JP2005242454A - Document information retrieval apparatus, document information retrieval method and program thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a document information retrieval device capable of retrieving document information based on an accepted input sentence and having a higher retrieval result accuracy than before.
A fixed relationship strength is calculated, and a variable relationship strength is calculated. In addition, the similarity between the input text information and the document information is calculated at the time of retrieval, and a similarity order document information list that holds the document information in descending order of similarity is generated and described in the document information of the similarity order document information list. The relationship strength total value indicating the strength of the relationship between the link destination document information of the selected link destination and all the document information in the similarity order document information list in which the link destination document information is designated as the link destination is calculated. . Further, after inserting linked document information between the document information in the similarity order document information list so as not to overlap based on the similarity, the order of the document information and the linked document information is changed, and the document information and A search result list including link destination document information is generated.
[Selection] Figure 1

Description

  The present invention relates to a document information retrieval apparatus, a document information retrieval method, and a program for retrieving document information.

2. Description of the Related Art Conventionally, a document information retrieval device that retrieves document information is used for a search site such as Yahoo (registered trademark) or Google (registered commercial law) on the Internet, or a function of a FAQ (Frequently Asked Question) database incorporated in application software. Has been. Examples of the document information retrieval apparatus include a technique called a vector search for improving the accuracy of a search, a technique called a page rank for improving the accuracy of a search, and a classification of document contents of document information. A technique for improving the accuracy of search is disclosed (for example, see Patent Document 1).
US Pat. No. 6,278,992

  Here, the document information search device accepts input of a sentence from a user, for example, searches for document information similar to the input sentence, and outputs the document information as an output result. In such a search process, The other document information (link destination document information) of the link destination of the document information output as the higher order as the search result is present in the lower order of the search result, or the link destination document information does not appear in the search result. It may become. However, since the linked document information often contains important information that should be output as a search result, the accuracy of the search result is not good.

  Therefore, an object of the present invention is to provide a document information search device, a document information search method, and a program thereof that can search document information with higher accuracy of search results than the conventional one based on received input text.

The present invention has been made to solve the above-described problem, and is a document information search apparatus that holds document information and link destination document information of a link destination of the document information, the document information and the link destination document. A fixed relationship strength calculating means for calculating a fixed relationship strength indicating the strength of a fixed relationship between information, and a variable relationship strength indicating a strength of a variable relationship between the document information and the linked document information A variable relation strength calculating means for calculating the similarity, calculating similarity between the input text information received at the time of search and the document information, and generating a similarity order document information list holding the document information in descending order of similarity A degree-by-degree document information list generating means, the link-destination document information at the link destination of the document information in the degree-of-similarity order document information list, Everything in A relationship strength total value calculating means for calculating a relationship strength total value indicating the strength of the relationship with the document information based on the fixed relationship strength and the variable relationship strength, and each document information in the similarity order document information list After the link destination document information is inserted so as not to overlap based on the similarity, the threshold value of the predetermined relationship strength total value and the link destination document information are included in the similarity order document information list as document information. Search result list generation means for generating a search result list including the document information and the linked document information by switching the order of the document information and the linked document information based on whether or not the document information is included. Is a document information retrieval apparatus characterized by
As a result, the document information retrieval apparatus is linked to the document information other than the document information similar to the input sentence, and is determined to be linked to the input sentence more deeply than the document information. Information can be output as search results.

  Also, the invention is characterized in that the similarity order document information list holds document information having a predetermined similarity or higher. As a result, document information having a low similarity to the input sentence can be deleted from the search result.

  According to the present invention, the calculation of the fixed relationship strength is performed to determine whether the document content classification of the document information and the linked document information matches, the similarity between the document information and the linked document information, the link The pre-document information is calculated based on any or all of the numbers linked from the document information. Thereby, it is possible to calculate the fixed relationship strength indicating the strength of the fixed relationship between the fixed document information and the linked document information until the document information is corrected.

  According to the present invention, the calculation of the variable relationship strength is any one of a number of accesses from the document information to the link destination document information in a search process and an instruction to output the link destination document information to a forced search result. Or it is calculated based on all. Thereby, it is possible to calculate the variable relationship strength indicating the strength of the variable relationship between the document information and the link destination document information.

