JP3913495B2 - Route search method, route search device, program and recording medium - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a shortest path search method, a shortest path search apparatus, a recording medium, and a program for searching for a shortest path in a large amount of document data sets having hyperlink information referring to other documents.
[0002]
[Prior art]
In recent years, it has become possible to exchange a large amount of electronic documents and to send information to an unspecified number of others through a computer network such as the Internet. Therefore, a service that can search for information required by an individual for such document information has been realized on a network for an unspecified number of people.
[0003]
Documents expressed on computer networks use expressions that take advantage of their characteristics. Among them, a document called WWW (World Wide Web), as shown in FIG. 2, not only writes some information but also refers to a document written by another person existing on another computer. There is a function of "Hyperlink". This is used when other documents are trusted to supplement the information written by the user, or when referring to documents having the same contents. In particular, when referring to each other's documents, the contents of those documents or document groups are very relevant.
[0004]
FIG. 3 shows visible information having a link function. Conventionally, in order to arrive at a target document, a user selects a link based only on experience and intuition, relying on the visible information given to the link, and arrives at the target document. In FIG. 3, the link was selected based on the visible information such as “professional financial technology know-how” and “practical financial technology university” in document A.
[0005]
[Problems to be solved by the invention]
In the past, links were selected based only on experience and intuition. Users want to reach the target document with as few links as possible. A path that follows a link is called a path, and a path that has the minimum number of times is called a shortest path. However, conventionally, a link is selected only by experience and intuition, and as a result, there is a high probability that the number of times of following the link is high, and the convenience for the user is reduced.
[0006]
FIG. 4 shows an example of hypertext space. For example, in order to reach the document L from the document A in FIG. 4, the links A, B, E, D, G, K, and L may be followed six times.
As shown in FIG. 4, if the search space is finite (here, finite means finding a path within a time acceptable by the user), the search is performed so as to cover the search space, The shortest path can be found. However, the search space such as WWW is not finite, and the shortest path cannot be obtained.
The present invention has been made in view of such circumstances, and when searching for another document from a document that follows a hyperlink between documents, the shortest route that can be searched by the shortest route. An object is to provide a search method, a shortest path search device, a recording medium, and a program.
[0007]
[Means for Solving the Problems]
In order to achieve the above object, the present invention comprises a document category storage means for storing documents belonging to a category, a path search means, and a result output means, and provides a hyperlink between documents belonging to a category. A route search method of a route search device for obtaining a path from a document as a starting point to a document as an end point, wherein the path search unit assigns a weight to a hyperlink and uses a value of the weight of the hyperlink A path search process for searching the weighted directed graph to obtain a path from the document as the starting point to the document as the end point, and a result of outputting the path obtained in the path searching process performed by the result output unit and the output process, consists of the weight of the hyperlink, in the feature vector of the feature vector and the link destination document of the link source document category category And relevance is the similarity, a route search method characterized by the matching degree according to the feature vector of the feature vector and the destination document destination document category is similarity is due to the multiplication of.
[0009]
In the route search method described above, when the link destination document belongs to a plurality of categories, the hyperlink weight used in the route search is calculated for each category to which the link destination document belongs. Among the weights, the maximum weight value is used.
[0010]
The present invention is also a route search apparatus for obtaining a path from a source document to a destination document by following hyperlinks between documents belonging to a category, wherein the document category stores documents belonging to the category A storage means , a relevance degree that is a similarity between a feature vector of a category of a link source document and a feature vector of a category of a link destination document that are linked by the hyperlink, a feature vector of a category of the link destination document, and a link destination document The weight calculation means for calculating the weight of the hyperlink by multiplication with the fitness that is the similarity based on the feature vector of the feature vector, and the route search of the weighted directed graph using the value of the weight of the hyperlink, Path search means for obtaining a path from a starting document to the end document, and a path obtained in the path searching process. And result output means for outputting a route search device, characterized in that it comprises a.
[0012]
In the route search device according to the present invention, the path search unit may calculate the weight of the hyperlink used in the route search when the link destination document belongs to a plurality of categories. The maximum weight value among the weights calculated for each category is used.
[0013]
Further, the present invention is a program for causing a computer to execute the route search method described above.
[0014]
The present invention is also a computer-readable recording medium that records the above-described program.
[0015]
The present invention uses a method of making the entire path by repeating the search in the partial search space and connecting the paths having the highest weight between documents in the partial search space. Although there is no guarantee of the shortest path, a path closer to the shortest path is found, which improves the convenience for the user.
Specifically, when a directory for classifying documents is created and two documents for which a route is desired are input, a weight between documents is calculated using the directory, a path is searched based on the weight, Output the final path as a result.
