JP2008071198A - Document retrieval device, document retrieval method, document retrieval program and storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document retrieval device outputting an appropriate retrieval result to a retrieval request without unnecessarily limiting selection of a seed document or expansion word. <P>SOLUTION: The document retrieval device 10 for retrieving a document from a document database 15 based on an input retrieval condition comprises a seed document acquisition means 12 for acquiring a seed document based on an input seed document acquisition character string; a relevant document acquisition means for acquiring other documents used by a user of the seed document as relevant document; a word extraction means 13 for extracting a word relevant to the retrieval condition from the seed document and the relevant documents; and a retrieval means 14 for retrieving the document based on the retrieval condition and the word extracted by the word extraction means 13. The word extraction means 13 determines relevancy of each word with a predetermined keyword based on the distance between each word and the predetermined keyword, and extracts a predetermined number of words in descending order of the relevancy. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a document search device, a document search method, a document search program, and a recording medium, and in particular, a document search device that searches a set of documents that match a search condition based on an input search condition, The present invention relates to a document search method, a document search program, and a recording medium.

  In the field of document search, whether or not the search result matches the search request of the user (searcher) is one of important evaluation criteria. 2. Description of the Related Art Conventionally, a document search apparatus that obtains a degree of matching with a search request (hereinafter referred to as “fitness”) for each document based on a search term specified in the search request and outputs a search result in descending order of suitability has been proposed. (For example, Patent Document 1).

  In addition, in order to obtain a high-quality search result, there is a technique (hereinafter referred to as “related word expansion”) in which not only a search term specified by a user but also a related term is added as a search term. Various proposals have been made for a method of selecting a search term (hereinafter referred to as “extended word”) added by expansion of related terms.

  For example, a method called relevance feedback is known. This method first presents the result of a search (primary search) with a search term designated by the user to the user, and the document presented as a result is a relevant document ( Documents desired by the user) and non-conforming documents. After that, the result is obtained, and the result of the search (secondary search) using the expanded word selected from the words included in the matching document is output as the final result. Hereinafter, a document used for selecting an extended word is referred to as a “seed document”.

  In addition, there is a technique called pseudo-compatibility feedback in order to reduce the burden imposed on the user by conformity feedback. In this method, an extended word is obtained by using a document positioned higher in the result of the primary search as a seed document.

  However, in the conventional relevance feedback and pseudo relevance feedback as described above, since the seed document is selected from the document group to be searched (the result of the primary search), the selection of extended words is limited to the result of the primary search. This may reduce the quality of the final search result.

  Several methods have been proposed to compensate for this drawback. For example, in Patent Document 2, even if the quality of the primary search result is poor by feeding back the result of the suitability calculation of the primary search to the suitability calculation of the secondary search. Reducing the negative impact on the quality of the final search results.

  In Patent Document 3, the seed document obtained as a result of the primary search is divided into a plurality of groups based on bibliographic items such as authors and dates, and the quality of the final search result is selected by selecting extended words from various viewpoints. Has improved.

On the other hand, a method has been proposed in which related words are registered in advance for each word, and related words are expanded based on the correspondence. For example, Patent Document 4 proposes a method of registering related words in the form of a co-occurrence word database.
Japanese Patent Laid-Open No. 11-224264 JP 2003-242170 A JP 2004-192374 A JP 2003-022275 A

  However, the document search apparatuses in Patent Document 2 and Patent Document 3 are still greatly affected by the primary search when selecting a seed document. In addition, the document search apparatus described in Patent Document 4 needs to register the correspondence relationship of extended words in advance, so that it is necessary to maintain the correspondence relationship and is applied to a field where terms are added one after another. There is a problem that is difficult.

  The present invention has been made in view of the above points, and is a document search that can output an appropriate search result in response to a search request without providing an extra restriction on the selection of a seed document or an extended word. An object is to provide an apparatus, a document search method, a document search program, and a recording medium.

  In order to achieve the above object, a document search device according to a first invention is a document search device that searches a document from a predetermined document database based on an input search condition, and acquires an input seed document. A seed document acquisition unit that acquires a seed document based on a character string for use; a related document acquisition unit that acquires another document used by a user of the seed document acquired by the seed document acquisition unit; and A word extracting unit that extracts a word related to the search condition from the seed document and the related document, and a search unit that searches the document based on the search condition and the word extracted by the word extracting unit. Features.

  A second invention is a document search apparatus according to the first invention, wherein the related document is borrowed by a borrower of the seed document in addition to other documents used by a user of the seed document. And other documents purchased by a purchaser of the seed document or other documents browsed by a viewer of the seed document.

