JP2004054882A - Synonym retrieval device, method, program and storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a synonym retrieval device, method, program and storage medium capable of inexpensively acquiring a synonym with universality when determining a synonym of a search term of a document group. <P>SOLUTION: A primary retrieval part 5 searches a document storing part 5 by using an object term of an object term buffer 2, and it attaches a document score to a retrieved document. The document score is larger in proportion to more number of search terms, shorter document length, and smaller number of documents. High ranking R pieces of documents with high document scores are considered as seed documents, the seed documents are disassembled into words, a degree of association with the object word is determined by predetermined calculation in regard to each word as a related term candidate, and high ranking T pieces with high degree of association are regarded as related terms. A secondary retrieval part 8 determines a seed document by searching the document storing part 3 again in the same way as the primary retrieval by using a related term group stored in a related term buffer 7 as a search term group, and it extracts related terms from the seed document. A synonym selecting part 9 selects high ranking S pieces of extracted related terms as synonym candidates. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a synonym search device and method for obtaining a synonym of a search word in a document search, a synonym search program used in the device, and a storage medium storing the synonym search program.
[0002]
[Prior art]
In a document search for extracting a document related to a search term (keyword) from a group of documents such as a large number of electronic documents stored in a database or the like, when a searcher gives a search term A, the search target document includes A synonym B of the search word A may be used. In that case, as long as the search process is performed based on the word form, the document cannot be searched and extracted even though it is highly likely that the document matches the search request. This problem has been generally pointed out as a lexical gap problem.
Here, the synonyms are, for example, synonyms at the notation level: “computer” and “computer”
・ Vocabulary level synonyms: “Book” and “Book”
Would say something like
[0003]
Therefore, if synonyms for the search term in the search request can be obtained, search performance can be expected to be improved by including those synonym groups in the search request. For this reason, for example, the following proposal has been made regarding a method of automatically acquiring a synonym group, for example.
A synonym is acquired from a synonym file prepared in advance for each constituent word of the target character string, and a synonym expression for the target character string is configured by combining them (for example, Patent Document 1).
A synonym is extracted from a group of documents by using a pattern such as "is a ..." (for example, Patent Document 2).
[0004]
A relation between a compound word and its constituent words is regarded as a synonymous relation, and a compound word having the input word as a constituent word is output as a synonym (for example, Patent Document 3).
-When registering an input word in an existing thesaurus, the constituent words of the input word are obtained, and the registration position of the input word is determined based on the registration positions in the thesaurus (for example, Patent Document 4).
-Search word groups input by the user in the past are stored, and synonyms (co-occurring words) for the current input word are extracted from them (for example, Patent Document 5).
A pair between a parenthesis and a preceding word is extracted as a synonymous relationship (for example, Patent Document 6).
[0005]
[Patent Document 1]
JP-A-6-162098 (Patent No. 3025724)
[Patent Document 2]
Japanese Patent Application Laid-Open No. 6-266770 [Patent Document 3]
Japanese Patent Application Laid-Open No. 7-319884 [Patent Document 4]
Japanese Patent Application Laid-Open No. 8-222427 [Patent Document 5]
JP-A-9-319767 [Patent Document 6]
JP-A-11-328205
[Problems to be solved by the invention]
However, the above-described conventional technology requires a synonym dictionary, a thesaurus, or a rule group for synonym extraction, and the like. However, these data groups must be manually developed and created in advance, which requires a large cost. There was a problem. Further, even if rules can be easily prepared by using parenthesis expressions or synonym expressions, the frequency at which those expressions actually appear is generally low, and therefore, the number of synonyms obtained is small. That is, there is a problem that the generality is lacking.
[0007]
SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and provides a synonym search apparatus, a method, a program, and a storage medium that can acquire general synonyms at a lower cost than conventional ones. With the goal.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, in a synonym search device according to the present invention, an input unit for inputting a target word, a document storage unit for storing a group of documents, and the document storage unit using the input target word as a search word A means is searched, a document score is given to the document in which the search term appears by the first operation, the document score is ranked in descending order of the document score, a predetermined number of documents having the highest ranking are taken out as a seed document, and the seed document is constructed. Is extracted as a related word candidate, the seed document is searched using the extracted related word candidate as a search word, and the degree of relevance between the target word and the search word is determined by a second operation for the document in which the search word appears. Primary search means for determining and extracting, from the related word candidates, a predetermined number of search words in the descending order of the degree of relevance as related words, and using the extracted related words as search words in the document Secondary search means for searching the storage means and extracting the related words by the first and second operations with respect to the document in which the search word appears, and from the related words extracted by the secondary search means in descending order of relevance. Synonym selecting means for selecting a predetermined number of related words as synonym candidates.
