JP4813312B2 - Electronic document search method, electronic document search apparatus and program - Google Patents

Description

  The present invention relates to an electronic document search technique that enables a search for an electronic document in a range that exceeds a search condition specified by a searcher.

  There is an increasing demand for efficient retrieval of important information contained in large-scale electronic documents such as digitized newspaper articles and documents on the web. In order to meet such demand, a search rule for searching a desired electronic document is created in advance, a morphological analysis is performed on the electronic document extracted from the electronic document group, and a search rule is used for the morphological analysis result. A search technique for searching for an electronic document that conforms to a search rule by performing pattern matching is known.

  FIG. 15 is a diagram for explaining an example of a conventional search method using morphological analysis. In the example shown here, in order to extract an electronic document (article) that matches the condition of “a proper noun and represents a region” and includes the word “bank”, “ A search rule of BANK = noun−proper noun−region “bank” ”is used. This is a search rule classified into the category “BANK” and is expressed by a logical product of a plurality of words. By performing pattern matching on the result of morphological analysis using this search rule, “AA bank” (AA is a proper noun representing a region name) and “BB bank” (BB Is a proper noun representing an area name different from AA).

  The search rules used in this search method are expressed by regular expressions using words and their part-of-speech names as constituent elements, and acquisition of grammar knowledge is indispensable for the creation. In order to alleviate this, various techniques for efficiently creating search rules have been proposed. For example, there is a technique for analyzing a component of an electronic document to be searched for and a combination pattern thereof and automatically outputting a frequently searched pattern as a search rule (for example, Patent Document 1).

  In addition, a similar document search is performed using the words searched by the existing search rules as search words, and a word having a high similarity to the search word is selected and selected from the searched similar documents. There is also a technique for replacing similar words with components of existing search rules (for example, Patent Document 2).

JP 2001-318792 A JP 2004-295797 A

  In the technique described in Patent Document 1, when a searcher does not include a part of the conditions necessary for searching an electronic document including a desired word in the search rule, the electronic document is searched without omission. Therefore, it becomes difficult to create a search rule. This situation will be specifically described with reference to FIG.

  Now, assume that electronic newspaper articles include the words “CC bank”, “DD bank”, “EE bank”, “XX bank”, and “FF securities” as financial institution names. CC, DD, EE, and FF are proper nouns that represent different area names, and OO is a proper noun that represents a meaning other than the area name. If the searcher wants to search for an article containing the above financial institution name, and the search terms specify the terms “CC bank”, “DD bank”, and “EE bank”, these three terms are all region names. Therefore, a search rule for searching only articles including “CC bank”, “DD bank”, and “EE bank” is generated. In other words, since “XX” is not a proper noun representing a region name, articles including “XX bank” are not searched by this search rule. Similarly, in this search rule, since the characters “securities” do not match the characters “bank”, articles including “FF securities” are not searched. As described above, unless the searcher inputs an appropriate search condition, a search omission occurs and an article including a desired word cannot be searched.

  In the technique described in Patent Document 2, there are cases where an inappropriate word is included as a constituent element of a search rule among words expanded in similar terms. In this case, an appropriate search rule is created. It becomes difficult. For example, as shown in FIG. 17, when searching for an article including the name of a financial institution, in the technique described in Patent Document 2, as an analogy of “bank” input by a searcher, for example, “exchange” ”,“ Interest rate ”and“ deposit ”are extracted. However, it is unlikely that these words represent financial institution names. For this reason, the search rule generated by replacing the search words with these words cannot appropriately search for articles including financial institution names.

  An object of the present invention is to provide a technique capable of searching for an electronic document even when a searcher does not appropriately specify a search condition for searching for an electronic document including a desired word. is there.