[0016]
The path search will be described in more detail. The origin document of the search is set, the link destination document is extracted from the origin document, and the category (one field of the directory) to which these documents belong with respect to the origin document and the link destination document. The weight between the origin document and the linked document is obtained, and the optimum path is determined based on the weight.
[0017]
The directory input unit creates a directory. A document belongs to at least one category (one field of a directory) and is used for calculating weights.
The search condition input unit inputs search conditions such as two documents for which a route is desired.
The path search unit obtains the paths of the two input documents,
The result output unit displays the result obtained by the path search unit.
In addition, the origin setting part of the path verification part sets the search origin document,
The link destination document extraction unit extracts a link destination document from the origin document.
The category determination unit determines the category to which the origin document and linked document belong,
The weight extraction unit obtains the weights of the origin document and the link destination document.
In the path determination recording unit, the optimum path is determined from the weights, stored,
The search end determination unit determines whether or not to continue the search.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that, in all drawings illustrating the embodiment of the present invention, the same elements are denoted by the same reference numerals, and redundant description is omitted.
FIG. 1 shows a configuration of a shortest path search device according to an embodiment of the present invention. In the figure, the shortest path search apparatus has a directory input unit 101, a search condition input unit 102, a path search unit 103, and a result output unit 104.
The path search unit 103 includes a starting point setting unit 201, a link destination document extraction unit 202, a category determination unit 203, a weight calculation unit 204, a path determination recording unit 205, and a search end determination unit 206. ing.
[0019]
5 and 6 are processing flows of the shortest path search apparatus according to the embodiment of the present invention. The numbers correspond to those in the block diagram of FIG. 1, and the same number means that performs the processing. And steps.
The directory input unit 101 inputs directory configuration information prepared separately and a sample document assigned to a category for automatic classification. For example, a directory as shown in FIG. 7 is created.
The search condition input unit 102 designates two documents whose paths are to be obtained. For example, documents A and B are entered.
[0020]
In the path search unit 103, the shortest path between the documents A and B input by the search condition input unit 102 is used for each hyperlink by using the degree of association between the categories of the directory and the degree of matching between the documents and categories. The weight is calculated, and the search is performed using the weight. When finding the shortest path from an arbitrary document A to an arbitrary document B, an A * algorithm that searches for the shortest path from the weighted effective graph using the hyperlink weight value ("Basic Knowledge of Artificial Intelligence") Adapted by Taihara, modern science company 1988). This makes it possible to search for a path more efficiently than the shortest path search method such as horizontal search or vertical search (see “Basic Knowledge of Artificial Intelligence” written by Taihara, Modern Science Co., 1988).
[0021]
The result output unit 104 outputs and displays information on the shortest path obtained by the path search unit 103.
The overall flow has been described above. The details of the path verification by the path search unit 103 will be further described. Before that, since the path search is performed based on the weight between documents, a method for calculating the weight will be described. The document feature vector, the category feature vector, the suitability indicating the degree of suitability of the document for the category, and the relevance indicating the strength of the relationship between the categories are also described.
[0022]
Documents are automatically classified into a directory prepared in advance. For automatic document classification, the method described in “Prior Art” of Japanese Patent Application No. 10-281621 (“Automatic Information Classification Method and Device and Recording Medium Recording Automatic Information Classification Program” Sugisaki et al.) Can be used. . Briefly, a feature vector is first created from words existing in a document and their appearance frequency information. The value ww _ik of the word k existing in the document i is ww _ik = number of occurrences of the word k in the document i log (total number of documents / number of documents in which the word k appears)
And the feature vector vec _i of document _i is
vec _i = (ww _i1 , ..., ww _ik , ..., ww _in ) (1)
And
[0023]
Where n is the total number of words that appear in all documents (for the creation of feature vectors, see “Automatic Text Processing” Gerard Salton, ADDISON-WESLEY pub. 1989).
Similarly, a feature vector “catev _x” of category “ _x” is created from words existing in the sample document and their appearance frequency information. The value of each element in the category feature vector is the average value of the corresponding elements of the feature vector of the assigned sample document.
[0024]
The trigonometric function cosθ is calculated using the angle between the feature vector of the document and the feature vector of the category, and the degree of fitness for each category of the document is defined from the calculated value (how much judgment of “belonging to that category” is appropriate) it can. The degree of matching between the document i and the category x is expressed as rel _ix . When the trigonometric function cos θ is used, the possible value of rel _ix is between 0 and 1, with 1 being the highest fitness.
