JPH10149370A - Document retrieval method and device using context information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the document retrieval accuracy by selecting a word having more intensive relationship as a keyword in a semantic sense when the relative words are added as keywords. SOLUTION: A keyword expansion part 4 extracts the relative words of an input keyword from a relative word data base, and a relative keyword selection part 5 outputs all relative keyword candidates expanded to the keyword in sequence and in a processing symmetric form. The part 5 also checks whether the relative keyword candidates have the same context information as the input keyword based on a word appearance frequency data base 3 including the context information. Thus, the part 5 selects the relative keywords having high appearance frequency. Then a document retrieval part 7 retrieves a document from a document data base 6 and outputs it based on the keyword that is inputted to a keyword input part 1 and the keyword that is selected at the part 5.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a document search method and apparatus, and more particularly, to a method and apparatus for searching a document by expanding an input keyword using word co-occurrence information.

[0002]

2. Description of the Related Art Conventionally, this type of document search method is used for setting a search character string from an input document or a keyword in a document search device or the like, and searching for a document including the search character string. I have.

As an example of a conventional document search method, for example, Japanese Patent Laid-Open Publication No. Hei 3-172966 discloses a similar document search apparatus which can efficiently and accurately search for a similar document from a full-text database. Proposed.

This conventional similar document search apparatus is provided with a syntax input unit for directly inputting a document and using it as code information, separating the input character string and adding morpheme information thereto, and providing a document based on the morpheme information. A dependency analyzing unit for determining a dependency structure between (phrases), an index extracting unit for determining a sentence structure from the result of the dependency analysis, extracting an index from the sentence structure, and assigning an importance of the index; A document storage unit for storing the input document, the dependency analysis result, and the index extraction result; a thesaurus expansion unit for expanding the index of the index extraction unit in a thesaurus dictionary; and a similarity index between the input document and the stored document. And a similar document output unit for outputting the searched similar document.

[0005] The thesaurus expansion unit is provided with a polysemy determination table, and distinguishes words having the same notation but different meanings according to the field of the document. In other words, the polysemy table is composed of information of word notation, reading, and field of use. When thesaurus expansion is performed, synonyms and synonyms that are most semantically correct in the field of the input document are output. For example, in the case of “CD”, synonyms and synonyms corresponding to the field are output, such as “cash dispenser” in the banking field and “compact disc” in the music field.

[0006]

However, in the above-described conventional document retrieval apparatus, the priorities of the expanded related vocabulary cannot be determined in thesaurus expansion using the polysemy determination table, so that an appropriate related vocabulary is not necessarily determined. Cannot be selected, and the search accuracy is not improved.

The reason is that in the polysemy determination table, the vocabulary of the same field is selected by specifying the field information of the document, but if the field information is not described in the document,
This is because it cannot be determined which vocabulary is a related vocabulary.

[0008] Therefore, the present invention has been made in view of the above problems, and has as its object to select a more semantically related word when adding a related word as a keyword. Another object of the present invention is to provide a document search method and apparatus that improves search accuracy by performing a search by adding a keyword.

[0009]

In order to achieve the above object, a first document search method according to the present invention (Claim 1) provides a synonym dictionary and a related word dictionary when a search keyword is extended from an input keyword. Using a dictionary, a thesaurus, etc. to develop related words, and using a word co-occurrence database that holds the co-occurrence relation of words, context information, and frequency, related words with high co-occurrence frequency with the same context information as the input keyword Is selected as a related keyword, and is added to the search keyword to perform a document search.

[0010] A second document search method according to the present invention (Claim 2)
In the first document search method, when a search keyword is expanded, only a characteristic input keyword whose appearance frequency is smaller than a predetermined value is expanded to a related word, and is added to the search keyword. And search.

A third document search method according to the present invention (claim 3)
Expands related words using synonym dictionaries, related word dictionaries, thesaurus dictionaries, etc. when expanding search keywords from input keywords, and word co-occurrence that retains word co-occurrence relations, context information and frequency Using the database, select a related word with high co-occurrence frequency with the same context information as the input keyword as a related keyword, and then recursively expand related words and select related keywords from the related keywords and add them to the search keywords. Document retrieval is performed.

A fourth document search method according to the present invention (claim 4)
In the third document search method, when a search keyword is expanded, recursive related word development and selection of a related keyword are performed only for characteristic related keywords whose appearance frequency is smaller than a predetermined value. And performing a search by adding the search keyword.