In order to solve the above problems, the present invention provides an electronic document search method, an electronic document search apparatus, and a computer program.
An electronic document search method provided by the present invention is an electronic document search method performed by an apparatus that enables a search that exceeds a range of search conditions specified in advance from a group of electronic documents, wherein the search condition is a primary search. A primary search step of accepting as a rule, searching the electronic document group by the primary search rule to obtain a primary document, extracting a word or phrase included in the primary document as a similarity search rule, By searching the electronic document group, one or a plurality of similar documents are obtained, and for each similar document, a similarity indicating how much the feature is similar to the primary document is calculated, and this similarity is associated with each similar document. A plurality of secondary search rule candidates that enable a search in a wider range than the search condition based on the similar search step and the primary search rule, The similar document is searched by each of the search rule candidates, and a word or phrase that is a search result is extracted together with the corresponding similarity, and the total value of the extracted similarities is obtained as a result of the search by the secondary search rule candidate. A search rule determining step that is derived as an evaluation value and determines a secondary search rule candidate satisfying a predetermined condition as a secondary search rule, and a second search for re-searching the electronic document group by the determined secondary search rule. And a next search step.
By this electronic document search method, for example, a secondary search rule for searching in a wider range than the search condition specified by the searcher can be obtained, so the search condition for searching for an electronic document containing a desired word is the searcher. Even in the case where it is not specified appropriately by the user, a desired electronic document can be searched. Further, the searchable range of a document similar to the primary document can be quantitatively known from the evaluation value.

  The primary search rule may include a logical condition of a plurality of words or phrases. In this case, the search rule determination step excludes, for example, some of the plurality of words or phrases when the logical condition is a logical product condition, and adds a new word or phrase when the logical condition is a logical sum. The secondary search rule candidate is generated by adding.

  The similar search step extracts, as the similar search rule, a word or phrase that is synonymous or similar to the word or phrase instead of or together with the word or phrase included in the primary document. You may make it do. By doing in this way, it is possible to search for many related documents whose contents approximate to those of the primary document.

The similarity search rule may be co-occurrence information whose frequency of co-occurrence in the primary document is higher than a predetermined value, instead of a word or phrase included in the primary document. This makes it possible to search for similar documents that are more relevant to the primary document, and the search range for the secondary search is further expanded. “Co-occurrence” refers to recalling other related words together with a certain word. “Co-occurrence information” is a co-occurrence word or phrase. The period for measuring the frequency may be arbitrary. For example, it can be one week.
In the search rule determination step, for example, a secondary search rule candidate having a relatively high evaluation value is determined as the secondary search rule.

  The electronic document search apparatus of the present invention extracts a primary document by searching an electronic document group using a primary search rule storage means for storing a search condition received from a searcher as a primary search rule, and the primary search rule. The electronic document group is searched by using words or phrases included in the document to find one or a plurality of similar documents, and for each similar document, a degree of similarity indicating how much the feature is similar to the primary document is calculated. A similar document management means for storing the similarity in association with each similar document, and generating a plurality of secondary search rule candidates that enable a search in a wider range than the search condition based on the primary search rule. The similar document is searched by each of the plurality of generated secondary search rule candidates, and a word or phrase as a search result is extracted together with the corresponding similarity, and the extracted similarity A total value is derived as an evaluation value of a search result by the secondary search rule candidate, and a secondary search rule candidate satisfying a predetermined condition for the evaluation value is used as a secondary search rule for re-searching the electronic document group. And a search rule determining means for determining.

  The computer program of the present invention includes a computer; primary search rule storage means for storing a search condition received from a searcher as a primary search rule; a primary document is extracted by searching an electronic document group according to the primary search rule, and the primary The electronic document group is searched by using words or phrases included in the document to find one or a plurality of similar documents, and for each similar document, a degree of similarity indicating how much the feature is similar to the primary document is calculated. A similar document management means for storing the similarity in association with each similar document; generating a plurality of secondary search rule candidates that enable a search in a wider range than the search condition based on the primary search rule; The similar document is searched by each of the plurality of generated secondary search rule candidates, and a word or a phrase as a search result is extracted together with the corresponding similarity. A total value of the extracted similarities is derived as an evaluation value of a search result by the secondary search rule candidate, and a secondary search rule candidate whose evaluation value satisfies a predetermined condition is re-searched for the electronic document group A computer program for functioning as search rule determination means for determining as a secondary search rule.

  According to the present invention, a secondary search rule related to the primary document searched in the primary search and enabling a wider range of search than the primary search is generated. Even if the search condition of the electronic document to be included is not correctly specified, a unique effect is obtained that the electronic document can be correctly searched.