  Further, the present invention is characterized in that, when a predetermined period has elapsed, the relationship strength total value is set to the value of the variable relationship strength. Document information that is finally accessed by the user is stored in a log from the document information output as a search result. Based on this log, the variable relationship strength is calculated and recorded in the relationship strength table of the search index. Therefore, when a predetermined period elapses, the document information requested by the user can be retrieved sufficiently with only the variable relation strength.

  The present invention is also a document information search method in a document information search apparatus that holds document information and link destination document information of a link destination of the document information, and fixes between the document information and the link destination document information. A fixed relationship strength calculating process for calculating a fixed relationship strength indicating the strength of a relationship, and a variable relationship strength for calculating a variable relationship strength indicating a strength of a variable relationship between the document information and the linked document information Similarity order document information list generation that calculates the similarity between the input text information received at the time of search and the document information, and generates a similarity order document information list that holds the document information in descending order of similarity And all the document information in the similarity order document information list in which the link destination document information of the document information in the similarity order document information list and the link destination document information are designated as the link destination Between the relationship strength total value calculation process for calculating the relationship strength total value indicating the strength of the relationship based on the fixed relationship strength and the variable relationship strength, and each document information in the similarity order document information list Whether the link destination document information is inserted based on the similarity so as not to overlap, and whether a predetermined threshold value of the relation strength total value and the link destination document information are included in the similarity order document information list as document information. A search result list generation step of generating a search result list including the document information and the link destination document information by switching the order of the document information and the link destination document information based on the information on whether or not This is a document information search method.

  The present invention is also a program for causing a computer of a document information search apparatus that holds document information and link destination document information of a link destination of the document information to execute between the document information and the link destination document information. Fixed relationship strength calculation processing for calculating a fixed relationship strength indicating the strength of a fixed relationship, and a variable relationship for calculating a variable relationship strength indicating the strength of a variable relationship between the document information and the linked document information Similarity-order document information list that calculates the similarity between the input text information received at the time of retrieval and the document information, and generates a similarity-order document information list that holds the document information in descending order of similarity All the links in the similarity order document information list in which the link destination document information of the link destination of the document information in the similarity order document information list and the link destination document information is designated as the link destination A relationship strength total value calculation process for calculating a relationship strength total value indicating the strength of the relationship with the document information based on the fixed relationship strength and the variable relationship strength, and each document in the similarity order document information list After inserting linked document information between the information so as not to overlap based on the similarity, the threshold value of the predetermined relation strength total value and the linked document information are used as document information as the document information list in order of similarity And a search result list generation process for generating a search result list including the document information and the link destination document information by switching the order of the document information and the link destination document information based on whether or not the information is included in the computer. This is a program to be executed.

  A document information retrieval apparatus according to an embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a schematic block diagram showing the configuration of the document information search apparatus of the embodiment. In this figure, reference numeral 10 denotes an index creation unit for creating a search index of the document information search apparatus. Reference numeral 11 denotes a search engine that performs document information search processing. A document information database 12 stores document information. Here, the document information is, for example, document data described in markup language HTML or the like. An extraction unit 13 extracts document information from the document information database 12 and transmits the document information to the index creation unit 10. Reference numeral 14 denotes a search index database for storing a search index. Reference numeral 15 denotes a morphological analysis dictionary database that stores words and parts of speech in association with each other, and the index creation unit 10 performs morphological analysis of sentences held in the document information by using information stored in the morphological analysis dictionary database 15.

  In the index creation unit 10, reference numeral 101 denotes a morpheme analysis unit that performs a morphological analysis on a sentence included in document information. Reference numeral 102 denotes an importance calculation unit that calculates importance (hereinafter referred to as TFIDF) for each word of the sentence included in the document information. Reference numeral 103 denotes a classification determining unit that determines the classification of each document information from the contents of each document information. Reference numeral 104 denotes a fixed relationship strength calculation unit that calculates a fixed relationship strength indicating the strength of a fixed relationship between document information and a link destination of the document information (hereinafter referred to as link destination document information). Reference numeral 105 denotes a variable relationship strength calculation unit that calculates a variable relationship strength indicating the strength of the variable relationship between the document information and the linked document information.