  A document search device according to a third aspect of the invention is a document search device that searches a document from a predetermined document database based on an input search condition, and is based on an input seed document acquisition character string. Seed document acquisition means for acquiring a document; word extraction means for extracting a word related to the search condition from the seed document acquired by the seed document acquisition means; the search condition and the word extracted by the word extraction means; Search means for searching for a document based on the word, and the word extraction means determines the degree of relevance of each word to the predetermined keyword based on the distance between each word and the predetermined keyword, A predetermined number of words are extracted in descending order of relevance.

  Further, a fourth invention is a document search device according to the third invention, wherein the word extracting means includes the appearance frequency of each word or each word in addition to the distance between each word and the predetermined keyword. The degree of relevance of each word to the keyword is determined based on the number of seed documents including, and a predetermined number of words are extracted in descending order of the degree of relevance.

  Further, the fifth invention is a document search device according to the third or fourth invention, wherein a reduction rate for setting a reduction rate of the degree of association that decreases as the distance between a word and the predetermined keyword increases. It has a rate setting means.

  The sixth invention is a document retrieval apparatus according to the fifth invention, wherein the reduction rate changes for each sentence.

  A document search method according to a seventh aspect of the invention is a document search method for searching a document from a predetermined document database based on an input search condition, wherein the seed search is performed based on an input seed document acquisition character string. A seed document acquisition step for acquiring a document, a related document acquisition step for acquiring another document used by a user of the seed document acquired in the seed document acquisition step as a related document, and the seed document and the related document. The method includes a word extraction step for extracting a word related to the search condition, and a search step for searching for a document based on the search condition and the word extracted in the word extraction step.

  A document search method according to an eighth aspect of the present invention is a document search method for searching a document from a predetermined document database based on an input search condition, wherein the seed search is based on an input seed document acquisition character string. A seed document acquisition step for acquiring a document; a word extraction step for extracting a word related to the search condition from the seed document acquired in the seed document acquisition step; and a search condition and the word extraction step extracted A search step for searching for a document based on a word, wherein the word extraction step determines a degree of relevance of the word to the predetermined keyword based on a distance between the word and the predetermined keyword, A predetermined number of words are extracted in descending order of the degree of association.

  A document search program according to the ninth invention is a document search program for causing a computer to search a document from a predetermined document database based on an input search condition, wherein the input seed document acquisition character A seed document acquisition step for acquiring a seed document based on a column; a related document acquisition step for acquiring another document used by a user of the seed document acquired in the seed document acquisition step; and the seed document And a word extraction step for extracting a word related to the search condition from the related document, and a search step for searching for a document based on the search condition and the word extracted in the word extraction step. To do.

  A document search program according to a tenth invention is a document search program for causing a computer to search for a document from a predetermined document database based on an input search condition, wherein the input seed document acquisition character is A seed document acquisition step of acquiring a seed document based on a column; a word extraction step of extracting a word related to the search condition from the seed document acquired in the seed document acquisition step; the search condition and the word extraction A search step for searching for a document based on the word extracted in the step, wherein the word extraction step relates the word to the predetermined keyword based on a distance between the word and the predetermined keyword. The degree is determined, and a predetermined number of words are extracted in descending order of the degree of association.

  The recording medium according to the eleventh invention is a computer-readable recording medium recording the document search program according to the ninth or tenth invention.

  According to the present invention, a document search apparatus, a document search method, a document search program, and a record capable of outputting an appropriate search result in response to a search request without placing an extra restriction on the selection of seed documents and extended words A medium can be provided.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram illustrating a functional configuration example of a document search apparatus according to an embodiment of the present invention. In FIG. 1, the document search apparatus 10 includes a search request input unit 11, a seed document acquisition unit 12, an extended word extraction unit 13, a document search unit 14, and a document database 15.

  The search request input unit 11 is a means for displaying a search request input screen and acquiring a search condition by inputting a search word, a search character string, a search expression, and the like for searching for a desired document.

  Further, the search request input unit 11 is a seed document acquisition character string (word, compound word, or sentence) for acquiring a seed document by allowing a user to input a word, a character string, or a sentence representing the content of a desired document. Etc.).

  FIG. 2 is a diagram showing a display example of the search request input screen. The search request input screen 110 includes a search condition input area 111, a seed document acquisition character string input area 112, a seed number input area 113, and a search button 114. Composed.

  The search condition input area 111 is a text box for inputting a search condition, and the seed document acquisition character string input area 112 is a text box for inputting a seed document acquisition character string or a sentence including the character string. .