Hereinafter, exemplary embodiments of the present invention will be described.
[Electronic document search device]
An example of the functional configuration of an electronic document retrieval apparatus to which the present invention is applied is shown in FIG. The electronic document retrieval apparatus 1 is realized by, for example, cooperation between a computer having an external storage device such as a hard disk and the computer program of the present invention.
That is, when the computer reads and executes the computer program of the present invention, the input / output I / F (“I / F” is an abbreviation of an interface, the same applies hereinafter) 2, rule management unit 3, search unit 4, evaluation unit 5. A search result output unit 6, a DB ("DB" is an abbreviation of database, the same applies hereinafter) management unit 7 functions, and a rule DB 11, a document DB 12, a file DB 13 in an external storage device accessible by the device. Each DB is constructed. The document DB 12 stores a large number of financial-related newspaper articles (electronic document group) that are digitized and managed so as to be identifiable by identifiers such as ID (Identification), and can be added, deleted, and changed as appropriate. It is like that.

  The input / output I / F 2 is an interface for accepting a word or phrase used for a search, for example, a type of part of speech or a word, input from a searcher through a keyboard or the like, and outputting a search result. In the present embodiment, when it is not necessary to particularly distinguish words or phrases used for the search, these are collectively referred to as “search words”.

  The rule management unit 3 selects a primary search rule, a similar search rule, and a secondary search rule candidate search rule that is a candidate for the secondary search rule based on a rule book developed in a predetermined area of the external storage device or a candidate thereof. This is generated and stored in the rule DB 11. The rule book stores a plurality of rules for generating a search rule, and a rule according to the purpose can be arbitrarily selected from these rules. When a plurality of search rule candidates are generated, the rule management unit 3 also performs a process for narrowing down each search rule candidate or determining a search rule.

The search unit 4 searches the various DBs 11 to 13 via the DB management unit 7. The evaluation unit 5 performs evaluation of the electronic document and the secondary search rule candidate, that is, calculation of similarity described later, derivation of an evaluation value, and the like. The search result output unit 6 temporarily stores the search results or stores the file management DB 13 via the DB management unit 7 and feeds it back to the rule management unit 3. The search result output unit 6 also outputs the search result via the DB management unit 7 to any output device such as a display device, an external recording device, or a printer (not shown). Output to the output device is when a final search result is obtained, for example, when an instruction input to that effect is received from the searcher.
The DB management unit 7 writes (accumulates) information in various DBs 11 to 13 and reads information from these DBs 11 to 13. Details of these functions will be described later.

Here, each search rule handled by the electronic document search apparatus 1 of the present embodiment will be described in more detail.
The primary search rule is a rule including a condition for performing a primary search with respect to a group of electronic documents stored in the document DB 12, and is expressed by a logical conditional expression such as a word or a phrase accepted by the input / output I / F2, for example. . Examples of words or phrases that accept input include part-of-speech types (nouns, proper nouns, adjectives, etc.), part-of-speech attributes (of proper nouns, regional names (including place names), specific words (generally bank names) ("Bank"", generally included in the name of the securities company), etc. The primary search rule is generated by taking the logical condition of the search term, for example, logical product (AND). An electronic document searched by the primary search rule is a “primary document”.

  The similarity search rule is a rule for searching for an electronic document similar to the primary document (this becomes a “similar document”) from the electronic document group stored in the document DB 12. In the present embodiment, a desired search word is extracted from one or a plurality of primary documents, and a logical conditional expression including the extracted search word as a search condition is set as a similar search rule.

  The secondary search rule candidate is a candidate for a secondary search rule for enabling a search in a wider range than the primary search rule while starting from the primary search rule. As a technique for expanding the search range, for example, the search term is increased in the case of the logical sum condition, the search term is deleted in the case of the logical product condition, or some of the search terms are searched using the primary search rule. It is conceivable to replace with words or phrases (such as those stored in advance in the internal memory) included in the document.

[Electronic document search method]
Next, an electronic document search method by the electronic document search apparatus 1 configured as described above will be described with reference to the overall procedure diagram of FIG. 2 and the operation explanatory diagrams of FIGS. Here, an electronic document that is truly desired is automatically selected beyond the search range specified by the searcher by inputting the search condition from among a group of electronic documents (many newspaper articles) stored in the document DB 12. An example for searching is given below.

