JP2012008680A - Program and information analyzing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent a node that is only infrequently connected to other nodes from tightening in link to any other node.SOLUTION: An information analyzing device 10 acquires linking information indicating directed links set between a plurality of nodes, updates the acquired linking information by adding to a plurality of nodes having directed links, virtual nodes that are connected from individual ones of a plurality of nodes and connected to individual ones of a plurality of nodes, selects nodes whose number of connections to other nodes are less than a first value out of a plurality of nodes, sets links to the virtual nodes until the numbers of connections of the selected nodes reach the first value, and thereby further updates the linking information. The information analyzing device 10 makes bidirectional the links indicated by the updated linking information set in each of the plurality of nodes and the virtual nodes to further update the linking information, and calculates the weight of each link indicated by the further updated linking information.

Description

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

  There are directed connections such as citations and references between documents such as papers, patent documents, web pages, etc., and network information using these documents as data elements (nodes) and connection relationships as links is significant. Getting information is done. For example, the following Non-Patent Document 1 describes a technique for calculating an evaluation value of each document so that a document collecting more citations has a higher value.

Lawrence Page, Sergey Brin, Rajeev Motwani, Terry Winograd, “The PageRank Citation Ranking: Bringing Order to the Web”, [online], January 26, 1998, Internet <URL: http://www-db.stanford.edu /~backrub/pageranksub.ps>

  In a network composed of documents that cite past documents, the evaluation value flows to the past document through the link. Therefore, it is possible to calculate the evaluation value of each document by virtually bidirectionalizing the link. Conceivable. At this time, a technique has been proposed in which the number of citations of each document is calculated so as to be equal to a specified value so that a document with a small number of citations but a large number of citations is not unreasonably high. However, even in such a case, if a document with a small number of citations cites a document with a high evaluation, an evaluation value higher than the value of the actual document may be calculated.

  An object of the present invention is to connect a data element having a small number of connections to other data elements when calculating the evaluation value of each data element by bidirectionalizing the connection relation between data elements having a directed connection relation. It is an object of the present invention to provide a program and an information analysis apparatus that suppress an excessively strong connection with a previous data element.

  In order to achieve the above object, the invention according to claim 1 is directed to a computer, an acquisition means for acquiring connection information indicating a directed connection relationship set between a plurality of data elements, and a directed connection relationship. Acquired by the acquisition means by adding a virtual data element connected to each data element of the plurality of data elements to a plurality of data elements having A first updating means for updating connection information; and a data element in which the number of connections to another data element is less than a predetermined first value among the plurality of data elements, and the selected data element Second updating means for setting connection relations to the virtual data elements until the number of connections reaches the first value and updating connection information updated by the first updating means; and the plurality of data The connection relation indicated in the connection information updated by the second update means set in each of the element and the virtual data element is bidirectionalized, and the connection information updated by the second update means is updated. 3 update means and a program for functioning as calculation means for calculating the weight of each connection relation indicated in the connection information updated by the third update means.

  In the invention according to claim 2, the computer further includes an adjacency matrix indicating a connection relationship set for each of the plurality of data elements and the virtual data element, and the number of connections from each data element is predetermined. The generation means for generating the second matrix so as to match the second value, and the calculation means updates the adjacency matrix generated by the generation means by the third update means, and then updates the updated adjacency matrix. The program according to claim 1, wherein the weight of connection relations from each data element is calculated to be 1 in total.

  In the invention according to claim 3, the first value is calculated by dividing the number of data elements of the plurality of data elements by the number of connections indicated in the connection information acquired by the acquisition unit. 3. The program according to claim 1, wherein the program is determined based on an average number of connections.

  According to a fourth aspect of the present invention, the computer further updates the evaluation value for each of the plurality of data elements according to the weight of each connection relation calculated by the calculation means, in a predetermined end. The program according to any one of claims 1 to 3, wherein the program is made to function as an updating unit that is repeatedly executed until a condition is satisfied.

  In addition, the invention according to claim 5 includes an acquisition unit that acquires connection information indicating a directed connection relationship set between a plurality of data elements, and a plurality of data elements having a directed connection relationship. First update means for updating connection information acquired by the acquisition means by adding a virtual data element connected from each data element of the plurality of data elements and connected to each data element of the plurality of data elements And selecting a data element in which the number of connections to other data elements is less than a predetermined first value among the plurality of data elements, and the number of connections of the selected data element is the first value A second update unit that sets connection relations to the virtual data elements until reaching the value and updates connection information updated by the additional update unit, and each of the plurality of data elements and the virtual data elements Third update means for updating the connection information updated by the second update means by bidirectionalizing the set connection relation indicated by the connection information updated by the second update means; and the third update Calculating means for calculating the weight of each connection relationship indicated in the connection information updated by the means.

