the weight of each concept by the same method as in the first embodiment.
Next, the path weight calculation unit 14 calculates path weights for all paths between the concept A and the concept B, respectively. As with the first embodiment, the weight itself of a certain path is a harmonic average of the weights of all the concepts (including the concepts at both ends) included in the path. Note that the path between the concept A and the concept B, which is a calculation target here, is a path that does not pass the same concept more than once in the middle. That is, it does not include a closed circuit in a part of the path or a backtrack in the middle of the path.
Note that when the size of the concept map is finite (the number of concepts included in the concept map is finite), the number of paths between the concept A and the concept B is finite.
In addition, regarding the path between the concept A and the concept B to be calculated here, the length may be limited. That is, at this time, a path that cannot be reached by the number of links of the parameter MAX_Distance set by the parameter setting unit 18 is excluded from the calculation.

  Then, the relevance calculation unit 16 obtains the sum of the weights of the paths between the concept A and the concept B, and outputs the sum as the relevance from the concept A to the concept B. Specifically, the relevance level calculation unit 16 writes the calculated relevance level in the relevance level storage unit 30.

  According to the third embodiment, the path between the concept A and the concept B is not represented by the path having the maximum weight, but the degree of association reflecting the weight of each path can be obtained. When the size of the concept map is large, a large amount of calculation may be required for the process of searching for a long distance. By introducing parallel processing as necessary, a large amount of calculation can be processed in a shorter time.

  In addition, you may make it implement | achieve part or all of the function of the conceptual processing apparatus in each embodiment mentioned above with a computer and a program. In that case, a program for realizing each function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible disk, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, a “computer-readable recording medium” dynamically holds a program for a short time, like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory inside a computer system serving as a server or a client in that case may be included and a program held for a certain period of time. The program may be a program for realizing a part of the above-described functions, or may be a program that can realize the above-described functions in combination with a program already recorded in a computer system.

  As described above, a plurality of embodiments of the present invention have been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes a design and the like within a scope not departing from the gist of the present invention. .

  The degree of association between the concept “disease” and another concept was calculated by the method described in the first embodiment using an actual concept map (separate from the simplified map illustrated in the drawing). . As a result, the concepts up to the top 10 in the value of relevance are as follows. Here, numerical values up to the fifth decimal place are obtained.

1st place Relevance: 0.00358 Concept: body 2nd place Relevance: 0.00325 Concept: inflammation 3rd place Relevance: 0.00314 Concept: medicine 4th place Relevance: 0.00309 Concept: cold 5th place Relevance: 0 .00307 Concept: Body 6th Relevance: 0.00300 Concept: Infectious Disease 7th Relevance: 0.00296 Concept: Cancer 8th Relevance: 0.00278 Concept: Infection 9th Relevance: 0.00267 Concept: Allergy 10th place Relevance: 0.00258 Concept: Cancer

  The present invention can be used for various types of information processing using the degree of association between concepts. As an example, the present invention can be used in fields such as information retrieval and program recommendation for recommending content such as television programs to users.

1, 2 Concept processor 12, 112 Concept weight calculator 14, 114 Path weight calculator 16, 116 Relevance calculator 18 Parameter setting unit 20 Concept map storage unit 22 Work storage unit 30 Relevance storage unit 140 Similarity acquisition unit

Claims (6)

Reads concept map data that indicates the presence or absence of relationships between concepts, and calculates the value of the concept weight so that the smaller the number of links that represent the relationship between one concept and the other, the greater the value for that concept A concept weight calculation unit for
In the concept map data, for a path formed by connecting a concept and another concept with the link, the weight value of the path corresponding to the weight value of the concept for the concept included on the path A path weight calculation unit for calculating
A degree-of-association calculating unit that calculates a degree of association between the concept and the other concept according to the value of the weight of the path for a path connecting a certain concept and another concept;
A conceptual processing apparatus comprising: The path weight calculation unit calculates a value of the weight of the path by taking a harmonic average or a geometric average of the values of the weight of the concept for each concept included on the path;
The conceptual processing device according to claim 1, wherein: The concept weight calculation unit calculates a value of the concept weight by calculating a logarithm of the reciprocal of the number of links between the concept and the other concept.
The conceptual processing apparatus according to claim 1, wherein When there are a plurality of paths formed by connecting the certain concept and the other concept by the link, the relevance calculation unit calculates the weight value of the plurality of paths as the number of the plurality of paths. By calculating the relevance between the concept and the other concept by multiplying the value of the maximum weight of the path among the above, or calculating the sum of the weights of the plurality of paths as the relevance,
The conceptual processing device according to claim 1, wherein the conceptual processing device is a device. A similarity acquisition unit that acquires similarity data representing the similarity between concepts,
The path weight calculation unit acquires, for the path, a degree of similarity related to a concept pair connected by a link included in the path from the similarity data, and the path weight value according to the acquired degree of similarity To calculate,
The conceptual processing device according to claim 1, wherein the conceptual processing device is a device.   The program for functioning a computer as a conceptual processing apparatus as described in any one of Claim 1-5.

Images