  For example, a natural sentence such as spoken language may be input to the seed document acquisition character string input area 112. In this case, the search request input unit 11 acquires a seed document from the input natural sentence by morphological analysis or the like. Extract the character string.

  In the seed document acquisition character string input area 112, a character string having the highest degree of relevance (for example, the searched document) among the search results (document group) based on the search condition input in the search condition input area 111 is displayed. The character string having the highest appearance frequency among them may be automatically extracted and input, or a character string arbitrarily selected by the user from the search results (document group) may be input.

  The seed number input area 113 is a text box for inputting the maximum number of seed documents acquired by the seed document acquisition character string. For example, when “10” is input as the maximum number of seed documents, the seed document is input. Even if there are 100 documents retrieved based on the acquisition character string, only 10 out of 100 documents are set as seed documents based on the appearance frequency of the seed document acquisition character string.

  The search button 114 is a button for starting a search of a document. When the search button 114 is pressed, a seed document is extracted based on a character string for obtaining a seed document, and an extended word is extracted from the seed document. Documents are searched based on conditions and extended words.

  The seed document acquisition unit 12 is a means for acquiring a seed document based on the seed document acquisition character string acquired by the search request input unit 11.

  In addition, the seed document acquisition unit 12 performs a primary search based on the seed document acquisition character string acquired by the search request input unit 11, and is used by a user of the document obtained by the primary search. Is acquired as a related document.

  In order to increase the number of seed documents by adding related documents to the seed documents, and to make the extended words extracted from the seed documents more appropriate, the users of the seed documents are similar in content to the seed documents. This is because there is a high possibility that the document to be used is used.

  The extended word extraction unit 13 is a means for selecting a predetermined number of extended words from the words constituting the seed document, and may be configured to select a predetermined number of extended words from the words constituting the seed document and the related document.

  The extended word extraction unit 13 extracts, for example, all the words included in the seed document by morphological analysis, calculates the appearance frequency of each word in the seed document, and calculates a predetermined number (for example, 5) in descending order of appearance frequency. Extract words as extended words.

  The document search unit 14 is adapted from a set of documents stored in the document database 15 (hereinafter referred to as “searched document”) based on the search condition and the extended word extracted by the extended word extraction unit 13. This is a means for searching a document to be searched and presenting a list of search results to the user. For example, a document including both a search condition and an extended word may be searched. You may search the document which contains.

  Further, the document search unit 14 may search for a document including all of the extended words, or may search for a document including a predetermined number (for example, three) or more of extended words.

  The document database 15 is a database in which searched documents are accumulated.

  The document retrieval apparatus 10 may be configured by a single computer, or may be configured by a plurality of computers by adopting a client / server type or the like. In the latter case, for example, the search request input unit 11 may be mounted on the client, and the seed document acquisition unit 12, the extended word extraction unit 13, the document search unit 14, and the document database 15 may be mounted on the server.

  FIG. 3 is a diagram illustrating a hardware configuration example of the document search apparatus 10 according to the embodiment of the present invention. 3 includes a drive device 100, an auxiliary storage device 102, a memory device 103, an arithmetic processing device 104, a display device 105, and an input device 106.

  The drive device 100 is a device for reading a program or the like recorded on the recording medium 101.

  The recording medium 101 is a portable recording medium for recording various data, and examples thereof include a CD-ROM and a DVD-ROM.

  The auxiliary storage device 102 is a non-volatile recording medium for storing a program for executing various processes in the document search device 10, and includes, for example, a hard disk. When the recording medium 101 on which the program is recorded is set in the drive device 100, the document search device 10 reads the program from the recording medium 101 of the drive device 100 and installs it in the auxiliary storage device 102.

  The memory device 103 is a volatile recording medium for loading a program for executing various processes in the document search device 10, and includes, for example, a RAM (Random Access Memory). When there is a program start command, the document search device 10 reads the program from the auxiliary storage device 102 and loads it into the memory device 103.

  The arithmetic processing unit 104 is a device for sequentially executing a program loaded in the memory device 103.

  The display device 105 is a device for displaying a GUI (Graphical User Interface) or the like by a program, and the input device 106 is configured by a keyboard, a mouse, and the like, and is a device for receiving various operation instructions.

  Next, a processing procedure in the document search apparatus 10 will be described with reference to FIG. FIG. 4 is a flowchart for explaining the document search processing by the document search apparatus 10 in the first embodiment.