  In the search engine 11, reference numeral 111 denotes a vector calculation unit that calculates the degree of similarity between the input text input and the document information stored in the document information database 12 by vector calculation. Reference numeral 112 denotes a search processing unit that performs a search process for determining sentence information similar to the input sentence as a search result and outputs the search result.

Then, this document information retrieval apparatus detects in advance the link destination document information of each document information stored in the document information database 12 and calculates the strength of the relationship between the document information and the link destination document information. deep. Here, the strength of the relationship between the document information and the linked document information includes the fixed relationship strength indicating the strength of the fixed relationship between the document information and the linked document information, and the relationship between the document information and the linked document information. And the variable relationship strength indicating the strength of the variable relationship. The strength of the fixed relationship is fixed until the document information stored in the document information database 12 is corrected. The variable relation strength gradually changes as document information search processing is performed. The variable relationship strength is also corrected by the administrator of the document information retrieval apparatus and the value changes. At the time of retrieval, the document information retrieval apparatus outputs linked document information linked to the document information as a retrieval result in addition to document information similar to the input sentence input by the user or the like. At this time, the document information search apparatus outputs important information in the link destination document information so that it appears at the top of the search result. At this time, search processing is performed using the above-described fixed relationship strength and variable relationship strength.
In the above description, the document information and the linked document information are distinguished from each other. However, some document information may be linked document information of other document information. Therefore, the document information stored in the document information database 12 is linked document information if a link is established from other document information.

FIG. 2 shows the data structure of the search index database.
The data structure of the search index will be described with reference to FIG. As shown in this figure, the search index database 14 holds a search index consisting of a document table, an in-document word table, a word table, a classification table, and a relationship strength table. In the document table, the document ID and document title for each document information stored in the document information database 12 are associated with the number of linked document information from other document information (hereinafter referred to as the number of links). keeping. The in-document word table includes the document ID, the word ID of each word included in the document information of the document ID, and the number of times that the word corresponding to the word ID appears in the document information of the document ID <TF (term frequency) value. > And the importance level <TFIDF (term frequency, inverse document frequency) value> of the word. The word table is created for each word extracted from all document information, and a word ID and a number indicating how many pieces of document information a word of the word ID appears in all document information. <DF (document frequency) value> is stored in association with each other.
If the arbitrary word is t and the number of document information is d, the TFIDF value can be calculated by the following equation.

  The classification table stores a document ID, a classification ID indicating the classification of the document content in which the document information of the document ID is classified, and a classification name of the classification ID in association with each other. In addition, the relationship strength table includes a document ID, a link destination document information ID that is an ID of a link destination document information of the document information of the document ID, document information indicated by the document ID and the link destination document ID, and a link destination document. It is the relationship strength table which memorize | stores the fixed relationship strength and variable relationship strength between information in correlation.

Next, search index creation processing in the index creation unit 10 will be described.
First, the index creation unit 10 instructs the extraction unit 13 to extract one piece of document information from the document information database 12. When one piece of document information is received, the morpheme analysis unit 101 of the index creation unit 10 reads the sentence described in the document information based on the correspondence table between words and parts of speech described in the morpheme analysis dictionary database 15. Perform morphological analysis and break it down into words. Then, the index creation unit 10 assigns a document ID to one piece of document information, and assigns a word ID to each word obtained by morphological analysis. Then, the index creation unit 10 records the word ID and the TF value of the word for each word decomposed by the morphological analysis of the document information of the sentence ID in the document word table in association with the document ID. Further, the index creating unit 10 records the word ID and the DF value of the word obtained by the morphological analysis in association with each other in the word table. The importance calculation unit 102 calculates the TFIDF value of each word and records it in the in-document word table. The processes of the morphological analysis unit 101 and the importance level calculation unit 102 are performed for all document information stored in the document information database 12, and an in-document word table and a word table are created.

  Next, the classification determination unit 103 of the index creation unit 10 determines the classification from the contents of each document information. For example, if the text information recorded in the document information database 12 is related to a computer-related document, the classification determining unit 103 assigns each document information to an OS (Operating System) using words included in the document information. And the computer model, etc., and the document ID, the classification ID of the document information of the document ID, and the classification name are recorded in association with each other in the classification table. Note that an identification number or the like for determining the classification is recorded in the document information in advance, and the classification determination unit 103 may create a classification table based on this. A plurality of classifications may be set for one document information and a classification ID may be attached.