[0009]
Further, the synonym search method according to the present invention searches a document group using the input target word as a search word, assigns a document score to the document in which the search word appears by a first operation, and ranks the documents in descending order of the document score. A predetermined number of documents in the top ranking are taken out as seed documents, words constituting the seed documents are extracted as related word candidates, and the extracted related word candidates are used as search words to search the seed document. The degree of relevance between the target word and the search word is calculated for the appearing document by a second operation, and a predetermined number of search words having a higher rank are extracted from the relevant word candidates in the descending order of the degree of relevance as related words. Searching the document storage means using a related word as a search word, extracting a related word from the document in which the search word appears by the first and second operations, and determining a degree of relevance from the extracted related word. Has a top predetermined number of related words is selected as synonyms candidates in descending order.
[0010]
Further, the program according to the present invention provides an input process for inputting a target word, a search for a document group using the input target word as a search word, and a document score is given to the document in which the search word appears by a first operation. A ranking process for ranking documents in descending order of document score, extracting a predetermined number of documents having the highest ranking as a seed document, and extracting words constituting the seed document as related word candidates; Is used as a search word to search the seed document, and for a document in which the search word appears, a second operation is performed to determine the relevance between the target word and the search word. An extraction process of extracting a search word as a related word, and searching the document group using the extracted related word as a search word. A program for causing a computer to execute an extraction process of extracting related words by a second operation, and a selection process of selecting a predetermined number of related words as synonym candidates in descending order of relevance from the extracted related words. is there.
[0011]
Further, a storage medium according to the present invention stores the above-mentioned program.
[0012]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
In the present embodiment, for a target word, a document group (collection) is searched by ranking, and a related word group of the target word is extracted from a word group appearing in a higher-order document group thereof, and only the related word group is used. The above object is achieved by performing a ranking search of the collection again and obtaining a synonym candidate using the degree of relevance from a group of related words appearing in a document group that does not include the target word, from among the higher-order document group.
[0013]
FIG. 7 is a block diagram showing a synonym search device according to an embodiment of the present invention.
As shown in the figure, the apparatus includes an input device 20 for performing a target word input operation by a user, an output device 30 for outputting a synonym candidate obtained as a search result, a storage device 40 such as a program storage ROM, a work RAM, and the like. Is controlled by the CPU 50.
[0014]
FIG. 1 is a block diagram showing a configuration of a control unit in a CPU according to the first embodiment of the present invention.
In FIG. 1, 1 is an input unit for receiving a target word (here, a search word for searching for a synonym) input from the input device 20 by the user, 2 is a target word buffer for storing the input target word, and 3 is a target word buffer. A document storage unit in which a group of documents is stored, 4 is a document search unit for searching the document storage unit 3, and 5 is a ranking search of the document storage unit 3 via the document search unit 4 using the target word of the target word buffer 2 as a search word. In addition, a primary search unit for obtaining a related word candidate group and obtaining a related word therefrom, a related word extracting unit for extracting a related word from the related word candidate group, and a related search unit for storing the extracted related word A word buffer, 8 is a secondary search unit that rank-searches a document group using related words and extracts related words, 9 is a synonym selection unit that selects synonym candidates from the secondary searched related words, and 10 is a synonym. An output unit for outputting.
[0015]
Next, the operation of the above configuration will be described.
FIG. 2 is a flowchart schematically showing the operation of the control unit.
First, a target word is input via the input unit 1, and the input target word is stored in the target word buffer 2 (step S1, hereinafter abbreviated to step). Next, a primary search of the document storage unit 3 is performed by the primary search unit 5 (S2), and a secondary search is performed by the secondary search unit 8 based on the search result (S3). Next, a synonym candidate is selected by the synonym selection unit 9 based on the result of the secondary search (S4), and the word form of the synonym candidate is output by the output unit 10 (S5).
[0016]
FIG. 3 is a flowchart showing the operation of the primary search unit 5.
