JP2001325293A - Method and device for retrieving whole sentences and storage medium with stored whole-sentence retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a whole-sentence retrieving method, its device and a storage medium for storing a whole-sentence retrieval program, by which structure information is adopted in calculation of a conformity degree and a filter by attribute is realized by a simple operation such as description in a format file and designation by a retrieval formula when the conformity degrees of the retrieval formula with respective documents to be retrieved are properly calculated, the documents are arranged in order from the one with the lowest conformity degree and whole sentences are retrieved with them as a retrieval result. SOLUTION: An index corresponds to each part of each document structure. Which one of the indexes is generated is described in the format file or a structurized document. A plurality of index files are generated based on the description of the format file, which of one or plural indexes among the indexes of the generated index file are used is designated by the retrieval formula and document structure information is used for calculating the conformity degrees.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a full-text search method and apparatus and a storage medium storing a full-text search program, and in particular, each document (structured document) having a structure to be searched is converted into a search formula. The present invention relates to a full-text search method and apparatus capable of appropriately calculating a value of “fitness” indicating the degree of matching, and a storage medium storing a full-text search program.

[0002]

2. Description of the Related Art The configuration of a conventional full-text search apparatus will be described.

FIG. 7 shows a configuration of a conventional full-text search device. The full-text search device shown in FIG.
0 and a search execution unit 20. Each function is as follows.

[0004] The document index section 10 creates an index by the following method.

A set N of documents (plain text) to be searched is input from a search target document database 11.

A word i used in each document j∈N is extracted, and the importance w ij of each word i is calculated.

A reverse index 12 is created and output. Here, the reverse index is a table having the following keys and values.

Key: word i value: ID of document j in which word i appears, and document j of word i
Here, an example of the reverse index 12 is shown in FIG. For example, in FIG. 8, the word “language” corresponds to documents 1, 3, 6
And the importance of 'language' in each document is 0.3, 0.3, 0.8.

The search execution section 20 performs a search by the following method.

A search expression q described by a word or a Boolean operator combination of words is input.

From the reverse index 12 created by the document index unit 10, a document set in which each word i included in the search formula q appears and the importance w ij of each word i in each document j are acquired.

[0012] Each document j and the search formula q are calculated using the Boolean operator and the word importance wij included in the search formula q.
Is calculated.

[0013] Documents are arranged in descending order of relevance to obtain search results.

For example, the document index unit 10 creates the reverse index 12 of FIG. 8 from the document to be searched, and shows a method of calculating the fitness when the following search formula is given.

(Language and processing) or knowledge Here, the processing method of and and or differs depending on the processing system. Here, it is determined as follows.

Or—The sum of the left and right evaluation values is taken as the evaluation value.

And-the smaller of the left and right evaluation values is taken as the evaluation value.

Here, in the case of a word, the degree of importance is an evaluation value, and the evaluation value obtained from the entire search formula is the degree of relevance.
If the importance wiij is not registered in the reverse index, the importance of the word i in the document j is “0”.

The relevance of document 1 is “0.4” as follows:
Is obtained.

(Language '0.3' and processing '0.
2 ") or knowledge '0.2' → ('0.2') or '0.
2 '→ 0.4 The relevance of document 3 is obtained as'0' as follows: (language '0.3' and processing '0') or knowledge '0' →
('0') or '0' → 0 The documents are arranged in descending order of the degree of matching calculated in this way, and the top k (k is a constant) items are set as search results.

[0021]

However, when a structured document is to be searched by the above-mentioned conventional full-text search apparatus, there are the following problems.

(1) Compatibility calculation that reflects the structure information of a document cannot be performed.

In recent years, structures such as XML and the like are described using tags in documents. FIG. 9 shows an example of a document having a structure. In the figure, the tag (<TEXT>
− </ TEXT>)).

In the conventional full-text search device, the input is assumed to be plain text having no structure.
Even if a document has structural information, such information cannot be appropriately incorporated into the relevance calculation. For example, taking into account the structure of titles, "Those containing 'artificial intelligence' somewhere in the document,
I want to increase the relevance of documents that contain. "Cannot be answered.

(2) A filter using an attribute attached to a document cannot be realized.

The filter refers to the following search request. I want to extract only Japanese documents that contain 'UNIX (registered trademark)'. If the document contains 'knowledge' and the author is 'Suzuki', we want to improve the relevance (ranking up the ranking).

In each case, the filter uses an attribute of "language to be used", and in the case of a filter, an attribute of "book". In particular, here, "and-type filters" are used to narrow down results, and "or-type filters" are used to correct the degree of conformity of results and raise the ranking of specific items. Type filter ".