  First, the search request input unit 11 displays the search request input screen 110 on the display device 105, and causes the user to input a search request (step S101).

  When a search condition, a character string for acquiring a seed document or a sentence including the same, a maximum number of seed documents, and the like are input, and the search button 114 is clicked, the seed document acquiring unit 12 inputs a character string for acquiring the seed document The sentence input to the area 112 is divided into words by morphological analysis (step S102).

  Next, the seed document acquisition unit 12 calculates the appearance frequency in the searched document for each word (step S103).

  Next, the seed document acquisition unit 12 selects the word having the highest appearance frequency (step S104), and based on the selected word, the search condition input in the search condition input area 111, and the maximum number of seed documents. A command statement indicating a search request for the document database 15 is generated (step S105). Note that the seed document acquisition unit 12 may select a plurality of words in descending order of appearance frequency.

  The command statement indicating the search request is described in a well-known SQL (Structured Query Language) syntax or its extended syntax, and is, for example, an extended syntax using the following subquery.

select title from document where body contains 'environmental protection' expand from (select title from document where document ID in (select document ID from document ID history where user ID in (select first user ID from user ID history where document ID in (select document ID from document where text contains 'warming' limit 10))))
In the following, the above-described command sentence is divided into a plurality of parts for explanation.

select Title from Document where Text contains 'Environmental protection' (1)
expand from ... (2)
(select title from document where document ID in (3)
(select document ID from document ID history where user ID in (4)
(select first user ID from user ID history where document ID in (5)
(select document ID from document where text contains 'warming' limit 10))))) (6)
The part (1) is a search command for a document table defined in the document database 15. More specifically, “extract the title of a document that includes the word“ environmental protection ”in the body of the document in the document table. "Means the command.

  The “document table” is a table that systematically forms various data related to the document specified by the document ID. For example, as shown in FIG. 4A, the document ID, title, author, publisher, translator Etc. fields.

  The “document ID” is an identifier for specifying a document stored in the document database 15 and is expressed by, for example, a number, a symbol, a character string, etc., and includes a document table, a user ID history table, and a document ID. Used as a common item in the history table.

  The “user ID history table” is a table in which the user history is recorded for each document. For example, the user ID history table is stored in the document database 15 and, as shown in FIG. 4B, the document ID and the first user ID. And a field such as a second user ID.

  The user ID history table is a table for recording the history of users who used each document in time series. For example, the management of library rental history, the management of bookstore sales history, the browsing history of websites, etc. It is used for management etc.

  The “user ID” is an identifier for specifying a user, and is represented by, for example, a number, a symbol, a character string, etc., and is used as a field in the user ID history table and the document ID history table. A “user” is a person who uses a document, for example, a borrower of a document (book) when the document search apparatus 10 is introduced into a library, and a document when the document search apparatus 10 is introduced into a bookstore. (Books) purchasers, documents (contents) viewers when the document search device 10 is installed on a website, and the like.

  The first user ID is the identifier of the most recent user who used the corresponding document, and the second user ID is the user who used the document before the user indicated by the first user ID. It is an identifier.

  The “document ID history table” is a table in which the history of documents used by each user is recorded for each user. For example, the document ID history table is stored in the document database 15 and, as shown in FIG. Fields such as ID, first document ID, and second document ID are included.

  The document ID history table is a table that records the history of documents used by each user in time series. For example, as with the user ID history table, the management of library rental history and the sales history of bookstores are recorded. Used for management, management of website browsing history, etc.

  The first document ID is an identifier of the latest document used by the corresponding user, and the second document ID is the document used before the document indicated by the first document ID used by the corresponding user. It is an identifier.

  The outermost select statement (3) in the subquery following the description (2) “expand from” is a search command for acquiring more seed documents. This means a command “extract the title of the record whose ID value matches the search result value of (4)”.

  Note that “expand from X” means an instruction “extract a predetermined number of expansion words from the document group indicated by X”.

  The second outer select statement (4) means an instruction “extract the document ID of a record whose user ID value matches the search result value of (5) in the document ID history table”. To do.

  The third outer select statement (5) says, “Extract the first user ID of a record whose document ID value matches the search result value of (6) in the user ID history table”. Means an instruction.

  The innermost select statement (6) means an instruction “search the top 10 document IDs of records including the word“ warming ”in the document body in the document table”. The ranking for determining the top 10 cases is determined based on, for example, the appearance frequency of “warming” in each document.

  The word “warming” is a word extracted from the character string for seed document acquisition, and “limit 10” indicates the maximum number of seed documents to be acquired. “Environmental protection” is a search term input as a search condition.