Referring to FIG. 2, the electronic document search apparatus 1 first performs a primary search process (step S101). The primary search process is performed, for example, according to the procedure shown in FIG.
That is, when the “proprietary noun representing the area name” and “bank” are received as the primary search rules for searching for the bank name “BANK” through the input / output I / F 2, the rule management unit 3 sets the primary search rules as rules. Store in DB11. In this example, the primary search rule is “of nouns (condition 1), proper noun (condition 2), representing region (condition 3)”, and the word “bank” as a keyword (condition 4) Are included as search conditions for logical product determination. That is, the primary search rule is represented by the following logical conditional expression.
"BANK = noun-proper noun-area" bank ""

  The search unit 4 reads the primary search rule and performs a search process on the electronic document group stored in the document DB 12 based on a search request automatically or according to a searcher's instruction. A specific example of the contents of this search process is shown in FIG. In the example of FIG. 8, the electronic document (newspaper article) including the words “GG bank”, “HH bank” and “II bank” from the electronic document group shown in the upper right as shown in the lower right of the figure. Indicates that it is searched as a primary document that conforms to the primary search rule in the upper left of the figure. Here, “GG”, “HH”, and “II” are proper nouns representing different regional names.

  The search result output unit 6 stores the primary document as a primary search result by the search unit 4 and a search word extracted from the primary document, that is, a word or a phrase, in the file DB 13 as a primary document file. The search term can be extracted by, for example, performing morphological analysis of the primary document by a known morphological analysis method.

Returning to FIG. 2, when the primary search process is completed, the electronic document search apparatus 1 performs a similarity search process (step S102). The similarity search process is performed, for example, according to the procedure shown in FIG.
First, the rule management unit 3 reads a search word from the primary document file, and generates a similar search rule including this search word as a search condition. And the search part 4 searches the electronic document group accumulate | stored in document DB12 by this similar search rule. By this search, one or a plurality of similar documents are searched. The evaluation unit 5 calculates the degree of similarity for each retrieved similar document. The degree of similarity is a measure that represents how similar a similar document is to the primary document. In the present embodiment, a numerical value that increases as the degree of similarity increases as a scale.

That is, the evaluation unit 5 extracts words or phrases included in each of the similar document and the primary document, calculates each feature vector, and numerically represents the degree of similarity between the feature vector in the similar document and the feature vector in the primary document. It expresses with. Such calculation of the similarity can be performed, for example, by a TF / IDF (Term Frequency Inverse Document Frequency) method. Regarding the TF / IDF method, for example, the description of “Language and Calculation 5 Information Retrieval and Language Processing”, Takenobu Tokunaga, University of Tokyo Press 1999/11 can be referred to.
Note that the calculation of the similarity is not limited to the calculation method based on the TF / IDF method, and may be calculated by other methods. When the evaluation unit 5 calculates the similarity as described above, the evaluation unit 5 performs processing for associating the similarity with each similar document. For example, an ID assigned to a similar document and a similarity are linked.

FIG. 9 shows a specific example of the contents of the processing so far. In the example shown in FIG. 9, “ATM”, “handling”, “finance”, “institution”, “product”, “unsecured”, “loan”, and “balance” are used as search terms extracted from the primary document. Is listed. This search term is an example, and one or a plurality of morphemes included in the primary document can be extracted as the search term. For example, only a noun phrase may be selected as a search word. The lower part of FIG. 9 represents the degree of similarity associated with each similar document and each primary document. Similar document (a) has similarity “72.1”, similar document (b) has similarity “68.6”, similar document (c) has similarity “66.4”, and similar document (d) has similarity “62.8” and the similar document (e) have the similarity “59.5”. The example of FIG. 9 shows that the similar document (a) is the closest to the primary document.
The search result output unit 6 stores the similar document and the similarity in the file DB 13 as a similar document file.

  Returning to FIG. 2, when the similarity search process is completed, the electronic document search apparatus performs a secondary search rule candidate generation process (step S <b> 103). In the secondary search rule candidate generation process, for example, as shown in FIG. 5, the rule management unit 3 enables a search in a wider range than the primary search rule based on the primary search rule stored in the rule DB 11. The process starts by generating a plurality (X) of secondary search rule candidates. An example of the generated secondary search rule candidate is shown in FIG. Each secondary search rule candidate is generated as follows.