Also, the category relevance level is defined between the categories. Assume that the degree of association between category x and category y is Caterel _xy . In the case of automatic calculation, the trigonometric function cos θ is used as the relevance between categories using the angle formed by the feature vector of the category (see “Introducing the distance between categories” in Japanese Patent Application No. 10-281621). .
[0025]
A weight is defined for a hyperlink attached to another document (j) from one document (i) using the relevance between categories and the fitness of each category of the document. The hyperlink weight w _ij is _extended from the document i assigned to the category x to the document j assigned to the category y.
w _ij = Caterel _xy・ rel _jy (2)
And
An example of the hyperlink weight will be described with reference to FIG.
[0026]
Document A belongs to category “Company T”, fitness level = 0.7, and document D belongs to category “pure literature”, fitness level = 0.9. If the degree of association between “Company T” and “pure literature” = 0.11, the weight of the link from document A to document D is 0.099 (0.11 * 0.9).
The category to which a document belongs is not necessarily one. There may be more than one. In such cases, the maximum weight is adopted. FIG. 9 shows an example of hyperlink weights when there are a plurality of paths between documents. In the figure, the document D belongs to the categories “pure literature” and “parts”. Since the weight via “pure literature” is 0.099 (0.11 * 0.9) and the weight via “parts” is 0.12 (0.3 * 0.4), the weight is large. 0.12 is adopted.
[0027]
Next, path search in the path search unit 103 will be described. As a whole, the optimum path is determined from the weight of the hyperlink in the partial search space, the search in the partial search space is repeated, and the entire path is determined.
A case where a path from the document A to the document L is obtained will be described with reference to FIG. The numerical values in FIG. 10 indicate weights, and larger values indicate more desirable paths.
In the path search unit 103, first, the starting point setting unit 201 sets the starting point of the search. The first time is a document input as a search condition, and the second and subsequent times are solutions of the previous partial search. For example, in FIG.
[0028]
Next, the link destination document extraction unit 202 extracts a document with a hyperlink from the origin document. In FIG. 10, document B and document C are extracted. The category determination unit 203 determines a category for the origin document and the link destination document using an automatic classification technique, and obtains a degree of conformity between the document and the category.
In this example, the category to which the document A, the document B, and the document C belong is determined, and the fitness is obtained.
Furthermore, the weight calculation unit 204 calculates the weight of the hyperlink from the result obtained by the category determination unit 203 and the degree of association between the categories, using Expression (2). FIG. 10 shows the calculated weights.
[0029]
The path determination recording unit 205 compares the calculated weights, determines an optimal path, and records the optimal path. In the example shown in FIG. 10, the weight of the hyperlink from document A to document B is 0.4, and the weight of the hyperlink from document A to document C is 0.5. Select the hyperlink and record it.
Further, the search end determination unit 206 checks whether or not the selected document is the target document. If the target document is the target document, the process ends. If not, the process returns to step 201 to search again.
[0030]
In the example shown in FIG. 10, since the document L has not been reached, the search is performed again.
In the second and subsequent searches, the process proceeds by reflecting the previous search result. In the example shown in FIG. 10, the second search is performed starting from the document C as the first search result. Document E and document F are compared, and document F is selected. The above processing is repeated until the document L is obtained.
In the second and subsequent times, the hyperlink weight is selected by multiplying all the already selected path weights and accumulating to search for a hyperlink having a larger weight.
Thus, in the example of FIG. 10, the routes A, C, F, I, and L are obtained as solutions. The search process is summarized in FIG. Shaded rows indicate the optimal solution for partial search.
[0031]
Note that at which stage the degree of association is calculated is determined by a trade-off between the calculation time and the storage memory amount. There is a method of calculating each time for each partial search (which takes a long time), and a method of calculating and recording at the directory input stage and using the recorded value (a large amount of storage memory).
In addition, the degree of association may be manually specified individually. In this case, it is performed in the directory input process. Since the partial search method is a conventional technique, detailed description will not be made any more. However, for a simple supplement, the search may be performed in a partial search space having a plurality of link destinations.
[0032]
Further, the search may be advanced while holding a plurality of solutions by partial search. Further, the processing time or the number of searches may be employed as the search end condition. These other search methods include a method in which the system is fixed, and a method in which the user designates in the search condition input process.
[0033]
In addition, the shortest path search program shown in FIGS. 5 and 6 for realizing the function of the shortest path search apparatus shown in FIG. 1 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is recorded. The path search may be performed by causing the computer system to read and execute.
[0034]
That is, a shortest path search program for searching a path so as to pass through a shortest path from a document following a hyperlink between documents to another document, A directory input step for classifying documents, a search condition input step for specifying a search condition including at least two documents for which a route is to be obtained, a path search step for obtaining a route between the two input documents, A shortest path search program comprising a result output step for outputting a result obtained in the path search process is recorded on a computer-readable recording medium, and the shortest path search program recorded on the recording medium is read by a computer system. The path search may be performed by executing.