  That is, the above SQL syntax searches for the first user ID (meaning the latest user ID) of the user who used the seed document searched in (6) in (5), and (5) The document ID of the document other than the seed document used by the user having the first user ID searched for in (4) is searched for in (4), and the document indicated by the document ID searched in (4) is (3) In step (2), a predetermined number of extended words are extracted from the related document extracted in (3), and the expanded word or the search term “environmental protection” extracted in (2) is included in the text. Extract the document title. "".

  For example, the document search apparatus 10 acquires the document ID value 2 of the record including the word “warming” in the body of the document from the document table in FIG. 5A, and refers to the user ID history table to obtain the document. The history of the user ID corresponding to the ID value 2 is acquired (in the case of FIG. 5B, the first user ID = 3 is acquired).

  Thereafter, the document search apparatus 10 refers to the document ID history table to obtain a document ID history corresponding to the user ID value 3 (in the case of FIG. 5C, the first document ID = 2, the second document ID). Document ID = 4 and third document ID = 5 are acquired).

  Thereafter, the document search apparatus 10 extracts a document having a document ID value of 2, 4, or 5 as a related document, extracts an extended word from the seed document and these related documents, and further extracts the extracted extended word or search word. Search for documents that contain "Environmental protection" in the text.

  Note that the document search apparatus 10 may use only the document indicated by the first document ID as a related document, or may use all documents in the history as related documents.

  As a result, compared to the case where only the seed document searched in (6) is the target of extended word extraction, more documents can be the target of extended word extraction, and the extracted extended words are more appropriate. Can be.

  Note that the above-described command sentence may be explicitly input by the user (searcher). However, it is desirable from the viewpoint of convenience for users unfamiliar with SQL that the system automatically creates a command sentence by providing a GUI such as the search request input screen 110.

  Subsequently, referring to FIG. 4 again, the seed document acquisition unit 12 actually acquires a seed document from the document database 15 based on the generated command statement (step S106). That is, the seed document acquisition unit 12 acquires the top 10 documents among the documents including the keyword “warming” as a seed document by executing the above-described instruction (6) on the document database 15.

  Subsequently, the seed document acquisition unit 12 acquires the first user ID of each seed document based on the command statement (5) (step S107).

  Thereafter, the seed document acquisition unit 12 acquires a document other than the seed document used by the user indicated by the first user ID as a related document based on the statements (3) and (4) (step S108).

  That is, the seed document acquisition unit 12 executes the above-described commands (3) to (5) on the document database 15, so that the top 10 seed documents among the documents including the keyword “warming” are included. Other documents used by other users are acquired as related documents.

  As described above, the document search apparatus 10 uses a seed document based on the user ID of a user (referred to as a borrower, purchaser, or viewer) of the seed document extracted by the seed document acquisition character string. The document IDs of other documents used (including borrowing, purchasing, browsing, etc.) are extracted, and the documents indicated by these document IDs are acquired as related documents.

  Subsequently, the extended word extracting unit 13 selects and extracts extended words from the seed document and related documents acquired by the seed document acquiring unit 12.

  That is, the extended word extraction unit 13 divides the seed document and the related document into words (step S109), and calculates the document frequency for each word (step S110). Here, “document frequency” for a word refers to the number of seed documents or related documents including the word, and is expressed as a ratio to the total number of seed documents, for example, and the total number of seed documents and related documents is 50. When a certain word is included in 25 of them, the document frequency is 0.5 (50%).

  Further, the extended word extraction unit 13 selects a predetermined number of words in descending order of document frequency, and extracts the selected words as extended words (step S111). Note that an appearance frequency (the number of words appearing in the seed document) may be used instead of the document frequency.

  In addition, the seed document and the related document may be divided into words by using units separated by white space or by using known morphological analysis. Or you may use what was simply divided by a fixed number of characters.

  Further, the extended word extraction unit 13 may implement a mechanism in which words inappropriate for use as extended words are registered in advance and these words are not extracted as extended words.

  Further, the extended word extraction unit 13 may set the number of words to be extracted as an extended word as a fixed value, and may cause the search request input unit 11 to specify a user (searcher) via the GUI or the like.

  Subsequently, the document search unit 14 stores in the document database 15 a document that includes the search condition (search word) input on the search request input screen 110 and all or part of the extended words extracted by the extended word extraction unit 13. A search is performed from the set of documents (step S112), and the search result is presented to the user. For this process, for example, a method described in JP-A-2003-281181 may be used.

  Further, the document search unit 14 may search for a document including all or part of the search word or the extended word.