  According to the first and fifth aspects of the present invention, when the evaluation value of each data element is calculated by bidirectionalizing the connection relation between data elements having a directed connection relation, the connection to other data elements It can be suppressed that the data elements with a small number are excessively connected to the data element of the connection destination.

  According to the second aspect of the present invention, when the weight of the connection relation between the data elements is expressed by the transition probability from the data element to the data element, the data element having a small number of connections to the other data elements is the connection destination. It is possible to prevent the connection with the data element from becoming excessively strong.

  According to the third aspect of the present invention, it is possible to prevent the data element that is less than the average number of connections in the entire data element from being excessively strongly linked to the connection destination data element.

  According to the fourth aspect of the present invention, when a data element having a small number of connections to other data elements is connected to a highly evaluated data element, it is possible to suppress obtaining an excessively high evaluation value.

It is a functional block diagram of the information analysis device concerning this embodiment. It is a figure which shows an example of a network. It is a figure which shows an example of the network expanded by adding a virtual node to a node group. It is a figure which shows an example of the network which the bidirectional | two-way adjacency matrix represents. It is a figure which shows the example (A) at the time of making the link between nodes simply bidirectional, and the example (B) at the time of bidirectionalizing by the method used by this embodiment. It is a flowchart which shows an example of a calculation process of the evaluation value of a node.

  Hereinafter, embodiments for carrying out the present invention (hereinafter referred to as embodiments) will be described with reference to the drawings.

  FIG. 1 shows a functional block diagram of an information analysis apparatus 10 according to the present embodiment. As illustrated in FIG. 1, the information analysis apparatus 10 includes a node data acquisition unit 100, a virtual node addition setting unit 102, an adjacency matrix generation unit 104, a regularization processing unit 106, a bidirectionalization processing unit 108, and a transition probability matrix generation. Section 110, evaluation value update section 112, and processing result holding section 114. The functions of the above-described units are such that a computer having a control unit such as a CPU, a storage unit such as a memory, and an input / output unit that transmits and receives data to and from an external device reads a program stored in a computer-readable information storage medium. It may be realized by executing. The program may be supplied to a computer functioning as the information analysis apparatus 10 by an information storage medium, or may be supplied to the computer via data communication means such as the Internet.

  The node data acquisition unit 100 acquires each data of a plurality of nodes (data elements) and acquires connection information indicating a link destination (connection destination) of each node. Note that a node (data element) in the present embodiment corresponds to a node (vertex) constituting a graph in graph theory, and a link (connection relationship) corresponds to an edge (edge) connecting nodes. In this embodiment, an example in which one data file such as an electronic document is associated with one node will be described. For example, when the information analysis apparatus 10 analyzes a citation network between documents, the document may be a node and the citation relationship between documents may be a link. In that case, the node data acquisition unit 100 may acquire a document by accessing a database storing the document, and information on a document (quoted document) cited by the acquired document is acquired from the content of the acquired document. Alternatively, it may be obtained by inquiring a database storing documents. For example, when the document is a patent document, the cited document may be specified from the column of the prior art document in the patent document, or the examination information is obtained based on the identification information of the patent document, and the obtained examination information is obtained. The cited document may be specified based on the above. When the document is a scientific and technical paper, the cited document may be extracted from the citation column of the scientific and technical paper.

FIG. 2 is a diagram illustrating an example of a network in which each node is connected according to the connection relation based on the node data acquired by the node data acquiring unit 100. In FIG. 2, one node corresponds to one document, and an arrow connecting from the node i (n _i ) to the node j (n _j ) represents a relationship in which the node i cites the node j.

  The virtual node addition setting unit 102 adds a virtual node to the node group acquired by the node data acquisition unit 100 and sets a link between the added virtual node and the node group. A virtual node is a node that is cited by all nodes in the node group and all nodes in the node group. Here, the virtual node addition setting unit 102 selects a virtual node until the number of citations reaches a specified value for a node in which the number of citations of other nodes is less than the specified value among the nodes included in the node group. Set the node connection information to quote. The specified value of the number of citations may be a given constant, an average number of citations in the node group, or the maximum value of the number of citations in the node group. If the average number of citations in the node group is not an integer, the average number of citations may be rounded, rounded down, or rounded up to an integer. The virtual node addition setting unit 102 adds a row and a column of the virtual node to the adjacency matrix indicating the connection relation of the node group acquired by the node data acquiring unit 100, and the virtual node, each node of the node group, and The adjacency matrix is updated by setting the connection relationship between the nodes, and among the nodes included in the node group, a node whose number of connections to other nodes is less than a predetermined first value is selected. The connection relationship to the virtual data element may be set until the number of connections of the selected node reaches the first value, and the updated adjacency matrix may be further updated.