A first document retrieval apparatus according to the present invention (Claim 5)
Is synonymous with a keyword input section for inputting a keyword, and a synonym dictionary, a related word dictionary, a thesaurus dictionary, etc., based on input keywords input from the keyword input section.
A keyword expansion unit that expands related words such as synonyms, and a word co-occurrence that has the same context information as the input keyword by referring to the word co-occurrence database that holds the word co-occurrence relationship, context information, and frequency from the expanded related words A related keyword selecting unit for selecting a related word having a high frequency of occurrence as a related keyword, and a document search unit for searching a document from a document database using the input keyword and the related keyword as search keywords.

A second document retrieval apparatus according to the present invention (claim 6)
Is characterized in that a keyword expansion section of the first document search apparatus is replaced by a selective keyword expansion section, and only a characteristic input keyword having a low appearance frequency is searched in addition to a search keyword for expanding a related word. And

A third document retrieval apparatus according to the present invention (claim 7)
Is a keyword input unit for inputting a keyword, and a keyword expansion unit for expanding synonymous / synonymous words and related words using a synonym dictionary, a related word dictionary, a thesaurus dictionary, and the like from the input keyword and the related keyword input from the keyword input unit. From the expanded related words, refer to the word co-occurrence database that holds the co-occurrence relation of the words, context information and frequency, and select the related words with high co-occurrence frequency with the same context information as the input keywords as related keywords A recursive related keyword selection unit that selects and outputs the selected keyword to a keyword expansion unit, and a document search unit that searches a document from a document database using the input keyword and the related keyword as search keywords.

A fourth document search device of the present invention (claim 8)
Performs a recursive related word development only on characteristic related keywords whose appearance frequency is smaller than a predetermined value in the keyword developing unit of the third document search apparatus, and adds the keyword to the search keyword. And search.

[0017]

Next, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the document search device according to the first embodiment of the present invention. Referring to FIG. 1, a document search apparatus according to a first embodiment of the present invention includes a keyword input unit 1, a word appearance frequency database with context information 3, a related word database 2,
It is configured to include a keyword expanding section 4, a related keyword selecting section 5, a document database 6, and a document searching section 7.

The keyword input unit 1 inputs one or more keywords as a search character string using a keyboard or the like.

The word appearance frequency database 3 with context information is used to determine what kind of words appear in the document database 6 or other document databases and how frequently the words appear in the context information on the left and right sides of the word. (Word co-occurrence database holding co-occurrence relations of words, context information and frequency).

An example of context information is that words present on the left and right of a certain word are composed of a character string in the case of a particle, and a part of speech in the case of a non-particle, such as "word (left context, right context), frequency" The appearance frequency is recorded in three stages of large, medium and small (see FIG. 3). Of course, the classification of “large”, “medium”, and “small” may be frequency numerical data.

FIG. 3 exemplifies the contents of personnel reduction in the word appearance frequency database 3 with context information. For example, "Reduction of staff (by name), large"
(The second line in FIG. 3) shows that the frequency of the word “reduced staff” appears together with the left context “name” and the right context “due” (that is, “by the staff reduction of...”). And "reduction of staff (noun), medium" (fifth line in FIG. 3) is the frequency of occurrence of "reduction of staff (noun)" (for example, "reduction plan of ..." Indicates “medium”.

The related word database 2 is a thesaurus dictionary, a synonym dictionary, and a related word dictionary.

The keyword developing section 4 outputs related words as related keyword candidates to the related keyword selecting section 5 from the single or plural input keywords provided from the keyword input section 1 using the related word database 2.

The related keyword selecting unit 5 uses the word frequency database 3 with context information to select a high-frequency word having the same context information as the input keyword from the related keyword candidates as a related keyword.

FIG. 2 is a flowchart for explaining a processing flow of the related keyword selecting section 5 in the embodiment of the present invention.

Referring to FIG. 2, the related keyword selecting section 5 includes a step Al for outputting related keyword candidates,
Step A2 for determining whether the related keyword candidate appears in the same context as the input keyword (referred to as a “context co-occurrence relationship determining step”). If the frequency of occurrence is high, a related keyword for selecting the candidate as a related keyword It comprises a selection step A3 and a step A4 for determining whether or not there are other related keyword candidates.

The document database 6 is a database that stores digitized documents.

The document search unit 7 searches and outputs a document from the document database 6 using the keyword input to the keyword input unit 1 and the related keyword selected by the related keyword selection unit 5.

Next, the operation of the thus configured document retrieval apparatus according to the first embodiment will be described with reference to FIGS.
This will be described with reference to FIGS. FIG. 4 exemplifies the contents of the related word database 2 and the contents of extracted context information having a high appearance frequency from the word appearance frequency database with context information 3 for each related word.

When an input keyword is input from the keyword input unit 1, it is supplied to the keyword development unit 4.