(A) The condition “region” which is “condition 3” in the primary search rule “BANK = noun-proper noun-region“ bank ”” is excluded.
(B) to (e) Excluding “condition 3” in the primary search rule “BANK = noun-proper noun-region“ bank ”” and “condition 4” (condition for extracting the word “bank”) Are replaced with other words to be extracted in the search terms “exchange”, “interest rate”, “securities” and “deposit” extracted in step S3.
In addition to the above, a secondary search rule candidate in which the condition for extracting all search terms other than “exchange”, “interest rate”, “securities” and “deposit” is replaced with “condition 4”. Good.

  After the secondary search rule candidate is generated, the process proceeds according to the procedure shown in FIG. That is, the rule management unit 3 extracts the first secondary search rule candidate and passes it to the search unit 4. The search unit 4 performs a secondary search process using the secondary search rule candidates in cooperation with the evaluation unit 5 (steps S104 and S105).

  The secondary search process is performed as follows. That is, the search unit 4 searches for similar document files using secondary search rule candidates, extracts search results and corresponding similarities, and temporarily holds these as secondary search results. The evaluation unit 5 takes the sum of the similarities (similarities given to the original similar document) given to the words or phrases as the search results held temporarily, and performs the secondary search on the total value. Derived as an evaluation value of a search result by a rule candidate. Then, the evaluation value is passed to the search result output unit 6 together with the secondary search result and the secondary search rule candidate.

  An example of the contents of this secondary search process is shown in FIGS. In the example of FIG. 11, as a search result by the secondary search rule candidate (A) (= “BANK = noun−proper noun“ bank ””), “XX bank”, “II bank”, “HH bank”, “ GG Bank "is obtained. From “XX Bank”, the similarity “72.1” of the extracted similar document (a), from “II Bank”, the similarity “68.6” of the similar document (b), from “HH Bank” Since the similarity “66.4” of the similar document c and the similarity “62.8” of the similar document (d) are extracted from “GG bank”, the evaluation unit 5 adds these similarities. Thus, the evaluation value of the search result by the secondary search rule candidate (A) is “269.9” which is the total value.

Similarly, in the example of FIG. 12, “JJ Securities” is obtained as a search result based on the secondary search rule candidate (d) (= “BANK = noun−proper noun“ securities ””). Since only one word is extracted, the evaluation value of the search result by the secondary search rule candidate (d) is the same value as the similarity “59.5” of the similar document (e) from which “JJ Securities” is extracted. Become.
The search result output unit 6 associates the evaluation value thus derived with the secondary search rule candidate and the secondary search result, and accumulates them in the file DB 13 as a secondary document file.
The above secondary search process is repeated for all X secondary search rule candidates (steps S106 and S107: No).

When the secondary search process for the X secondary search rule candidates is completed (step S107: Yes), the electronic document search apparatus 1 performs a secondary search rule determination process (step S108) and a re-search process (step S109). . These processes are performed, for example, according to the procedure shown in FIG.
The secondary search rule determination process is performed by the rule management unit 3. The rule management unit 3 extracts an evaluation value recorded in the secondary document file, and selects a secondary search rule candidate satisfying a predetermined condition, for example, a secondary search rule candidate having a relatively high evaluation value. The next search rule is determined and stored in the rule DB 11. It is also possible to set the predetermined condition to exceed the evaluation value and a threshold value stored in advance.

  An example of specific contents of the secondary search rule determination process is shown in FIG. FIG. 13 shows examples of evaluation values of five secondary search rule candidates (A) to (E) derived by the procedure described so far. The secondary search rule candidate (A) is “269.9”, the secondary search rule candidate (B) is “0.0”, the secondary search rule candidate (C) is “0.0”, and the secondary search rule candidate (D) is “59.5” and the secondary search rule candidate (e) is “0.0”. The evaluation value of secondary search rule candidates (b), (c), and (e) is “0.0”, which matches these secondary search rule candidates (b), (c), and (e) This is because the word to be found does not exist in the similar documents (a) to (e). If the predetermined condition is a relatively high one, the secondary search rule candidate (A) “BANK = noun-proper noun“ bank ”” is determined as the secondary search rule. On the other hand, if the predetermined condition is that the threshold value is set to “50”, there are two secondary search rule candidates (A) and (D) whose evaluation value exceeds the threshold value. Therefore, in this case, in addition to the secondary search rule candidate (b), the secondary search rule candidate (d) “BANK = noun-proper noun“ securities ”” is also determined as the secondary search rule.