The primary search unit 5 first performs the following document ranking search via the document search unit 4 using the target word in the target word buffer 2 as a search word (S11). That is, a document score defined next for each document stored in the document storage unit 3 is calculated.
score = {tf / (tf + dlen)} × weight (1)
weight = Log (N / n + 1) (2)
[0017]
In the above equation (1), tf is the frequency of occurrence of the search word (target word) in the document, and dlen is the document length. According to this score definition, the greater the number of search words, the shorter the document length, and the greater the weight (weight), the greater the score given to the document.
In the above equation (2), N is the total number of documents stored in the document storage unit 3, and n is the number of documents (document frequency) in which the search term appears in the total N documents. According to the weight definition formula, the weight of the search term increases as the number of occurrences in a small number of documents increases.
[0018]
When there are a plurality of search terms, the above-mentioned scores are obtained for each search term, and a final document score is obtained by adding them. A document in which no search word appears has a score of 0.
By the document ranking search, each document is sorted and output in the order of the document score.
[0019]
Next, the primary search unit 5 retrieves the documents ranked higher in the order of a predetermined number R from the document storage unit 3 in order from the higher order based on the result of the document ranking (S12). These R upper documents are referred to as seed documents here. Next, documents are extracted one by one from the seed document (S12), and a related word candidate group is obtained for the document D by the related word extracting unit 6 as follows (S13).
[0020]
First, the document D is decomposed into words using morphological analysis or word delimiters. Then, the following attribute is obtained for each word as a related word candidate.
Document frequency n: By performing a document search using the word as a search word, the document frequency of the word (the number of appearing documents in the total number N of documents) is obtained.
-Number of appearing seed documents r: Number of seed documents in which the word appears in the total number R of seed documents-Weight: obtained by the above equation (2).
Selection value (tsv) = weight × (r / R−n / N) (3)
[0021]
After obtaining the related word candidate group (word group) and its attributes for all the seed documents as described above (S14), the primary search unit 5 sorts the candidates in descending order of the selection value, and The top T items are selected as related words. However, in that case, related word candidates having the same word form as the target word are not regarded as related words. Next, the primary search unit 5 stores the pair of the word form of each related word and the selected value in the related word buffer 7. The selection value increases as the weight increases and as the probability of co-occurrence with the target word in the document increases. Therefore, the selection value indicates the degree of association between the target word and the related word candidate, that is, the degree of association.
[0022]
FIG. 4 is a flowchart showing the operation of the secondary search unit 8.
The secondary search unit 8 performs a document ranking search of the document storage unit 3 via the document search unit 4 using the related word group stored in the related word buffer 7 as a search word group (S21). At this time, the target word is not included in the search word group. A seed document is obtained in the same manner as in the case of the primary search unit 5 (S22), and related words are extracted from the seed document. However, a higher-level document including the target word is not regarded as a seed document (S23, S24). This is because, in general, a document including a target word uses the word consistently in the document, and a synonym is often not used. The related words obtained as a result of the secondary search for all the seed documents are stored in the related word buffer 7 (S25).
[0023]
As a result of the secondary search, the synonym selection unit 9 selects the top S words from the related word group obtained in the related word buffer 7 in descending order of relevance (selection value) as synonym candidates. This synonym is output from the output unit 10.
[0024]
FIG. 5 is a block diagram showing a configuration of a control unit according to the second embodiment of the present invention. Parts corresponding to those in FIG. 1 are denoted by the same reference numerals, and redundant description is omitted.
In the present embodiment, a determination unit 11 is added between the target word buffer 2 and the related word buffer 7.
[0025]
FIG. 6 is a flowchart showing the operation of the control unit.
First, a target word is input via the input unit 1, and the input target word is stored in the target word buffer 2 (S31). Next, the primary search section 5 performs a primary search on the document storage section 3 (S32). Next, the determination unit 11 determines whether or not to continue the processing (S33). If not, a message indicating that no synonym candidate was obtained is output from the output unit 10 (S34). Next, the secondary search unit 8 performs a secondary search of the document storage unit 3 based on the result of the primary search (S35). Then, based on the result of the secondary search, the synonym selection unit 9 selects a synonym candidate (S36), and the output unit 10 outputs the synonym candidate word form (S37). Steps S31, S32, and S35 to S37 are performed in the same manner as S1 to S5 in FIG.