When the above-described document table, in-document word table, word table, and classification table are created, the fixed relationship strength calculation unit 104 next performs processing between the document information and the linked document information of the document information by the following processing. Calculate the strength of the fixed relationship.
First, the fixed relationship strength calculation unit 104 reads the document information transferred from the extraction unit 13 and checks whether there is linked document information linked to the document information. For example, when the document information is document information described in a markup language such as HTML (Hyper Text Markup Language), the link destination is a URL (Uniform Resource Locator) described in a tag holding the href attribute. Document information is specified. Here, it is assumed that the document ID of certain text information is TN100006146, and the document IDs of the three linked text information described in the text information are TN100005389, TN100004412, and TN100008455. Further, it is assumed that these document information holds computer-related contents, and is classified according to the type of OS, classified according to the model name of the computer, and classified according to the part name of the computer.

  The fixed relationship strength calculation unit 104 reads the OS classification, the computer model name classification, and the component name classification of each document information whose document IDs are TN100006146, TN100005389, TN100004412, and TN100008455 from the classification table. Here, it is assumed that the TN 100006146 has the OS classification “NT” and “2000”, the model name classification “Endever 6100”, and the part name classification “HDD”. In TN100005389, the OS classification is “98”, the model name classification is “EDCUBE100”, and the part name classification is “DVD”. Further, it is assumed that the OS classification of the TN10000412 is “XP”, the model name classification is “Endever6100”, and the part name classification is “CPU”. In TN100008455, the OS classification is “2000”, the model name classification is “Endever 6100”, and the part name classification is “memory”.

FIG. 3 is a table showing the classification of document information.
The fixed relationship strength calculation unit 104 creates the table of FIG. 3 that holds the classification information in association with the document ID of one document information and the document ID of the linked document information of the document information based on the classification table, Next, based on this table, the fixed relationship strength by classification between the document information (TN100006146) and the linked document information (TN100005389, TN100004412, TN100008455) is calculated.

  The fixed relationship strength calculation unit 104 detects linked document information having the same classification as the document information TN100006146 from FIG. Here, the document information TN100006146 and the link destination document information TN100008455 have the same OS classification. The document information TN100006146, the link destination document information TN100004412, and the TN100008455 have the same model name classification. Accordingly, the fixed relationship strength calculation unit 104 determines the strength of the relationship between the document information TN100006146 and the link destination document information TN100004412 and the strength of the relationship between the document information TN100006146 and the link destination document information TN100008455. Judge that there is. When the OS classification is the same, the fixed relationship strength between the two document information by the classification increases by “1”, and when the model name classification is the same, the fixed relationship strength between the two document information by the classification increases by “2”. Then, the fixed relationship strength between the document information TN100006146 and the linked document information TN100005389 is “0”, and the fixed relationship strength between the document information TN100006146 and the linked document information TN100004412 is “2”. The strength of the fixed relationship between the document information TN100006146 and the linked document information TN100008455 is “3” because the two types of OS type and model name are the same.

FIG. 4 is a first table showing the strength of the fixed relationship between document information.
The fixed relationship strength calculation unit 104 creates a table as shown in FIG. 4 indicating the fixed relationship strength based on the classification between the document information and the linked document information, and stores the table in the memory.
Next, the fixed relationship strength calculation unit 104 calculates the similarity between the document information and the linked document information.

  FIG. 5 is a diagram showing an outline of the similar time calculation process. Here, the technique for calculating the above-described similarity will be described with reference to FIG. For example, when the A document and the B document include two words x and y, the two-dimensional vector (a) represented by the TFIDF values of the two words x and y in the A document and the B document A two-dimensional vector (b) represented by TFIDF values of two words x and y is calculated. Then, the fixed relationship strength calculation unit 104 calculates the angle formed by the vector (a) and the vector (b) (c). The value of Cos θ at this angle is the similarity. Then, when the similarity represented by Cos θ is smaller than a predetermined value, it is determined that the document A and the document B are similar. Actually, since one document includes many words, a vector in the dimension of the number of words is calculated, and the similarity between the two documents is calculated using the vector.