[0035]
In the shortest path search program recorded on the recording medium, the path search step includes a start point setting step for setting a search start document, a link destination document extraction step for extracting a link destination document for the start document, For both the origin document and the linked document, a category determining step for determining the category to which these documents belong, a weight calculating step for calculating the weight between the documents for all the linked documents, and the calculated weight Comparing and determining an optimum path from the origin document to the linked document, and recording the path determination recording step, and a search end determination step for determining whether or not to continue the search To do.
[0036]
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 general medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system.
[0037]
Furthermore, the “computer-readable recording medium” means that a program is dynamically held for a short time, like a communication line when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. It is also assumed that what holds a program for a certain period of time, such as a volatile memory inside a computer system that becomes a server or client in that case (transmission medium or transmission wave).
Further, the program may be for realizing a part of the above-described functions, and further, a program that can realize the above-described functions in combination with a program already recorded in a computer system, a so-called difference file ( Difference program).
[0038]
【The invention's effect】
As described above, in the present invention, since the path is obtained by paying attention to the weight between documents, there is a low probability that an unsuitable path (in which the number of selections becomes extremely large) is presented, and an accurate path ( The probability of presenting a path close to the shortest path is high.
In addition, since the search is performed while being divided into partial spaces, the processing time can be suppressed within a range that can be allowed by the user.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a shortest path search device according to an embodiment of the present invention.
FIG. 2 is an explanatory diagram showing a hyperlink function.
FIG. 3 is an explanatory diagram showing visible information for realizing a hyperlink function.
FIG. 4 is an explanatory diagram showing a search space for hypertext.
FIG. 5 is a flowchart showing the processing contents of the shortest path search device shown in FIG. 1;
6 is a flowchart showing details of a path search process in FIG. 5;
FIG. 7 is an explanatory diagram showing an example of a directory.
FIG. 8 is an explanatory diagram showing an example of hyperlink weights.
FIG. 9 is an explanatory diagram illustrating an example of the weight of a hyperlink when there are a plurality of search routes.
FIG. 10 is an explanatory diagram showing an example of a hypertext search space together with hyperlink weights.
FIG. 11 is an explanatory diagram showing a partial search process during hypertext path search.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 101 Directory input part 102 Search condition input part 103 Path search part 104 Result output part 201 Origin point setting part 202 Link destination document extraction part 203 Category determination part 204 Weight calculation part 205 Path determination recording part 206 Search end determination part

Claims (6)

A document category storage means for storing documents belonging to a category, a path search means, and a result output means, and a document that is a starting point is traced to a document that is a starting point by following a hyperlink between documents belonging to the category. A route search method of a route search device for obtaining a path to reach,
A path obtained by assigning a weight to a hyperlink, searching for a weighted directed graph using the hyperlink weight value, and obtaining a path from the starting document to the end document by the path search unit Search process,
A result output process performed by the result output means for outputting a path obtained in the path search process, and
The weight of the hyperlink is
The relevance that is the similarity between the feature vector of the category of the link source document and the feature vector of the category of the link destination document, and the fitness that is the similarity between the feature vector of the category of the link destination document and the feature vector of the link destination document And a route search method characterized by being multiplied by the above . When the linked document belongs to a plurality of categories, the maximum weight value of the weights calculated for each category to which the linked document belongs is calculated in the calculation of the weight of the hyperlink used in the route search . The route search method according to claim 1 , wherein the route search method is used. A route search device that obtains a path from a source document to a destination document by following a hyperlink between documents belonging to a category,
Document category storage means for storing documents belonging to the category;
A relevance that is a similarity between the feature vector of the category of the link source document and the feature vector of the category of the link destination document that are linked by the hyperlink, the feature vector of the category of the link destination document, and the feature vector of the link destination document Weight calculation means for calculating the weight of the hyperlink by multiplying the fitness that is the similarity by
Path search means for searching for a weighted directed graph using the value of the weight of the hyperlink and obtaining a path from the starting document to the ending document;
A result output means for outputting a path obtained in the path search process;
A route search apparatus comprising: The path search means, when the link destination document belongs to a plurality of categories, in calculating the weight of the hyperlink used in the route search , out of the respective weights calculated for each category to which the link destination document belongs The route search device according to claim 3 , wherein a maximum weight value is used. The program which makes a computer perform the route search method of any one of Claim 1 or Claim 2 . A computer-readable recording medium on which the program according to claim 5 is recorded.

Images