[0026]
In step S33, the determination unit 11 determines whether or not to continue the processing as described below. That is, the related word group obtained as a result of the primary search is sorted in the order of the degree of relevance, and if the target word is not in the top k places in the ranking of the related words, it is determined that the processing is not continued. Otherwise, it is determined that the processing is to be continued. In other words, the word that should have the highest degree of relevance to the target word is the target word itself, and it is determined that a related word with a lower validity is obtained as the word goes out of the situation, and in that case, the processing is not continued.
[0027]
According to the present embodiment, for a target word for which a synonym cannot be obtained due to a lack of a document group or the like, a message to that effect is output, so that the quality of the synonym acquisition result can be improved.
In S5 of FIG. 2 and S37 of FIG. 6, only the synonymous word form is output by the output unit 10, but the relevance stored in the related word buffer 7 may be output together with the synonymous word form. Good. By doing so, the relevance of each synonym candidate is output as information for further narrowing down synonym candidates, so that the quality of the synonym acquisition result can be improved.
[0028]
Next, third to sixth embodiments of the present invention will be described. FIGS. 8, 12, 13, and 16 show the configuration of the control unit according to each embodiment. Parts corresponding to those in FIGS. 1 and 5 are denoted by the same reference numerals, and redundant description is omitted. The configuration of the synonym search device is the same as the configuration in FIG.
[0029]
FIG. 8 is a block diagram showing a configuration of a control unit according to the third embodiment of the present invention.
The primary search unit 5 and the secondary search unit 8 in FIGS. 1 and 5 are not shown.
The synonym selecting section 9 is illustrated as a synonym extracting section 12.
[0030]
FIG. 9 is a flowchart showing the operation of the control unit.
The target word is received using the input unit 1, and the target word is stored in the target word buffer 2 (S41). Next, a primary search is performed (S42), and a secondary search is performed (S43). Next, a similar word group is output using the output unit 10 (S44).
[0031]
FIG. 10 is a flowchart showing the operation of the primary search.
In the primary search, first, a document ranking search is performed using the document search unit 4 with the target word as a search word (S51). The document ranking search is performed by calculating a document score using the above equations (1) and (2). Next, from the result of the document ranking, a predetermined number (referred to as R) of upper documents (documents ranked higher) are extracted from the document storage unit 3 in order from the upper one (S52). These higher-level documents are referred to herein as seed documents. Next, a related word candidate group is obtained for each seed document D using the related word extracting unit 6 (S53). The related word candidate group is obtained by decomposing the document D into words using morphological analysis or word delimiters or the like, and using the selection value obtained by the above equation (3).
[0032]
Next, in the primary search, after a related word candidate group and its attributes have been obtained for all seed documents (S54), each candidate is sorted in descending order of selection value, and based on a predetermined number of related words T, Let T be related words. However, a related word candidate having the same word form as the target word is not regarded as a related word. Next, the pair of the word form and the selected value of each related word is stored in the related word buffer. The selection value increases as the weight increases and as the probability of co-occurrence with the target word in the document increases. Therefore, the selection value indicates the degree of association between the target word and the related word candidate (referred to as the degree of association).
[0033]
FIG. 11 is a flowchart showing the operation of the secondary search.
In the secondary search, first, a document ranking search is performed using the related word group stored in the related word buffer 7 as a search word group using the document search unit 4 (S61). Next, based on the result of the document ranking, a predetermined number (referred to as R2) of upper documents (documents ranked higher) are extracted from the document storage unit 3 in order from the upper one (S62). These upper documents are referred to as second seed documents D here. Next, a synonym candidate group is obtained for each second seed document D using the synonym extraction unit 12 as follows (S63).
[0034]
That is, the document D is decomposed into words using morphological analysis or word delimiters. Then, the following attributes are obtained for each word s.
Co-occurrence degree c: The total number of related words that appear at positions within W words before and after each occurrence position of word s in the second seed document. Here, W is a predetermined constant.
That is, the co-occurrence degree c for the word s is the total number of related words that appeared near s in the second seed document group.
[0035]
Note that c may be a relative frequency where the appearance frequency tf of s in the second seed document group is 1. In this case, c '= c / tf
It is.
[0036]
Next, words s and c or s and c ′ are pooled as synonym candidate expressions. Then, when synonym candidates have been obtained for all the second seed documents (S64), each candidate is sorted in descending order of the co-occurrence degree c, and the top S words are defined as synonyms based on a predetermined number S of synonyms. I do. The output unit 10 outputs the above S synonyms.