  By calculating the similarity, the fixed relationship strength calculation unit 104 determines whether or not the document information (TN100006146) and the linked document information (TN100005389, TN100004412, TN100008455) are similar. If the document information (TN100006146) is similar, “2” is added to the fixed relationship strength in the table of FIG.

  FIG. 6 is a second table showing the strength of the fixed relationship between document information. As shown in this figure, it is determined that the document information (TN100006146) and the link destination document information (TN100004412) are similar, and the document information (TN100006146) and the link destination document information (TN100008455) are similar. In this case, the fixed relationship strength corresponding to the combination of the document information (TN100006146) and the link destination document information (TN100004412) shown in FIG. 4 and the fixed relationship corresponding to the combination of the document information (TN100006146) and the link destination document information (TN100008455) The relationship strength is increased by “2” by the fixed relationship strength calculation unit 104.

  In addition, the fixed relationship strength calculation unit 104 reads the number of linked document information from other document information (number of links) from the document table of the search index, and if it is equal to or greater than a predetermined number of links, It is determined that the relationship with the document information (TN100006146) is strong, and “2” is added to the table of FIG. This is because the more the number of links, the more important document information is considered.

FIG. 7 is a third table showing the strength of the fixed relationship between document information.
In this figure, since the linked document information (TN100008455) is linked from many other document information, “2” is added to the strength of the relationship between the document information (TN100006146) and the linked document information (TN100008455). It shows that.

Through the above processing, the fixed relationship strength calculation unit 104 calculates the fixed relationship strength indicating the strength of the fixed relationship between the document information (TN100006146) and the link destination document information (TN100005389, TN100004412, TN100008455).
Note that the strength of the relationship based on the classification of the document information and the linked document information, the strength of the relationship based on the similarity, and the strength of the relationship based on the number of linked documents are document information (TN100006146) and linked document information (TN100005389, TN100004412 and TN100008455) are fixed. Therefore, in this embodiment, the strength of these relationships is called the fixed relationship strength.
The fixed relationship strength calculation unit 104 performs the above-described processing for calculating the fixed relationship strength for each combination of the document ID and the link destination document ID in the relationship strength table.

Next, the processing when the variable relationship strength calculation unit 105 calculates the variable relationship strength whose value gradually changes due to occurrence of document information search processing or correction of the document search system by the administrator is the same as described above. The document information (TN100006146) and link destination document information (TN100005389, TN100004412, TN100008455) will be described.
First, the variable relationship strength calculation unit 105 extracts a log that accesses linked document information from the document information based on the log information accumulated by the document information search process. Note that the log information is stored in a log storage unit or the like by the log acquisition unit of the document information search apparatus, for example. Then, the variable relationship strength calculating unit 105 determines that the document information, the link destination document information, and the link destination document information (TN100006146), the link destination document information (TN100005389, TN100004412, TN100008455) are And the variable relationship strength is increased by “2”.

FIG. 8 is a first table showing the strength of variable relationship between document information.
In this figure, it is determined that the relationship between the document information (TN100006146) and the link destination document information (TN100004412) is strong due to the number of accesses described above, and the relationship between the document information (TN100006146) and the link destination document information (TN100008455). It is shown that the variable relationship strength has increased by 2 when it is determined that is strong.

  In addition, the variable relationship strength calculation unit 105 performs a process of increasing the variable relationship strength for the link destination document information that the administrator of the document information retrieval apparatus wishes to forcibly output to the higher rank of the retrieval result. For example, when the administrator forcibly inputs information on a list of linked document information to be output to the top of the search results, the variable relationship strength calculation unit 105 performs processing to increase the variable relationship strength based on the list.

FIG. 9 is a second table showing the strength of variable relationship between document information.
This figure shows a case where the strength of the variable relationship shown in FIG. 8 is modified based on the list received by the administrator described above. In this example, linked document information (TN100004412) is described in the list. Therefore, the variable relation strength is “3” added.

  In addition to the variable relationship strength between the document information (TN100006146) and the link destination document information (TN100005389, TN100004412, TN100008455), the variable relationship strength calculation unit 105 stores the document information and the link destination document stored in the document information database 12. All variable relationship strengths between information are calculated and written into the relationship strength table of the search index database 14.