  FIG. 3 shows an example of a network expanded by adding a virtual node to the node group acquired by the node data acquisition unit 100. As shown in FIG. 3, the virtual node has at least one bidirectional link with all the nodes of the node group. Among these, a node with a small number of citations is provided with a plurality of links to the virtual node. For example, if the specified value of the citation count is 3 and the citation count of the node i is 2, one citation link is set from the node i to the virtual node, and the citation count of the node j is 1. In this case, two citation links are set from the node j to the virtual node.

  The adjacency matrix generation unit 104 adds a virtual node to the node group acquired by the node data acquisition unit 100 by the virtual node addition setting unit 102 and sets a link to the virtual node and indicates an extended network. A matrix is generated.

Specifically, the adjacency matrix generation unit 104 represents the component A _ij of the adjacency matrix A by the following formula (1) when the node j (n _j ) cites the node i (n _i ).
A _ij = 1 (1)
Here, i and j are natural numbers of 1 to N (N is the number of nodes including virtual nodes). However, if node j cites node i L times, A _ij = L. When node j does not cite node i, A _ij is expressed by the following equation (2).
A _ij = 0 (2)
Since each node does not quote itself, the diagonal component of the adjacency matrix A is expressed by the following equation (3).
A _ii = 0 (3)
In the present embodiment, the adjacency matrix A is generated by the above processing.

Here, the number of _outlinks (number of citations) k _out (j) of the node j is the following expression (4), and the number of inlinks (number of citations) k _in (j) of the node j is the following expression (5). It is represented by

また、ｉ＝ｊのときには以下の式（７）となる。

The regularization processing unit 106 performs processing (normalization) for aligning the number of outlinks of each node to m (m is a natural number). Specifically, the normalization processing unit has a number of citations equal to or greater than a specified value in all nodes, so that k _out (j) ≠ 0 when i ≠ j, and the adjacency matrix A is expressed by the following formula ( 6) to generate a normalized adjacency matrix A (tilde).

When i = j, the following equation (7) is obtained.

また、ｉ＝ｊのときには、双方向化された隣接行列Ａ（オーバーライン）は以下の式（９）により表される。

なお、双方向化された隣接行列Ａ（オーバーライン）は、式（１０）に示されるように対称行列である。

The bi-directional processing unit 108 bi-directionalizes links in the normalized adjacency matrix A (tilde) and updates the adjacency matrix. The bi-directional processing unit 108 generates a bi-directional adjacency matrix A (overline) using the normalized adjacency matrix A (tilde) by the following equation (8) when i ≠ j. .

When i = j, the bidirectional adjacency matrix A (overline) is expressed by the following equation (9).

Note that the bidirectionalized adjacency matrix A (overline) is a symmetric matrix as shown in Equation (10).

ただし、κ（ｊ）は以下の式（１２）により表される。

Here, the number of outlinks k (j) of the node j based on the bidirectionalized adjacency matrix A (overline) is obtained by the following equation (11).

However, κ (j) is expressed by the following equation (12).

  FIG. 4 shows an example of a network represented by a bidirectional adjacency matrix. In FIG. 4, a solid line link from the node j indicates a link originally set between nodes, and a dotted line link indicates a link newly set by bidirectional processing. Note that the first item of the number of outlinks k (j) of the node j is a term based on the solid line link, and the second item is a term based on the dotted line link added by the bidirectional processing.

なお、一般にＴ_ｉｊは対称行列ではない。式（１３）を式（８）〜（１２）に基づいて展開すると、式（１４）のようになる。

式（１４）からも明らかなようにＴ_ｉｊはｍに依存しない。 The transition probability matrix generation unit 110 normalizes the bi-directional adjacency matrix so that the sum of the weights of the outlinks at each node is 1 (that is, the link weights represent the probabilities). is there. Specifically, the transition probability matrix generation unit 110 generates a transition probability matrix T _ij by the following equation (13) based on the bidirectional adjacency matrix.

In general, T _ij is not a symmetric matrix. When Expression (13) is expanded based on Expressions (8) to (12), Expression (14) is obtained.

As is clear from equation (14), T _ij does not depend on m.