  The re-search process is performed in cooperation with the search unit 3 and the search result output unit 6. The search unit 3 re-searches the electronic document group stored in the document DB 12 using the determined secondary search rule, and thereby, the electronic document (newspaper article, this search document that the searcher really wants is searched. ”) Is retrieved. The search unit 4 passes these search documents to the search result output unit 6. The search result output unit 6 stores the search document as a re-search result in the file DB 14.

  An example of the contents of this re-search process is shown in FIG. In the example of FIG. 14, the above-described two secondary search rule candidates (“BANK = noun-proper noun“ bank ”” and “BANK = noun-proper noun“ securities ””) are determined as secondary search rules. The re-search results are shown. The search documents retrieved by the two secondary search rules in this way include newspaper articles including “GG Bank”, “II Bank” and “HH Bank”, as well as “XX Bank” and “JJ Securities”. Newspaper articles that contain From the viewpoint of financial institution name, it can be seen that the scope of newspaper articles to be searched is expanded because banks and securities are the same.

  As described above, according to the electronic document search apparatus 1 of the present embodiment, the secondary search rule having the search range wider than the search condition input by the searcher is automatically generated. Even if it is not properly specified, the newspaper article that the searcher really wants can be easily searched.

  Embodiments of the present invention are as described above, but the present invention is not limited to the above-described embodiments. For example, in the present embodiment, an example in which the electronic document group is stored in the document DB 12 in the system has been described. However, the electronic document group may be any area as long as the electronic document search apparatus 1 is accessible. May be present. For example, it may exist in an arbitrary server of a computer network to which the electronic document search apparatus 1 is connected.

  In the present embodiment, an example in which the search unit 3 searches for the primary document from the electronic document group stored in the document DB 12 has been described. However, a specific primary specified in advance in the electronic document group is described. The document may be read by the rule management unit 3.

  Further, in the present embodiment, for example, in the primary search rule (= “BANK = noun-proprietary noun-region“ bank ””) classified into the category “BANK”, “of nouns (condition 1), The case where the search condition is that a plurality of conditions, “proprietary nouns (condition 2) and representing the region (condition 3)” are proper nouns representing the region, is explained. The part of speech is not limited to a noun, and another part of speech may be used as a search condition. Furthermore, the search condition is not limited to the part of speech as long as it is a condition for searching for a desired electronic document.

  Moreover, although the case where the secondary search rule candidate is based on the replacement of the search word has been described, it may be generated by replacing the search word with a word that is synonymous or similar to the search word. In addition, the search rule generation unit 2 extracts a word or phrase whose co-occurrence frequency in the primary document is higher than a predetermined threshold and stores it, and is expressed by “condition 4” in the primary search rule. The keyword may be replaced with this word or phrase to generate a secondary search rule candidate.

  In addition, when the evaluation value is derived, each secondary search rule candidate and the evaluation value derivation result are presented to the searcher so that the secondary search rule candidate designated by the searcher can be determined as the secondary search rule. May be. In this case, the searcher can select an arbitrary secondary search rule candidate having a high evaluation value by his / her own judgment, and an advantage that a search rule more reflecting the searcher's taste can be generated.

1 is a functional configuration diagram of an electronic document search apparatus according to an embodiment of the present invention. FIG. 6 is a processing procedure diagram of an electronic document search method according to the present embodiment. Explanatory drawing of the procedure of a primary search process. Explanatory drawing of the procedure of a similar search process. Explanatory drawing of a secondary search rule candidate production | generation process. Explanatory drawing of the procedure of a secondary search process. Explanatory drawing of the procedure of a secondary search rule determination process and a re-search process. Explanatory drawing which shows the search result example of a primary document. Explanatory drawing which shows the outline | summary of search result extraction of a search word, and a similar document. Explanatory drawing which shows the example of the secondary search rule candidate produced | generated from a primary search rule. Explanatory drawing which shows the relationship between the word extracted from a similar document, similarity, and an evaluation value. Explanatory drawing which shows the relationship between the word extracted from a similar document, similarity, and an evaluation value. Explanatory drawing which shows the relationship between a secondary search rule candidate and evaluation value. Explanatory drawing which shows the search document searched out from the electronic document group by the secondary search rule. The figure which shows the conventional search method using a morphological analysis. The figure which shows the conventional search method using a morphological analysis. The figure which shows the search method using the conventional similar word expansion | deployment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Electronic document search apparatus, 2 ... Input / output I / F, 3 ... Rule management part, 4 ... Search part, 5 ... Evaluation part, 6 ... Search result output part, 7 ... DB management part, 11 ... Rule DB, 12 ... document DB, 13 ... file DB.