  As described above, the document search apparatus 10 according to the first embodiment selects an extended word based on a character string (a character string for seed document acquisition) specified by a user (searcher). High-quality search results closer to the user's intention.

  In addition, since the document search apparatus 10 according to the first embodiment can input the seed document acquisition character string together with the input of the search condition, it is easy to obtain high quality by a single input operation by the user (searcher). Search results can be provided.

  In addition, the document search apparatus 10 according to the first embodiment adds a document searched for based on a seed document acquisition character string specified by a user (searcher) and a document common to the user to the seed document. The set for extracting extended words (the number of seed documents) can be increased, and search results that meet the expectations of users (searchers) based on extended words carefully selected from more documents. Can be provided.

  Next, a second embodiment will be described. The second embodiment is characterized in that the extended word extraction unit 13 determines the degree of association of each word with the keyword based on the distance between the word and the keyword, and extracts a word with a high degree of association as an extended word. Have

  In the second embodiment, the functional configuration and hardware configuration of the document search apparatus 10 are the same as those shown in FIGS. 1 and 2, respectively.

  FIG. 6 is a flowchart for explaining the document search process by the document search apparatus 10 in the second embodiment, and steps S201 to S206 are common to the process flow described in the flowchart of FIG.

  When the seed document is acquired by the processing from step S201 to step S206, the extended word extraction unit 13 divides the seed document into words by morphological analysis or the like (step S207), and the distance between each word and the keyword (for example, there is If there are a plurality of words in the same document, the shortest distance from the keyword is acquired) (step S208).

  The keyword may be the search term “environmental protection” input to the search condition input area 111, and is a word selected from the seed document acquisition character string (for example, “earth”, “warming”, etc.). It may be.

  Here, “distance” is an interval expressed by the number of characters, the number of words, the number of sentences, etc. between each word and the keyword, and the smaller the distance, the higher the degree of association between each word and the keyword. It is said.

  The extended word extraction unit 13 calculates the document frequency or appearance frequency (the number of occurrences of each word in the seed document) for each word, and extracts a predetermined number of words as an extended word in descending order of the document frequency or appearance frequency. , A weighting factor is derived based on the distance between each word and the keyword, and the degree of association between each word and the keyword is obtained by multiplying the document frequency or appearance frequency by the weighting factor (relevance = weighting factor × document frequency or Appearance frequency) is determined (step S209).

  The extended word extraction unit 13 expresses the weighting coefficient in the range of 0 to 1 and sets the case of the keyword itself to 1 (refers to a state where the distance is 0). Is set to 0. A method for calculating the weighting coefficient will be described later.

  Further, the extended word extraction unit 13 has a document frequency or appearance frequency in a range from 0 to 1 (for example, in the case of a document frequency, a value obtained by dividing the number of documents including each word by the number of seed documents). Represented by

  The extended word extraction unit 13 may calculate the relevance by dividing the value obtained by adding the weighting coefficient to the document frequency or the appearance frequency by 2.

  Furthermore, the extended word extraction unit 13 may use the weighting coefficient as it is without regard to the appearance frequency of each word and the document frequency (relationship degree = weighting coefficient × 1). Even when the appearance frequency or document frequency does not make sense due to the nature of the searched document or word (for example, the searched document is biased to a specific field, or the word is commonly used in any field) This is because it is possible to calculate an appropriate degree of relevance.

  As described above, the document search device 10 uses the relevance level based on the distance between the word and the keyword (a value between 0 and 1) and the relevance level based on the document frequency or appearance frequency (a value between 0 and 1). Since the final relevance is derived and the extended words are extracted, it is possible to output a high-quality search result that is closer to the intention of the user (searcher).

  When the extended word extraction unit 13 determines the degree of relevance of each word to the keyword and extracts extended words, the document search apparatus 10 uses, for example, an extended syntax using a subquery as shown below as a command statement indicating a search request. And

select title from document where text contains 'environmental protection' expand from (select title from document where text contains 'warming' limit 10) distance factor 0.2
In the following, the above-described command sentence is divided into a plurality of parts for explanation.

select Title from Document where Text contains 'Environmental protection' (7)
expand from ... (8)
(select title from document where text contains 'warming' limit 10) (9)
distance factor 0.2 (10)
The part (7) is a search command for the document table defined in the document database 15. More specifically, “extract the title of a record that includes the word“ environmental protection ”in the body of the document in the document table. "Means the command.