The keyword developing section 4 extracts a related word of the input keyword from the related word database.

The related keyword selecting section 5 sequentially outputs all the related keyword candidates that have been developed as keywords as processing targets (step Al in FIG. 2).

Next, the related keyword selecting section 5 uses the word appearance frequency database 3 with context information,
It is checked whether the related keyword candidate is a word having the same context information as the input keyword (step A2 in FIG. 2). If the frequency of appearance is high as a result of the determination in step A2, the keyword is selected as a related keyword (step A2 in FIG. 2).
3). Subsequently, the related keyword developing unit 4 determines whether or not there is another related keyword candidate (step A4 in FIG. 2). If any other related keyword candidate remains, the control returns to step A1 until there is no more related keyword candidate. And the above steps Al to A4 are repeated.

For example, referring to FIG. 4, the keyword developing unit 4 determines, as a related keyword candidate of “manpower reduction”,
Outputs "Large airline", "Management", "Rationalization", "Personnel arrangement", "Employees", "Voluntary retirement", "Retirement allowance", and "Restructuring".

Referring to FIG. 3, the related keyword selecting unit 5 determines from the word appearance frequency database 3 with context information as “higher frequency context of“ reduction of personnel ”, (By name) and "Reduction of staff (to name)".

The reason why these frequencies are large is, for example, "...
... "," ... for reducing the number of people ... ",
The phrase "... by reducing the number of people ..." indicates that the expression is commonly used.

The related keyword selection unit 5 outputs the candidates (the above-mentioned “major airline”,
Management, rationalization, staffing, etc.)
A context having a high frequency is output from the word appearance frequency database 3 with context information.

Subsequently, the related keyword selecting section 5 selects, from among the related keyword candidates, the frequency of the same context as “reduction of staff”, “organization of staff”, “voluntary retirement”, “restructuring”.
Is selected as a related keyword.

The document search section 7 includes an input keyword input from the keyword input section 1 and a related keyword selection section 5.
A document is retrieved from the document database 6 by using the related keyword selected in the above step and output.

As described above, in the first embodiment of the present invention, the related word is expanded from the input keyword by using the related word database 2, and the related word is used by using the word appearance frequency database with context information 3 from the related word. A related word that appears in the same context as the input keyword can be selected and used for document search.

Also, in the first embodiment of the present invention, even when there are a plurality of related words, words with low relevance can be eliminated by using the context information and the co-occurrence relationship. Therefore, even when the field information is not described in the document, the retrieval accuracy of the document is significantly improved.

FIG. 5 is a block diagram showing a configuration of a document search device according to the second embodiment of the present invention.

Referring to FIG. 5, in the document search device according to the second embodiment of the present invention, the keyword expansion unit 4 of the document search device according to the first embodiment shown in FIG. The difference is that it is replaced by a selective keyword expansion unit 4 ', and the other configuration is the same.

FIG. 6 is a flowchart for explaining a processing flow of the selective keyword expanding section 4 'according to the second embodiment of the present invention. Referring to FIG. 6, the selective keyword developing unit 4 'includes a step B1 for outputting an input keyword, a step B2 for determining whether the appearance frequency of the input keyword is low (referred to as a "development determining step"), And a step B4 for outputting to the related keyword selecting section 5 (referred to as a "keyword developing step"), and a step B4 for determining whether or not there is another input keyword.

That is, in the selective keyword developing section 4 ', the input keyword is output in step B1, and if the appearance frequency does not exceed the set threshold in step B2, the related word is developed in step B3. A process for outputting to the related keyword selection unit 5 is performed. Then, when the input keyword remains, similarly, step B1
Do from.

As described above, in the second embodiment of the present invention, the keyword expansion for an extremely general word having a high frequency of appearance can be suppressed by the selective keyword expansion section 4 ', and the number of keywords can be increased. It is possible to suppress a decrease in search accuracy due to too long.

FIG. 7 is a block diagram showing a configuration of a document search device according to the third embodiment of the present invention.

Referring to FIG. 7, in the document search device according to the third embodiment of the present invention, the related keyword selecting section 5 of the document search device according to the first embodiment shown in FIG. The difference is that it is replaced by the recursive related keyword selection unit 5 'and that the output of the recursive related keyword selection unit 5' is also output to the keyword expansion unit 4, and the other configurations are the same. ing.

The recursive related keyword selecting section 5 'selects a high-frequency word having the same context information as the input keyword from the related keyword candidates using the word frequency database 3 with context information. At the same time, the related keyword is output to the keyword developing unit 4.