It is difficult to realize these filters by using a reverse index used in a conventional full-text search device. Therefore, if a filter is required, it is necessary to prepare a completely different index or leave the full text of the document independently of the full text search device and scan these attributes at the time of search.

Further, in the conventional technique, there is also a method of directly describing a partial structure in a retrieval formula for a structured document. However, this method is not used for a full-text search device, but is used for the purpose of extracting a part that satisfies a condition described in a search formula from one target structured document. For example, for the structured document in Fig. 9, search expression: // DOC / AUTROR (<AUT nested in <DOC>
If the search is performed using (the part enclosed by RO> is extracted), "Suzuki Tanaka" is obtained as a result. As described above, this method cannot be used for the purpose of calculating the relevance between the retrieval formula and each structured document.

The present invention has been made in view of the above points, and appropriately calculates the relevance between a search formula and each document to be searched, arranges the documents in descending order of relevance, and obtains a full-text as a search result. When performing a search, a full-text search method and apparatus and a full-text that can incorporate structural information into the calculation of relevance and realize a filter by attributes by a simple operation of description in a format file and specification in a search formula It is an object to provide a storage medium storing a search program.

[0031]

FIG. 1 is a diagram for explaining the principle of the present invention.

According to the present invention (claim 1), to what degree each document (structured document) having a structure to be searched matches a word or a search expression formed by combining Boolean operators of words. In a full-text search method that calculates the “relevance” that indicates whether a document is in order and sorts documents in descending order of relevance, the search result is used to determine which part of the structure of each document is to be indexed. Or in an independent format file (step 1)
A plurality of index files are created based on the description of the format file (step 2), and one or more of the indexes of the created index files are
Which of a plurality of indexes is to be used is specified by a search formula (step 3), and the structural information of the document is used for calculating the degree of matching.

According to the present invention (claim 2), when a partial structure of a document to be indexed is described in a structured document or in an independent format file, the structure of the target document is nested. By analyzing the document, creating a tree structure with elements representing each structure as nodes and elements representing the entire document as root nodes, and describing the path from the root node in the format file, Specify which part of the document structure is to be indexed.

According to the present invention (claim 3), when an index file is created, a word is used as a key, and a reverse index having a set of a pair of a document ID in which the word appears and the importance of the word as a value is used. One or both of a forward index and a document ID are used as a key, and a value of a set of pairs of words appearing in the document and importance of the word is created based on the description of the format. An index ID is assigned to each.

According to the present invention (claim 4), a search formula is specified by using an index ID assigned to each index to determine which index to use from among a plurality of created indexes.

According to the present invention (claim 5), the relevance between a search expression described by a word, a Boolean operator combination of words, and a filter expression using an index file, and each document to be searched. Is calculated using a reverse index, and a search using a forward index is performed.

According to the present invention (claim 6), when performing a search using a forward index,
An 'and-type filter' that retrieves only documents that match the conditions specified by the search expression from among a specified number of documents with high relevance and that matches the conditions specified by the search expression, and matches the conditions specified by the search expression An "or-type filter" is used in which the relevance of the document is increased, the documents are rearranged in descending order of the corrected relevance, and the retrieval result is obtained.

FIG. 2 is a diagram showing the principle of the present invention.

According to the present invention (claim 7), to what extent each document (structured document) having a structure to be searched matches a word or a search expression formed by combining words with Boolean operators. This is a full-text search device that calculates the “fitness” that indicates whether a document is arranged and sorts the documents in descending order of the matchability and obtains a search result. The structured document determines which index of the structure of each document is to be created. A format file creating means 300 described in a separate or independent format file, and a format file 120
Document index creating means 100 for creating a plurality of index files 150 based on the description of
A search execution unit that specifies which one or a plurality of indexes to use by a search formula, and uses the document structure information for calculating the degree of relevance.

According to the present invention (claim 8), when the partial structure of the document to be indexed is described in the structured file or in the independent format file 120 by the format file creating means 300, it is an object. When the structure of the document is nested, means for analyzing the document and creating a tree structure having elements representing the respective structures as nodes and elements representing the entire document as root nodes, Means for specifying which part of the document structure is to be indexed by describing the path from the root node.

According to the present invention (claim 9), in the document index creating means 100, a reverse index in which a word is used as a key, and a set of a pair of a document ID in which the word appears and the importance of the word is used as a value, Based on the description of the format file 120, one or both of a forward index and a document index are used as a key, and a value of a set of pairs of words appearing in the document and importance of the words is created. Means for assigning an index ID to each of the indexes is included.

According to the present invention (claim 10), the search execution means 2
00, means for designating a search formula by using an index ID assigned to each of the indexes to be used among the plurality of created indexes.