  Further, the select statement (9) in the sub-query following the description “expand from” (8) is a search command for acquiring more seed documents. More specifically, extract the top 10 titles of records that contain the word “warming” in the body of the document in the document table. "Means the command.

  The part (10) indicates the degree of association based on the distance (number of characters or number of words) between each word in the seed document searched in (9) and a keyword (for example, the character string “warming”). This is a command for calculation, and the value “0.2” means a decreasing rate of the weighting coefficient that decreases as the distance between each word and the keyword increases.

  For example, the weighting factor is “weighting factor = 1 ÷ ((distance−1) ^ decrease rate)” or “weighting factor = 1−decrease rate × distance” (in this case, the minimum value of the weighting factor is 0). It is calculated using the mathematical formula shown by In any formula, the weight factor decreases more rapidly as the decrease rate is larger.

  Note that the reduction rate may be a fixed value, or a value may be directly input on the search request input screen 110, or a three-stage radio of “High”, “Middle”, and “Low”. You may make it select with a button. When any one of the “High”, “Middle”, and “Low” radio buttons is selected, the reduction rates are, for example, 0.8, 0.5, and 0.2, respectively.

  As described above, the document search apparatus 10 provides a simple method for designating the reduction rate, thereby enabling the degree of relevance of each word to the keyword to be adjusted in the degree of influence on the extraction of the expanded word, and the tendency of the search result. The convenience of the user (searcher) when adjusting (quality) can be improved.

  Further, the document search device 10 changes the decrease rate for each sentence (sentence), and the decrease rate is increased at a predetermined rate (for example, 10%) for each sentence as the sentence including the word moves away from the sentence including the keyword. You may make it reduce (For example, a reduction rate reduces for every sentence like 0.2%, 0.18%, 0.162% ...). In this case, the predetermined ratio may be set as the second argument of “distance factor”. In this case, the part (9) is, for example, a reduction rate of 0.2% and a predetermined ratio of 10%. It is described as “distance factor 0.2, 10”.

  For example, the document search apparatus 10 sets the weighting factor to 1 when the sentence including the word is the same as the sentence including the keyword, and sets the weighting coefficient at a predetermined rate as the sentence including the word moves away from the sentence including the keyword. The weight coefficient is set to 0 when a sentence including a word is separated from a sentence including a keyword by a predetermined number of sentences or more.

  As described above, the document search apparatus 10 includes words included in the same sentence as the sentence including the keyword but having a large distance to the keyword, but included in a sentence different from the sentence including the keyword but having a small distance to the keyword. High-quality search that is closer to the intention of the user (searcher) by preventing the relevance of the keyword from becoming lower than the generated word and setting the appropriate relevance and extracting the appropriate expanded words The result can be output.

  Subsequently, referring to FIG. 6 again, the extended word extraction unit 13 extracts words as extended words in descending order of relevance (weight coefficient × document frequency) (step S210). A document including the search condition (search word) input in the condition input area 111 and all or part of the extended word extracted by the extended word extracting unit 13 is searched from the set of documents in the document database 15 (step S211), the search result is presented to the user.

  As described above, the document search apparatus 10 according to the second embodiment determines the degree of relevance of a word to a keyword based on the distance between the word and the keyword included in the seed document, and extracts an extended word. A high-quality search result closer to the intention of the user (searcher) can be output.

  Note that the document search apparatus 10 according to the second embodiment extracts another document used by the user of the seed document as a related document, and sets the distance between the word included in the seed document or the related document and the keyword. Based on this, the degree of relevance of the word to the keyword may be determined to extract the extended word.

  As mentioned above, although the Example of this invention was explained in full detail, this invention is not limited to the above specific embodiment, In the range of the summary of this invention described in the claim, Various modifications or changes can be made.

  For example, in the above-described embodiment, the expanded word extraction unit 13 extracts a predetermined number of words as expanded words in descending order of relevance, but may extract all words having a relevance of a predetermined value or more as expanded words. Good.

  In the above-described embodiment, the seed document acquisition unit 12 acquires all or a part of other documents used by the user of the seed document as related documents, but the user of the seed document used it within a predetermined period. Only a document may be acquired as a related document. This is because it is considered that the relevance with the seed document is low when the use dates are separated in time.

  In the second embodiment, the extended word extraction unit 13 calculates the degree of association based on the shortest distance (minimum value) between each word and the keyword. However, the average distance (average value) and the longest distance (maximum) Value) and intermediate distance (intermediate value), the degree of association may be calculated.