FIG. 10 shows the data structure of the relationship strength table.
As shown in this figure, in the relationship strength table, a fixed relationship strength and a variable relationship strength are stored in association with each combination of document information (TN100006146) and link destination document information (TN100005389, TN100004412, TN100008455). Yes.

Next, a process for performing a search based on an input sentence input by the document information search apparatus will be described.
First, the document information search apparatus accepts an input sentence. Then, the vector calculation unit 111 of the search engine 11 calculates the TFIDF value of each word (obtained by morphological analysis) included in the input sentence, and each of the words included in the document information stored in the document information database 12 The TFIDF value of a word (this word is also obtained by morphological analysis) is read from the search index database 14. Then, the vector calculation unit 111 determines whether the input sentence and the document information are based on the TFIDF value of each word included in the input sentence and the TFIDF value of each word included in the document information stored in the document information database 12. Similarity is performed in the same manner as the above-described calculation of similarity. Then, the similarity between all the document information stored in the document information database 12 and the input sentence is calculated.

FIG. 11 is a diagram showing a similarity order document information list.
The vector calculation unit 111 arranges document IDs of document information in descending order of similarity to the input sentence, and creates a similarity order document information list (FIG. 11) in which the similarity is associated with these document IDs. .
Next, the search processing unit 112 checks whether there is link destination document information corresponding to each document information in the similarity order document information list. This is easy by referring to the relationship strength table and reading the link destination document ID corresponding to one document ID. The search processing unit 112 temporarily stores a list of link destination document information related to each document information in the similarity order document information list on a memory or the like.

FIG. 12 shows a linked document information list.
As shown in this figure, the linked document information list stores an ID of each document information in the similarity order document information list and an ID of linked document information related to the document information in association with each other. Then, the search processing unit 112 performs the following processing for the link destination document information in the link destination document information list.

  The search processing unit 112 reads one link destination document ID from the link destination document information list of FIG. Then, the fixed relationship strength and the variable relationship strength that are associated with the link destination document ID and recorded in the relationship strength table are read, and all values are summed. That is, all the linked document information is detected with respect to the linked document information of the linked document ID read from the linked document information list, and between the document information and the read linked document information is detected. A total relationship strength value is calculated by summing up all of the fixed relationship strength and the variable relationship strength indicating the strength of the relationship. Then, the relation strength total value is calculated for the linked document information in all linked document information lists by the same processing.

FIG. 13 is a first diagram of a relation strength total value list for each linked document information.
As shown in FIG. 13, the search processing unit 112 calculates the relationship strength total value of each link destination document information in the link destination document information list, and temporarily stores the list in the memory.

  Next, the search processing unit 112 deletes information about link destination document information whose relationship strength total value is lower than a predetermined value in the relationship strength total value list for each link destination document information of FIG.

  FIG. 14 is a second diagram of the relation strength total value list for each linked document information. This figure shows a list when the IDs of linked document information whose relation strength total value is lower than a predetermined value are deleted from FIG.

  Next, the search processing unit 112 inserts each link destination document ID shown in FIG. 14 into the similarity order document information list shown in FIG. At this time, the search processing unit 112 sets the link destination document ID of the link destination document information linked from the document information next to the document ID indicating the document information having a high similarity in the similarity order document information list. insert. Even if the link destination document ID inserted in the similarity order document information list is the link destination document ID of the link destination document information that is the link destination of the other document information, it is next to the other document information. Not inserted.

  FIG. 15 is a diagram illustrating a first example in which link destination document information is inserted into the similarity order document information list. As shown in this figure, the ID of the linked document information of the document information is inserted next to the ID of the document information having a high similarity in the similarity order document information list. Here, the document ID of TN100008889 in FIG. 15 is the ID of document information appearing in the similarity order document information list of FIG. 11 and the linked document information list of FIG. Therefore, since it originally exists in the similarity order document information list, the link destination document information is not inserted into the similarity order document information list.