According to the present invention (claim 11), the search execution means 2
At 00, using the index file 150,
A relevance calculation means for calculating relevance between a word, a search expression described by a Boolean operator combination of words and a filter expression, and each document to be searched; It includes a normal search means for performing a search using a pull-down index and a filter search means for performing a search using a forward-looking index.

According to a twelfth aspect of the present invention, the filter search means matches a condition specified by a search formula from a predetermined number of documents having a high degree of relevance among acquired search results. An 'and-type filter' that retrieves only documents and uses them as search results, and increases the relevance of documents that match the conditions specified by the search expression, and sorts the documents again in descending order of the corrected relevance to obtain search results 'o
and means using an 'r-type filter'.

According to the present invention (claim 13), to what extent each document (structured document) having a structure to be searched matches a word or a search expression formed by combining Boolean operators of words. This is a storage medium storing a full-text search program which calculates the "relevance" indicating the degree of relevance, arranges the documents in descending order of the relevance, and obtains a search result, and creates an index corresponding to which part of the structure of each document Format file creation process that describes whether or not a document is in a structured document or in an independent format file, a document index creation process that creates multiple index files based on the description of the format file, and an index of the created index file Specify which index or multiple indexes to use in the search expression , And a search execution process using the structural information of the document in the calculation of fitness.

According to the present invention (claim 14), when the partial structure of a document to be indexed is described in a structured document or in a separate format file in the format file creation process, If the structure is nested, analyze the document and create a tree structure with elements representing each structure as nodes and elements representing the entire document as root nodes;
Specifying the portion of the document structure to be indexed by describing the path from the root node in the format file.

According to the present invention (claim 15), in the document index creation process, a reverse index in which a word is used as a key and a set of a pair of a document ID in which the word appears and the importance of the word is used as a value, Using the ID as a key, create one or both of the forward index, which has a value of a set of pairs of words appearing in the document and the importance of the word, based on the description of the format file. And assigning an index ID to the.

According to the present invention (claim 16), in the search execution process, a process of designating a search expression by using an index ID assigned to each index to determine which index is used among a plurality of created indexes. Including.

According to the present invention (claim 17), in the search execution process, using an index file, a search expression described by a word, a Boolean operator combination of words and a filter expression, and each document to be searched are provided. A fitness calculation process of calculating a fitness with a normal search process of performing a search using a reverse index, and a filter search process of performing a search using a forward index. Including.

According to the present invention (claim 18), in a filter search process, a condition specified by a search formula is selected from among a predetermined number of documents having a high degree of relevance among acquired search results. An 'and-type filter' that retrieves only documents and uses them as search results, and increases the relevance of documents that match the conditions specified by the search expression, and sorts the documents again in descending order of the corrected relevance to obtain search results Includes a process using an 'or-type filter'. As described above, the present invention provides a full-text search method that properly calculates the relevance between a search formula and each document to be searched, arranges the documents in descending order of relevance, and obtains a search result. By a simple operation of description and specification in a search formula, it is possible to incorporate structural information into the calculation of the degree of fitness. Further, it is possible to realize a filter based on attributes.

[0051]

FIG. 3 shows the configuration of a full-text search device according to the present invention. The full-text search device illustrated in FIG. 1 includes a document index unit 100 and a search execution unit 200. Data from the search target document database 110 and the format file 120 is input to the document index unit 100.
The index file group 150 is output, and the search execution unit 200 performs a search using the files of the index file group 150.

The index file group 150 includes a reverse index 151 and a forward index 152.

First, the operation of the document index unit 100 will be described.

A set N of structured documents to be searched is input from the search target document database 110. It is assumed that the structure shown in FIG. 9 is described in each document.

The structure of each document j∈N is analyzed, and a structure tree is created. FIG. 4 shows an example of the structural tree. FIG.
When the structure is described by nesting tags as shown in FIG. 4, when the structure is analyzed, a tree structure as shown in FIG. 4 is obtained.
Here, a tree representing such a structure is called a structural tree. A method of creating a structure tree from a document having a structure uses an existing technology.

Based on the specification of the index target described in the format file 120, which part of the structure tree is to be indexed is determined. The description format of the format file 120 is not particularly limited, but is specified here as shown in FIG.

The designation of the index target is shown in (<FIE
LD_DEFINITION>-</ FIELD_DEFINITION>). Tag path from the root of the structure tree (target_ta
g) specifies which part of the document structure is to be indexed. '//' in target_tag
Indicates a descendant, and '/' indicates a direct child.