It is a figure which shows the function structural example of the document search device in embodiment of this invention. It is a figure which shows the example of a display of a search request input screen. It is a figure which shows the hardware structural example of the document search apparatus in embodiment of this invention. It is a flowchart for demonstrating the document search process by the document search apparatus in 1st embodiment. It is a figure which shows the various tables utilized by the document search apparatus in 1st embodiment. It is a flowchart for demonstrating the document search process by the document search apparatus in 2nd embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Document search device 11 Search request input part 12 Seed document acquisition part 13 Extended word extraction part 14 Document search part 15 Document database 100 Drive apparatus 101 Recording medium 102 Auxiliary storage apparatus 103 Memory apparatus 104 Arithmetic processing apparatus 105 Display apparatus 106 Input apparatus 110 Search request input screen 111 Search condition input area 112 Character string input area for seed document acquisition 113 Seed number input area 114 Search button

Claims (11)

A document search device for searching a document from a predetermined document database based on an input search condition,
Seed document acquisition means for acquiring a seed document based on the input seed document acquisition character string;
Related document acquisition means for acquiring other documents used by the user of the seed document acquired by the seed document acquisition means as related documents;
Word extraction means for extracting words related to the search condition from the seed document and the related document;
Search means for searching for a document based on the search condition and the word extracted by the word extraction means;
A document search apparatus characterized by comprising: The related document includes, in addition to other documents used by a user of the seed document, other documents borrowed by a borrower of the seed document, other documents purchased by a purchaser of the seed document, or the seed document Including other documents viewed by
The document search apparatus according to claim 1. A document search device for searching a document from a predetermined document database based on an input search condition,
Seed document acquisition means for acquiring a seed document based on the input seed document acquisition character string;
A word extracting unit that extracts a word related to the search condition from the seed document acquired by the seed document acquiring unit;
Search means for searching for a document based on the search condition and the word extracted by the word extraction means,
The word extraction means determines a degree of association of each word with respect to the predetermined keyword based on a distance between each word and the predetermined keyword, and extracts a predetermined number of words in descending order of the degree of association;
A document search apparatus characterized by that. The word extraction means determines the relevance of each word to the keyword based on the frequency of appearance of each word or the number of seed documents including each word, in addition to the distance between each word and the predetermined keyword, Extracting a predetermined number of words in descending order of the degree of association;
The document search apparatus according to claim 3. A decrease rate setting means for setting a decrease rate of the degree of association that decreases as the distance between the word and the predetermined keyword increases;
5. The document search apparatus according to claim 3, further comprising: The rate of decrease varies from sentence to sentence,
The document search apparatus according to claim 5, wherein: A document search method for searching a document from a predetermined document database based on an input search condition,
A seed document acquisition step of acquiring a seed document based on the input seed document acquisition character string;
A related document acquisition step of acquiring, as a related document, another document used by a user of the seed document acquired in the seed document acquisition step;
A word extracting step of extracting a word related to the search condition from the seed document and the related document;
A search step for searching for a document based on the search condition and the word extracted in the word extraction step;
A document search method characterized by comprising: A document search method for searching a document from a predetermined document database based on an input search condition,
A seed document acquisition step of acquiring a seed document based on the input seed document acquisition character string;
A word extraction step of extracting a word related to the search condition from the seed document acquired in the seed document acquisition step;
A search step for searching for a document based on the search condition and the word extracted in the word extraction step,
The word extraction step determines a degree of association of the word with the predetermined keyword based on a distance between the word and the predetermined keyword, and extracts a predetermined number of words in descending order of the degree of association.
A document search method characterized by the above. A document search program for causing a computer to search for a document from a predetermined document database based on an input search condition,
A seed document acquisition step of acquiring a seed document based on the input seed document acquisition character string;
A related document acquisition step of acquiring, as a related document, another document used by a user of the seed document acquired in the seed document acquisition step;
A word extracting step of extracting a word related to the search condition from the seed document and the related document;
A search step for searching for a document based on the search condition and the word extracted in the word extraction step;
A document search program characterized by comprising: A document search program for causing a computer to search for a document from a predetermined document database based on an input search condition,
A seed document acquisition step of acquiring a seed document based on the input seed document acquisition character string;
A word extraction step of extracting a word related to the search condition from the seed document acquired in the seed document acquisition step;
A search step for searching for a document based on the search condition and the word extracted in the word extraction step,
The word extraction step determines a degree of association of the word with the predetermined keyword based on a distance between the word and the predetermined keyword, and extracts a predetermined number of words in descending order of the degree of association.
A document search program characterized by that.   A computer-readable recording medium in which the document search program according to claim 9 or 10 is recorded.

Images