  Next, the search processing unit 112 determines whether or not the total relationship strength value of the inserted link destination document information is higher than a predetermined threshold in the list shown in FIG. When the total relationship strength value is higher than the predetermined threshold, the search processing unit 112 moves the ID of the link destination document information above the ID of the document information linked to the link destination document information. Let In other words, for linked document information with a high relationship strength total value, it is determined that the linked document information is more closely related to the input text than the document information linked to the linked document information. Move up.

  FIG. 16 is a diagram showing a second example in which link destination document information is inserted into the similarity order document information list. This figure shows the above-described processing in which the ID of the linked document information is moved over the ID of the document information linked to the linked document information when the total relationship strength value is higher than a predetermined threshold value. A list of document information in order of similarity into which link destination document information is inserted later is shown.

  Further, the search processing unit 112 considers that the document information of the document ID appearing in both the similarity order document information list in FIG. 11 and the linked document information list in FIG. 13 has a large relationship with the input sentence. , Move to the top of the list of FIG. In the present embodiment, the document information whose document ID is TN100008889 corresponds. However, since the relationship strength total value of the link destination document information with the document ID TN100008889 is lower than the relationship strength total value of the link destination document information with the document ID TN100001983, the document ID of the TN100001983 is Next, the document ID of TN100008889 is moved. Then, the search processing unit 112 sets the document information list of the search results as the similarity order document information list in which the link destination document information is inserted, obtained by the above processing.

FIG. 17 is a diagram showing a list of document information of search results.
Then, the search processing unit 112 searches the titles of the document information (or linked document information) in the order of the document IDs shown in the document information list (search result list) of the search results in FIG. It displays on the display part with which the apparatus was equipped. As a result, a search result indicating a list of document information similar to the input document input to the document information search apparatus is output.

  As a result, the document information retrieval apparatus is linked to the document information other than the document information similar to the input sentence, and is determined to be linked to the input sentence more deeply than the document information. Information can be output as search results.

  In the present embodiment, the document information search apparatus holds the document information database 12, and searches for document information that is similar to or has a large relationship with the input document from the document information recorded in the document information database 12. For example, a search process may be performed for document information that is similar or closely related to the input document from document information stored in other document information storage devices connected via a communication network. Good.

  Further, the document information search processing of the present embodiment uses the fixed relationship strength and the variable relationship strength, but when a predetermined period has elapsed, the value of the variable relationship strength is set as the relationship strength total value. The search process may be performed. In other words, the document information accessed by the user from the document information output as the search result is held in the log, and is recorded in the relationship strength table of the search index as the variable relationship strength calculated based on this log. This is because the document information requested by the user can be retrieved sufficiently with only the variable relationship strength.

  The document information apparatus described above has a computer system inside. The process described above is stored in a computer-readable recording medium in the form of a program, and the above process is performed by the computer reading and executing this program. Here, the computer-readable recording medium means a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Alternatively, the computer program may be distributed to the computer via a communication line, and the computer that has received the distribution may execute the program.

  The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is a schematic block diagram which shows the structure of a document information search device. It is a figure which shows the data structure of the index database for search. It is a table | surface which shows the classification | category of document information. It is a 1st table | surface which shows the fixed relationship strength between document information. It is a figure which shows the outline of the process of calculation at the time of similarity. It is a 2nd table | surface which shows the fixed relationship strength between document information. It is a 3rd table | surface which shows the fixed relationship strength between document information. It is a 1st table | surface which shows the variable relationship strength between document information. It is a 2nd table | surface which shows the variable relationship strength between document information. It is a figure which shows the data structure of a relationship strength table. It is a figure which shows a similarity order document information list. It is a figure which shows a link destination document information list. It is a 1st figure of the relation strength total value list for every link destination document information. It is a 2nd figure of the relation strength total value list for every link destination document information. It is the 1st example by which the link destination document information was inserted in the similarity order document information list. It is a 2nd example by which the link destination document information was inserted in the similarity order document information list. It is a figure which shows the document information list of a search result.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Index creation part, 11 ... Search engine, 12 ... Document information database, 13 ... Extraction part, 14 ... Index database for search, 15 ... Morphological analysis dictionary database, 101. ..Morphological analyzer, 102 ... Importance calculator, 103 ... Classification determination unit, 104 ... Fixed relationship strength calculator, 105 ... Variable relationship strength calculator, 111 ... Vector calculator 112 ... Search processing unit