式（１４）から、ノードｊのノードｉに対する影響は、ノードｉのアウトリンク数ｋ（ｉ）に反比例する。ここで、図５（Ａ）の場合には、ｋ（ｉ）＝１であるのに対して、図５（Ｂ）の場合には、ｋ（ｉ）＝Ｌとなる（Ｌは引用数の規定値（最低本数）でＬ＞１である）から、（Ａ）の場合に比べて（Ｂ）の場合には、ノードｊからノードｉへの影響は１／Ｌに低減される。なお、（Ｂ）の場合には、（Ｌ−１）本のリンクにより仮想ノードからノードｉへの影響が及ぼされるが、仮想ノードは、全てのノードを引用し、また全てのノードから引用されるノードであるため、全ノードにおける平均的な評価値を有する。そのため、（Ａ）では（Ｂ）よりもノードｊからの影響をＬ倍で受けることに比べると、ノードｉが引用している全ノードから受ける評価値の総計は減少することとなり、ノードｉが、被引用数が少ないにもかかわらず、引用先のノードの価値が高いことにより大きな評価を受けることが回避される。 FIG. 5 shows an example (A) when the link between nodes is simply bidirectionalized, and an example (B) when bidirectionalized by the method used in this embodiment. In FIG. 5, it is assumed that the node i quotes only the node j that is highly worth collecting many citations. Here, the influence on the node i from the node j, that is, the value of the transition probability matrix T _ij is obtained. Since node i cites node j and there is no reverse citation, T _ij is expressed by the following equation (15) based on equation (14).

From equation (14), the influence of node j on node i is inversely proportional to the number of outlinks k (i) of node i. Here, in the case of FIG. 5A, k (i) = 1, whereas in the case of FIG. 5B, k (i) = L (L is the number of citations). In the case of (B), the influence from the node j to the node i is reduced to 1 / L compared to the case of (A). In the case of (B), (L-1) links affect the node i from the virtual node, but the virtual node quotes all nodes and is quoted from all nodes. Therefore, it has an average evaluation value in all nodes. Therefore, in (A), the total evaluation value received from all the nodes cited by node i is reduced compared to receiving the influence from node j by L times than in (B). Even though the number of citations is small, it is avoided to receive a large evaluation due to the high value of the cited node.

  Based on the transition probability matrix generated by the transition probability matrix generation unit 110, the evaluation value update unit 112 repeatedly updates the evaluation value of each node until a predetermined end condition is satisfied. The rank (rank) of each node is determined in descending order of the evaluation value of each node updated by the evaluation value update unit 112. Hereinafter, an example of processing by the evaluation value update unit 112 will be described.

ここで、ｄは０≦ｄ≦１を満たす実数定数であり、評価値更新部１１２は、全てのノードｉにおいてｐ_ｉの値の変化量が予め定められた条件（例えば、ｐ_ｉの変化量が閾値以下となる）を満たすまで、繰り返し式（１５）による評価値の更新を繰り返すこととしてよい。なお、各ノードが有する評価値の初期値は予め付与することとしてよい。 The evaluation value update unit 112 calculates the evaluation value p _i of the node _i by the following equation (16).

Here, d is a real number constant satisfying 0 ≦ d ≦ 1, the evaluation value updating unit 112, all conditions the amount of change values of p _i at node i is predetermined (e.g., the variation of p _i It is possible to repeat the update of the evaluation value according to the iterative expression (15) until the value satisfies a value equal to or less than the threshold. The initial value of the evaluation value possessed by each node may be given in advance.

  The processing result holding unit 114 holds the evaluation value (rank value) of each node updated by the evaluation value update unit 112. The evaluation value of each node held in the processing result holding unit 114 may be used for various processes. For example, the information analysis apparatus 10 searches for a node that matches a specified search condition from among a plurality of nodes to be processed, and stores the searched node for each node held in the processing result holding unit 114. Display data to be displayed in the order based on the evaluation value may be generated and output.

  Next, an example of a node evaluation value calculation process performed by the information analysis apparatus 10 will be described with reference to a flowchart shown in FIG.

  As illustrated in FIG. 6, the information analysis apparatus 10 acquires node data including information of a node group including a plurality of nodes and a directed link (quotation relationship) defined between the nodes (S1001). The information analysis apparatus 10 adds a virtual node to the acquired node group (S1002). Here, the information analysis apparatus 10 selects one unselected node among the nodes included in the node group (S1003), and the number of outlinks (quotation count) of the selected node does not reach the specified value. (S1004: N), a link to the virtual node is set until the number of outlinks reaches a specified value (S1005). When the number of outlinks of the selected node is greater than or equal to the specified value (S1004: Y), or when the processing of S1005 is completed, the information analysis apparatus 10 determines whether there is an unselected node in the node group ( In S1006), if there is an unselected node (S1006: Y), the process returns to S1003 and the subsequent processing is repeated. If there is no unselected node (S1006: N), the process proceeds to S1007.