Claims (6)

An electronic document search method performed by an apparatus that enables a search that exceeds a range of search conditions specified in advance from an electronic document group,
A primary search step of accepting the search condition as a primary search rule and obtaining the primary document by searching the electronic document group according to the primary search rule;
A word or phrase included in the primary document is extracted as a similar search rule, and the electronic document group is searched by the extracted similar search rule to obtain one or a plurality of similar documents. For each similar document, the primary document and A similarity search step for calculating a similarity indicating how similar the features are, and storing the similarity in association with each similar document;
Based on the primary search rule, a plurality of secondary search rule candidates that enable a search in a wider range than the search condition are generated, and the similar document is searched by each of these secondary search rule candidates to perform a search. A word or phrase as a result is extracted together with the corresponding similarity, and a total value of the extracted similarities is derived as an evaluation value of a search result by the secondary search rule candidate, and the evaluation value satisfies a predetermined condition A search rule determination step for determining secondary search rule candidates as secondary search rules;
A secondary search step of re-searching the electronic document group according to the determined secondary search rule. The primary search rule includes a logical condition of a plurality of words or phrases,
The search rule determination step excludes some of the plurality of words or phrases when the logical condition is a logical product condition, while adding a new word or phrase when the logical condition is a logical sum. Generate secondary search rule candidates,
The electronic document search method according to claim 1. The similar search step extracts, as the similar search rule, a word or phrase that is synonymous or similar to the word or phrase instead of or together with the word or phrase included in the primary document. To
The electronic document search method according to claim 2. The search rule determination step determines a secondary search rule candidate having a relatively high evaluation value as the secondary search rule.
The electronic document search method according to claim 1. Primary search rule storage means for storing search conditions received from a searcher as a primary search rule;
The electronic document group is searched by the primary search rule to extract a primary document, the electronic document group is searched by a word or phrase included in the primary document, and one or a plurality of similar documents are searched, and each similar document is searched. A similarity document management means for calculating a similarity indicating how similar the feature to the primary document is, and storing the similarity in association with each similar document;
Based on the primary search rule, a plurality of secondary search rule candidates that enable a search in a wider range than the search condition are generated, and the similar document is searched by each of the generated plurality of secondary search rule candidates. , A word or phrase that is a search result is extracted together with the corresponding similarity, and a total value of the extracted similarities is derived as an evaluation value of the search result by the secondary search rule candidate, and the evaluation value is a predetermined condition A search rule determining means for determining a secondary search rule candidate that satisfies the above as a secondary search rule for performing a re-search of the electronic document group,
Electronic document search device. A computer;
Primary search rule storage means for storing a search condition received from a searcher as a primary search rule;
The electronic document group is searched by the primary search rule to extract a primary document, the electronic document group is searched by a word or phrase included in the primary document, and one or a plurality of similar documents are searched, and each similar document is searched. A similar document management means for calculating a similarity indicating how much the feature is similar to the primary document, and storing the similarity in association with each similar document;
Based on the primary search rule, a plurality of secondary search rule candidates that enable a search in a wider range than the search condition are generated, and the similar document is searched by each of the generated plurality of secondary search rule candidates. , A word or phrase that is a search result is extracted together with the corresponding similarity, and a total value of the extracted similarities is derived as an evaluation value of the search result by the secondary search rule candidate, and the evaluation value is a predetermined condition A search rule determining means for determining a secondary search rule candidate satisfying as a secondary search rule for re-searching the electronic document group;
Computer program to function as.

Images