[0037]
When outputting the synonym, a synonym having the same form as the target word may be output. Therefore, in the fourth embodiment of the present invention, such trivial synonyms are output. FIG. 12 shows a control unit according to the fourth embodiment, in which the output unit 10 compares the word form of a target word stored in the target word buffer 2 with a synonymous word form, and outputs no output if they are the same.
[0038]
As described above, according to the third and fourth embodiments, a document group is ranked and searched for a target word, a related word group of the target word is obtained from a higher order document group, and the document is re-used using the related word group. The group is searched by ranking, and a synonym group is obtained from the upper document group and the related word group.
[0039]
FIG. 13 is a block diagram showing a configuration of a control unit according to the fifth embodiment of the present invention, which is obtained by adding a supplementary word buffer 13 to FIG.
[0040]
FIG. 14 is a flowchart showing the operation of the control unit.
First, a target word and a supplementary word group are received using the input unit 1, the target word is stored in the target word buffer 2, and the supplementary word group is stored in the supplemental word buffer 13 (S71). Next, a primary search is performed, and further a secondary search is performed (S72, S73). Then, a similar word group is output from the output unit 10 (S74).
[0041]
FIG. 15 is a flowchart showing the operation of the primary search.
The difference from FIG. 10 in S81 to S84 is that the document search unit 4 performs a document ranking search using not only the target word but also a supplementary word group in S81.
[0042]
Here, supplementary words will be described.
For example, the target word “TCP” is a polysemy having the following two meanings.
Meaning 1: Network protocol Meaning 2: Tricalcium phosphate When a synonym of meaning 1 is obtained, "network, communication" or the like may be given as a supplementary word group.
[0043]
The secondary search is performed in the same manner as in FIG. 11, and the output unit 10 outputs the S synonyms. In this case, since a synonym having the same form as the target word may be output, in the sixth embodiment of the present invention, such a synonym is output excluding such obvious synonyms. FIG. 16 shows a control unit according to the sixth embodiment. The output unit 10 compares the word form of the target word stored in the target word buffer 2 with the synonymous word form, and does not output if they are the same.
[0044]
As described above, in the fifth and sixth embodiments, the document group is ranked and searched for the target word and the supplementary word group, and the related word group of the target word is obtained from the higher order document group. Then, the document group is searched again for ranking, and a synonym group is obtained from the upper document group and the related word group.
[0045]
The program for executing the processing described in each flowchart by the control units of FIGS. 1, 5, 8, 12, 13, and 16 and the CPU of FIG. 7 constitutes a program according to the present invention. In addition, a storage medium such as the storage device 40 in FIG. 7 that stores the program constitutes a storage medium according to the present invention. The storage medium may be an optical disk, a magneto-optical disk, a magnetic recording medium, a semiconductor storage device, or the like.
[0046]
【The invention's effect】
As described above, according to the present invention, it is sufficient to use only a document group for synonym search, so that a general synonym can be obtained at lower cost than in the past.
[0047]
Also, by outputting a message to that effect for a target word for which a synonym cannot be obtained due to a shortage of documents, the quality of the synonym to be obtained can be improved.
In addition, by outputting the relevance (including co-occurrence) for each synonym candidate as information for further narrowing down synonym candidates, the quality of the synonym acquired can be improved.
Furthermore, it is possible to acquire low-cost, general synonyms for polysynonyms, and to improve the quality of the synonyms to be acquired.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of a control unit in a synonym search device according to a first embodiment of the present invention.
FIG. 2 is a flowchart illustrating processing of the entire control unit according to the first embodiment.
FIG. 3 is a flowchart illustrating processing of a primary search unit.
FIG. 4 is a flowchart illustrating processing of a secondary search unit.
FIG. 5 is a block diagram illustrating a configuration of a control unit in a synonym search device according to a second embodiment of the present invention.
FIG. 6 is a flowchart illustrating processing of the entire control unit according to the second embodiment.
FIG. 7 is a block diagram showing a synonym search device according to an embodiment of the present invention.
FIG. 8 is a block diagram illustrating a configuration of a control unit in a synonym search device according to a third embodiment of the present invention.
FIG. 9 is a flowchart illustrating processing of the entire control unit according to the third embodiment.
FIG. 10 is a flowchart illustrating processing of a primary search unit.
FIG. 11 is a flowchart illustrating processing of a secondary search unit.