The related keywords sent to the keyword developing unit 4 are extracted from the related word database 2 in the same manner as the input keywords, and the related keywords of the related keywords are selected by the recursive related keyword selecting unit 5 '. Is done.

This process is repeated until a predetermined number of keywords are obtained.

For example, referring to FIG. 4, the keyword developing unit 4 determines, as a related keyword candidate of “manpower reduction”,
Outputs "Large airline", "Management", "Rationalization", "Personnel arrangement", "Employees", "Voluntary retirement", "Retirement allowance", and "Restructuring".

Referring to FIG. 3, the recursive related keyword selecting section 5 ′ uses the word frequency database 3 with context information as a high-frequency context
“Reduction of staff (for one of the people)”, “Reduction of staff (for one of the people)”, and “Reduction of staff (for one of the people)” are extracted.

The recursive related keyword selection unit 5 ′ selects the candidates (the above-mentioned “large airline”, “management”, “rationalization”, “personnel arrangement”, etc.) output by the keyword development unit 4. A context having a high frequency is output from the word appearance frequency database 3 with context information.

Subsequently, a recursive related keyword selecting section 5 '
Selects “restructuring”, “arrangement of staff”, and “voluntary retirement”, which have a high frequency of the same context as “reduced staff”, from the related keyword candidates as related keywords, and output the keyword to the keyword developing unit 4.

FIG. 9 exemplifies the contents of the related word database 2 and the contents of extracted context information having a high appearance frequency from the word appearance frequency database 3 with context information for each related word.

Referring to FIG. 9, the keyword developing unit 4
Outputs “layoff”, “unification and abolition”, and “reconstruction” as related keyword candidates for “restructuring”.

Referring to FIG. 3, the recursive related keyword selecting section 5 ′, based on the word frequency database 3 with context information, sets
Extract "layoff (by name)".

Referring to FIG. 9 again, the recursive related keyword selecting section 5 ′ outputs a high frequency context from the word appearance frequency database with context information 3 to the candidates output by the keyword expanding section 4.

Subsequently, a recursive related keyword selecting section 5 '
Selects “layoff”, which has the same context frequency as “restructuring”, from the related keyword candidates as the related keyword.

As described above, in the third embodiment of the present invention, the recursive related keyword selecting section 5 '
Even when the number of input keywords is extremely small, a sufficient number of search keywords can be obtained, and a decrease in search accuracy due to too few keywords can be suppressed.

As a modification of the third embodiment of the present invention, the recursive related keyword selecting section 5 ′ exceeds the set appearance frequency threshold in order to prevent the number of related keywords from diverging. Only words that are missing are processed.
You may do so.

FIG. 8 is a flowchart for explaining a processing flow of the recursive related keyword selecting section 5 'in the third embodiment of the present invention and its modification. FIG.
, The recursive related keyword selecting unit 5 ′ outputs the related keyword and the step C2 of determining whether the appearance frequency of the related keyword is low or not.
(Referred to as an "expansion determination step"), a step C for expanding the related word and outputting it to the recursive related keyword selection unit 5 '
3 (referred to as “keyword development output step”) and step C4 of determining whether there are any other related keywords
And consisting of

In step Cl, the related keyword is output, and if the frequency of appearance does not exceed the set threshold in step C2, the related word is expanded and output to the recursive related keyword selecting section 5 '(step C).
3). Then, when the related keyword remains, the process is similarly performed from step Cl.

As described above, in the third embodiment and its modification of the present invention, it is possible to obtain a sufficient search keyword even when the number of input keywords is extremely small, and it is possible to obtain an extremely general keyword having a high appearance frequency from related keywords. Key words can be suppressed, and a decrease in search accuracy due to too few keywords and too many keywords can be suppressed.

[0067]

As described above, according to the present invention, the following effects can be obtained.

A first effect of the present invention is that the retrieval accuracy can be improved because a keyword having a strong relationship with the input keyword is selected. As a result, the present invention improves the document retrieval accuracy.

The reason is that, in the present invention, a word having a high appearance frequency having the same context information as the input keyword is selected as the related keyword, and is added to the search keyword to perform the search.

Further, as a second effect of the present invention, the selective keyword expansion unit can suppress the keyword expansion of an extremely general word having a high appearance frequency, and the search accuracy is reduced due to an excessive number of keywords. Can be suppressed.

Further, as a third effect of the present invention, even when the number of input keywords is extremely small, a sufficient number of search keywords can be obtained by the recursive related keyword selection section. That is, the reduction can be suppressed.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a document search device according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating a related keyword selection process according to the first embodiment of the present invention.