  The information analysis apparatus 10 generates an adjacency matrix indicating network information configured by links set between the node groups including the virtual nodes by the above processing (S1007). Then, the information analysis apparatus 10 performs regularization processing for aligning the number of out-links from each node to a constant number for the generated adjacency matrix (S1008), and for the regularized adjacency matrix, Bi-directional processing for bi-directionalizing each directional link is performed (S1009). Based on the bidirectional adjacency matrix, the information analysis apparatus 10 normalizes the weight of the out-link from each node to indicate the transition probability that the data value transitions to another node via the out-link. Then, a transition probability matrix is calculated (S1010).

  The information analysis apparatus 10 executes the process of sequentially updating the evaluation value (rank value) of each node according to the calculated transition probability matrix until the amount of change in the evaluation value of each node falls below a given threshold (S1011). Then, the evaluation value of each node is obtained and the process is terminated.

  In the information analysis apparatus 10 according to the present embodiment, when quoting relationships between nodes are bidirectionalized, and the number of citations of each node is calculated and the evaluation value of each node is calculated, all nodes are connected, By quoting the virtual nodes in the connected relationship until the number of citations reaches a specified value, it is avoided that a node with a small number of citations is strongly linked to the node of the citation destination. In this way, even though the number of citations is small, there is no node that is highly evaluated due to the high value of the quoted node.

  In this embodiment, an example in which a document is a node and a citation relationship is a link has been described. However, the present invention is not limited to the above example, and is similarly applied to all nodes connected by directed links. Of course.

  DESCRIPTION OF SYMBOLS 10 Information analysis apparatus, 100 Node data acquisition part, 102 Virtual node addition setting part, 104 Adjacency matrix production | generation part, 106 Regularization processing part, 108 Bidirectionalization processing part, 110 Transition probability matrix generation part, 112 Evaluation value update part, 114 Processing result holding unit.

Claims (5)

Computer
An acquisition means for acquiring connection information indicating a directed connection relationship set between a plurality of data elements;
The acquisition is performed by adding a virtual data element connected to each data element of the plurality of data elements to a plurality of data elements having a directed connection relationship and connected to each data element of the plurality of data elements. First update means for updating connection information acquired by the means;
A data element whose number of connections to other data elements is less than a predetermined first value is selected from the plurality of data elements, and the number of connections of the selected data element reaches the first value. Second update means for setting connection relations to the virtual data elements until the connection information updated by the first update means is updated;
Connection information that is set by each of the plurality of data elements and the virtual data element and that is updated by the second update unit by bidirectionalizing the connection relationship indicated by the connection information that is updated by the second update unit. A third updating means for updating
The program for functioning as a calculation means for calculating the weight of each connection relationship indicated in the connection information updated by the third update means. The computer further
Generation means for generating an adjacency matrix indicating connection relations set for each of the plurality of data elements and the virtual data element so that the number of connections from each data element matches a predetermined second value Make it work
The calculating means updates the adjacency matrix generated by the generating means by the third updating means, and then, based on the updated adjacency matrix, the weight of connection relation from each data element becomes 1 in total. The program according to claim 1, wherein the program is calculated as follows. The first value is determined based on an average number of connections calculated by dividing the number of connections indicated in the connection information acquired by the acquisition unit by the number of data elements of the plurality of data elements. The program according to claim 1 or 2, characterized in that The computer further
A function for updating the evaluation value for each of the plurality of data elements in accordance with the weight of each connection relationship calculated by the calculation unit until the predetermined end condition is satisfied is functioned as an update unit. The program according to any one of claims 1 to 3. An acquisition means for acquiring connection information indicating a directed connection relationship set between a plurality of data elements;
The acquisition is performed by adding a virtual data element connected to each data element of the plurality of data elements to a plurality of data elements having a directed connection relationship and connected to each data element of the plurality of data elements. First update means for updating connection information acquired by the means;
A data element whose number of connections to other data elements is less than a predetermined first value is selected from the plurality of data elements, and the number of connections of the selected data element reaches the first value. Second update means for setting connection relations to the virtual data elements until the connection information updated by the additional update means is updated;
Connection information that is set by each of the plurality of data elements and the virtual data element and that is updated by the second update unit by bidirectionalizing the connection relationship indicated by the connection information that is updated by the second update unit. A third updating means for updating
Calculating means for calculating a weight of each connection relation indicated in the connection information updated by the third updating means.

Images