FIG. 12 is a block diagram illustrating a configuration of a control unit in a synonym search device according to a fourth embodiment of the present invention.
FIG. 13 is a block diagram illustrating a configuration of a control unit in a synonym search device according to a fifth embodiment of the present invention.
FIG. 14 is a flowchart illustrating processing of the entire control unit according to a fifth embodiment.
FIG. 15 is a flowchart illustrating processing of a primary search unit.
FIG. 16 is a block diagram illustrating a configuration of a control unit in a synonym search device according to a sixth embodiment of the present invention.
[Explanation of symbols]
Reference Signs List 1 Input unit 2 Target word buffer 3 Document storage unit 4 Document search unit 5 Primary search unit 6 Related word extraction unit 7 Related word buffer 8 Secondary search unit 9 Synonymous word selection unit 10 Output unit 11 Judgment unit 12 Synonymous word extraction unit 13 Supplement Word buffer 20 Input device 30 Output device 40 Storage device 50 CPU

Claims (32)

Input means for inputting a target word;
Document storage means for storing a group of documents;
The document storage unit is searched by using the input target word as a search word, a document score is given to the document in which the search word appears by a first operation, and the documents are ranked in descending order of the document score. Is extracted as a seed document, words constituting the seed document are extracted as related word candidates, and the seed document is searched using the extracted related word candidates as search words. Primary search means for calculating the degree of relevance between the target word and the search word by calculation, and extracting a predetermined number of search words from the relevant word candidates in the descending order of the degree of relevance as related words
Secondary search means for searching the document storage means using the extracted related word as a search word, and extracting a related word by the first and second operations for a document in which the search word appears;
A synonym selecting means for selecting a predetermined number of related words having higher relevance from the related words extracted by the secondary searching means as synonym candidates. 2. The synonym search apparatus according to claim 1, wherein the processing is stopped when the target word does not fall within a predetermined upper rank of the degree of relevancy obtained by the primary search means. 3. The synonym search device according to claim 2, further comprising an output unit that outputs a message indicating that the processing is stopped. The synonym search device according to any one of claims 1 to 3, wherein the related word candidate group does not include the target word. The first operation is:
Document score = {tf / (tf + dlen)} × weight
weight = Log (N / n + 1)
Here, tf: frequency of occurrence of a search word (target word) in the document, dlen: document length, N: total number of documents stored in the document storage means, and n: search word appears in the total number N of documents. Number of documents (document frequency)
The synonym search device according to any one of claims 1 to 4, wherein The second operation is
Relevance = weight × (r / R−n / N)
The synonym search device according to any one of claims 1 to 5, wherein R: the number of seed documents, and r: the number of documents in which the search words in the seed document appear. 7. The method according to claim 1, wherein when the input target word, that is, the search word is plural, the document score is obtained for each search word, and a final document score is obtained by adding them. The synonym search device according to claim 1. The synonym search device according to any one of claims 1 to 7, wherein the synonym selection unit outputs the degree of relevance together with the synonym candidate. A document group is searched using the input target word as a search word, a document score is given to the document in which the search word appears by the first operation, and the documents are ranked in descending order of the document score. It is extracted as a seed document, the words constituting the seed document are extracted as related word candidates, and the seed document is searched using the extracted related word candidates as search words. Determine the degree of relevance between the target word and the search term, and extract a predetermined number of search terms as related terms from the related word candidate in the descending order of the degree of relevance,
Searching the document storage means using the extracted related word as a search word, extracting a related word by the first and second operations for a document in which the search word appears,
A synonym search method, comprising selecting a predetermined number of related words having higher relevance from the extracted related words as synonym candidates. 10. The synonym search method according to claim 9, wherein the processing is stopped when the target word does not fall within the upper predetermined rank of the relevance. The synonym search method according to claim 10, wherein a message indicating that the processing is stopped is output. 12. The synonym search method according to claim 9, wherein the related word candidate group does not include the target word. The first operation is:
Document score = {tf / (tf + dlen)} × weight
weight = Log (N / n + 1)
Here, tf: frequency of occurrence of a search word (target word) in the document, dlen: document length, N: total number of documents stored in the document storage means, and n: search word appears in the total number N of documents. Number of documents (document frequency)
The synonym search method according to any one of claims 9 to 12, wherein: The second operation is
Relevance = weight × (r / R−n / N)