FIG. 3 is a diagram illustrating a specific example of the contents of a word appearance frequency database with context information according to the first embodiment of the present invention;

FIG. 4 is a diagram for explaining the first embodiment of the present invention, which illustrates contents of a related word database;
This is an example of the content of context information having a high appearance frequency extracted from a word appearance frequency database with context information for each related word.

FIG. 5 is a block diagram illustrating a configuration of a document search device according to a second embodiment of the present invention.

FIG. 6 is a flowchart illustrating processing of a selective keyword expanding unit according to the second embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration of a document search device according to a third embodiment of the present invention.

FIG. 8 is a flowchart illustrating a process of a recursive related keyword selection unit according to the third embodiment and the modification of the present invention.

FIG. 9 is a diagram for explaining a third embodiment of the present invention, which illustrates the contents of a related word database;
This is an example of the content of context information having a high appearance frequency extracted from a word appearance frequency database with context information for each related word.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Keyword input part 2 Related word database 3 Word frequency database with context information 4 Keyword development part 4 ′ Selective keyword development part 5 Related keyword selection part 5 ′ Recursive relevant keyword selection part 6 Document database 7 Document search part

Claims (9)

[Claims] When a search keyword is extended from an input keyword, related words are expanded using a synonym dictionary, a related word dictionary, a thesaurus dictionary, and the like, and the co-occurrence relationship of the words, context information, and frequency are retained. Using a word co-occurrence database, selecting a related word having a high co-occurrence frequency having the same context information as the input keyword as a related keyword, adding the related word to a search keyword, and searching for a document. Document search method. 2. The method according to claim 1, wherein when the search keyword is extended, the related word is expanded only for a characteristic input keyword whose appearance frequency is smaller than a predetermined value.
2. The document search method according to claim 1, wherein a document is searched in addition to the search keyword. 3. When a search keyword is extended from an input keyword, related words are expanded by using a thesaurus, a related word dictionary, a thesaurus dictionary, etc., and the co-occurrence relation of the words, context information, and frequency are retained. Using a word co-occurrence database, select a related word having a high co-occurrence frequency having the same context information as the input keyword as a related keyword, further recursively develop related words from the related keyword, A document search method characterized by making a selection and adding a search keyword to search for a document. 4. When the search keyword is extended, recursive development of related words and selection of related keywords are performed only for characteristic related keywords whose appearance frequency is smaller than a predetermined value. 4. The document search method according to claim 3, wherein a document is searched in addition to the keyword. 5. A keyword input section for inputting a keyword, and a keyword expansion for expanding related words such as synonyms and synonyms from an input keyword input from the keyword input section using a thesaurus, a related word dictionary, a thesaurus dictionary, or the like. And, from the expanded related words, refer to a word co-occurrence database holding the co-occurrence relation of the words, context information and frequency, and associate the related words having the same context information as the input keyword with a high co-occurrence frequency. A document search device comprising: a related keyword selection unit that selects a keyword; and a document search unit that searches for a document from a document database using the input keyword and the related keyword as search keywords. 6. A keyword input unit for inputting a keyword, and when a related word such as a synonym or a synonym is developed from an input keyword input from the keyword input unit using a thesaurus, a related word dictionary, a thesaurus dictionary, or the like. A selective keyword expansion unit that expands related words only for characteristic input keywords with a low frequency of appearance, and a word co-occurrence database that holds word co-occurrence relationships, context information, and frequency from expanded related words A related keyword selecting unit for selecting, as a related keyword, a related word having a high co-occurrence frequency having the same context information as the input keyword; and searching for a document from a document database using the input keyword and the related keyword as search keywords. A document search unit for performing the search. 7. A keyword input unit for inputting a keyword, and synonymous / synonymous words and related words are developed from the input keyword and the related keyword input from the keyword input unit using a thesaurus, a related word dictionary, a thesaurus dictionary, and the like. A keyword expansion unit, from the expanded related words, referring to a word co-occurrence database that holds the co-occurrence relation of the words, context information and frequency, and related words having a high co-occurrence frequency having the same context information as the input keyword. A recursive related keyword selecting unit that selects a related keyword as a related keyword and outputs the related keyword to the keyword expanding unit; and a document searching unit that searches for a document from a document database using the input keyword and the related keyword as search keywords. A document search device characterized by the following. 8. The document search according to claim 7, wherein keyword expansion is performed only when the related keyword is a characteristic related keyword whose appearance frequency is smaller than a predetermined value. apparatus. 9. The selective keyword expanding section compares an appearance frequency of the input keyword with a predetermined threshold value.
6. The document search apparatus according to claim 5, wherein the related word is developed only for the input keyword whose appearance frequency does not exceed the threshold, and is output to the related keyword selecting unit.