Claims (7)

A document information search apparatus that holds document information and link destination document information of a link destination of the document information,
Fixed relationship strength calculating means for calculating a fixed relationship strength indicating the strength of a fixed relationship between the document information and the linked document information;
Variable relationship strength calculating means for calculating a variable relationship strength indicating the strength of a variable relationship between the document information and the linked document information;
A similarity order document information list generating means for calculating a similarity between the input sentence information received at the time of search and the document information, and generating a similarity order document information list holding the document information in descending order of similarity;
Relationship between the link destination document information of the document information in the similarity order document information list and all the document information in the similarity order document information list specifying the link destination document information as the link destination A relationship strength total value calculating means for calculating a relationship strength total value indicating the strength of the relationship based on the fixed relationship strength and the variable relationship strength;
After inserting linked document information between the document information in the similarity order document information list so as not to overlap based on the similarity, a predetermined threshold value of the relation strength total value and linked document information are A search result list including the document information and the linked document information is obtained by switching the order of the document information and the linked document information based on whether or not the document information is included in the similarity order document information list. A search result list generating means for generating;
A document information retrieval apparatus comprising: The document information search apparatus according to claim 1, wherein the similarity order document information list holds document information having a predetermined similarity or higher. The calculation of the fixed relationship strength is as follows:
Determining whether the document content classification of the document information and the linked document information matches;
Similarity between the document information and the linked document information;
The number of linked document information linked from the document information;
The document information search device according to claim 1, wherein the document information search device is calculated based on any or all of the following. The calculation of the variable relationship strength includes the number of accesses from the document information to the linked document information in the search process, an instruction to output the forced search result of the linked document information,
The document information search device according to claim 1, wherein the document information search device is calculated based on any or all of the following. If the specified period has passed,
The document information search apparatus according to claim 1, wherein the relation strength total value is the value of the variable relation strength. A document information search method in a document information search apparatus that holds document information and link destination document information of a link destination of the document information,
A fixed relationship strength calculating step of calculating a fixed relationship strength indicating the strength of a fixed relationship between the document information and the linked document information;
A variable relationship strength calculating step for calculating a variable relationship strength indicating the strength of a variable relationship between the document information and the linked document information;
Calculating the similarity between the input sentence information received at the time of search and the document information, and generating a similarity order document information list generation process for generating a similarity order document information list holding the document information in descending order of similarity;
Relationship between the link destination document information of the document information in the similarity order document information list and all the document information in the similarity order document information list specifying the link destination document information as the link destination A relationship strength total value calculation process for calculating a relationship strength total value indicating the strength of the relationship based on the fixed relationship strength and the variable relationship strength;
After inserting linked document information between the document information in the similarity order document information list so as not to overlap based on the similarity, a predetermined threshold value of the relation strength total value and linked document information are A search result list including the document information and the linked document information is obtained by switching the order of the document information and the linked document information based on whether or not the document information is included in the similarity order document information list. The search result list generation process to be generated,
A document information search method characterized by comprising: A program to be executed by a computer of a document information retrieval apparatus that holds document information and link destination document information of a link destination of the document information,
A fixed relationship strength calculation process for calculating a fixed relationship strength indicating the strength of a fixed relationship between the document information and the linked document information;
A variable relationship strength calculation process for calculating a variable relationship strength indicating the strength of a variable relationship between the document information and the linked document information;
Similarity order document information list generation processing for calculating similarity between the input sentence information received at the time of search and the document information, and generating a similarity order document information list that holds the document information in descending order of similarity,
Relationship between the link destination document information of the document information in the similarity order document information list and all the document information in the similarity order document information list specifying the link destination document information as the link destination A relationship strength total value calculation process for calculating a relationship strength total value indicating the strength of the relationship based on the fixed relationship strength and the variable relationship strength;
After inserting linked document information between the document information in the similarity order document information list so as not to overlap based on the similarity, a predetermined threshold value of the relation strength total value and linked document information are A search result list including the document information and the linked document information is obtained by switching the order of the document information and the linked document information based on whether or not the document information is included in the similarity order document information list. Search result list generation processing to be generated,
A program that causes a computer to execute.

Images