14. The synonym search method according to any one of claims 9 to 13, wherein R: the number of seed documents, and r: the number of documents in which the search words in the seed document appear. 15. The method according to claim 9, wherein when the input target word, that is, a plurality of search words, is obtained, the document score is obtained for each search word, and a final document score is obtained by adding them. A synonym search method according to any one of the preceding claims. The synonym search method according to any one of claims 9 to 15, wherein a relevance is output together with the selected synonym candidate. An input process for inputting a target word,
A ranking process of searching a document group using the input target word as a search word, assigning a document score to the document in which the search word appears by a first operation, and ranking the document in descending order of the document score;
An extraction process of extracting a predetermined number of documents in the ranking as a seed document, and extracting words forming the seed document as related word candidates;
The seed document is searched by using the extracted related word candidate as a search word, and a relevance between the target word and the search word is obtained by a second operation for the document in which the search word appears, and the relevance is calculated from the relevant word candidate. Extraction processing for extracting a predetermined number of search terms in the descending order of the number as related words,
An extraction process of searching the document group using the extracted related word as a search word, and extracting a related word by the first and second operations for a document in which the search word appears;
A synonym search program for causing a computer to execute a selection process of selecting, as a synonym candidate, a predetermined number of related words in descending order of relevance of the extracted related words. 18. The synonym search program according to claim 17, further comprising a suspending process for suspending the process when the target word does not fall within the upper predetermined place of the relevance. 19. The synonym search program according to claim 18, further comprising an output process for outputting a message indicating that the process is stopped. The synonym search program according to any one of claims 17 to 19, wherein the related word candidate group does not include the target word. The first operation is:
Document score = {tf / (tf + dlen)} × weight
weight = Log (N / n + 1)
Here, tf: frequency of occurrence of a search word (target word) in the document, dlen: document length, N: total number of documents stored in the document storage means, and n: search word appears in the total number N of documents. Number of documents (document frequency)
The synonym search program according to any one of claims 17 to 20, wherein: The second operation is
Relevance = weight × (r / R−n / N)
22. The synonym search program according to claim 17, wherein R: the number of seed documents, and r: the number of documents in which search words in the seed document appear. 23. The method according to claim 17, wherein when the input target word, that is, the search word is plural, the document score is obtained for each search word, and a final document score is obtained by adding them. A synonym search program according to any one of the preceding claims. The synonym search program according to any one of claims 17 to 23, wherein the synonym selection unit outputs a relevance together with the synonym candidate. A storage medium storing the synonym search program according to any one of claims 17 to 24. The input means inputs, when the target word is a polysemous word, a supplementary word for giving a meaning together with the target word, and the primary search means sets the input target word and the supplementary word as the search word. The synonym search device according to any one of claims 1 to 8, wherein The synonym selecting means, at each occurrence position of the word s in the seed document obtained by the second operation, a co-occurrence degree c or c indicating a total number of related words appearing within a predetermined number W words before and after the word s. 28. The method according to claim 1, wherein a maximum number of related words are selected as synonym candidates in ascending order of c or c ′ = c / tf. Synonym search device of description. 17. The method according to claim 9, wherein when the target word is an ambiguous word, a supplementary word for meaning is input together with the target word, and the input target word and the supplementary word are used as the search word. A synonym search method according to any one of the preceding claims. At each occurrence position of the word s in the seed document obtained by the first and second operations, the co-occurrence degree c or c ′ = c indicating the total number of related words appearing within a predetermined number of W words before and after the word s. The synonym according to any one of claims 9 to 16 or 28, wherein / tf is determined, and a predetermined number of related words having a higher order are selected as synonym candidates in descending order of c or c '= c / tf. retrieval method. In the input process, when the target word is a polysemy, a supplementary word for its meaning is input together with the target word, and the ranking process sets the input target word and the supplementary word as the search word. The synonym search program according to any one of claims 17 to 24, characterized in that: The selection process includes a co-occurrence degree c indicating a total number of related words appearing at positions within a predetermined number of W words before and after each occurrence position of the word s in the seed document obtained by the extraction process for extracting the related word. Or c ′ = c / tf is determined, and a predetermined number of related words having a higher rank in the descending order of c or c ′ = ｃc / tf are selected as synonym candidates. Synonym search program described in section. A storage medium storing the synonym search program according to any one of claims 17 to 24, 30, and 31.

Images