For example, 1 of <FIELD_DEFINITION> in FIG.
Terget_tag = “// SECTION / SPEC” on the line As a descendant from the document root (DOC), 'SECTI
ON ', and indicates that a substructure that is a direct child' SPEC 'of the' ON 'is to be indexed. That is, a subtree below 'SPEC' in FIG. 4 is specified, and <SECTION> to </ SECTION> in the original document.
Indicates that the part enclosed by <SPEC> to </ SPEC> nested between is indexed. Also, index_id
= "Spec" specifies which index file to register a word in, as described below.

The index file group 150 is created based on the specification of the index method described in the format file 120. Each index file group 150 is one of the following.

Reverse index (reverse index file 151): This table has the following keys and values, and has the same format as the conventional reverse index. Key: word i Value: set of ID of document j in which word i appears and weight wij in document j of word i • Forward index (forward index file 1)
52): A table having the following keys and values. -Key: ID of document j-Value: Set of word i appearing in document j and importance wij of word i in document j Here, an example of the forward index is shown in FIG. In the figure, the document 3 includes the words "language" and "learning", and their importance is 0.3 and 0.6, respectively.

The specification of the index method is performed in the format file <INDEX_FILE_DEFINITION> to
</ INDEX_FILE_DEFINITION>.

For example, in FIG. 5, <INDEX_FILE id =
The description “spec” type = “INVERTED” />is' spec ”
<INDEX_FILE id = “lang'type”
The description “=“ SEQUENTIAL ”/> indicates that a forward index having an index ID of lang is created. These index IDs are based on the inde mentioned in
It corresponds to x_id.

Next, the operation of the search execution section 200 will be described.

The search execution unit 200 uses the index file group 150 to determine the relevance between a search expression (described by a word, a Boolean operator combination of words and a filter expression) and each document to be searched. calculate. A search is performed by performing a “filter search” following the “normal search”. Here, the ordinary search basically uses a conventional full-text search technique, but the present invention is different in that an index to be actually used from a plurality of indexes can be specified by a search formula.

Here, a normal search (search using a reverse index) in the search execution unit 200 of the present invention will be described.

Using each word i included in the search expression as a key, the reverse index specified in the search expression is subtracted.

From the values obtained from the reverse index, a set of document IDs in which each word i appears and the importance Wwij of the word i in each document j are obtained.

The Boolean operator included in the search expression is appropriately processed, and the “ordinary search relevance” of each document is calculated using the importance of the word.

The documents are arranged in descending order of “normal search relevance”.

Next, a filter search (search using a forward index) in the search execution unit 200 will be described.

Using the document IDs of the top k documents of “ordinary search relevance” as keys, a forward index specified by the search formula is retrieved.

From the values obtained from the forward index, a set of words appearing in each document and the importance of the words is obtained.

The Boolean operator included in the search expression is appropriately processed, and the “filter matching degree” is determined using the importance of the word.
Is calculated.

When the “and type filter” is specified in the search expression, and the “filter suitability” obtained by is 0, the “fitness of the entire document” is set to 0. When 'or-type filter' is specified in the search expression, the value obtained by adding the "filter suitability" obtained by "normal search suitability" to "the overall document suitability" is used.

Next, the normal search in the search execution section 200 will be described.

In the normal search, after the filter search is completed, the documents are arranged in descending order of “the degree of suitability of the entire document” to obtain a search result. The following is an example of using structure information designation, normal search, and filter search.

Example 1) Search formula (1): (UNIX and ti = (Network)) and _filter (lang = (japanes)
e or english)) where and_filter represents 'and-type filter',
ti is the ID of the reverse index for the title part of the document, and lang is the ID of the forward index for the language part of the document. The processing when the search expression (1) is given is as follows.

A document set that includes 'UNIX' in the default reverse index and 'Network' in the reverse index ti is obtained, and based on the word importance and the 'and' operator, the document set of each document is obtained. Calculate “normal search fitness”.

The documents are sorted in descending order of “ordinary search suitability”, and the following processing is performed for each of the top k documents j.

An index is set using the ID of the document j as a key.
Subtract lang and check if its value is 'japanese' or 'english'. If it exists, calculate the "fitness of the filter" using the importance of the words "japanese" and "english". 'Filter fit' for documents where 'japanese' or 'english' is not included in the value of index lang
Becomes 0.

Since the “and type filter” is specified, the “fitness of the entire document” is 0 for a document whose “filter suitability” is 0, and “the entire document” is a document whose “filter fit” is positive. The “fitness” is “normal search fitness”.

Finally, when the “degree of relevance of the entire document” is positive, the documents are arranged in descending order of this value and output as a search result.

As described above, using the retrieval formula (1),
A "document set including UNIX and a title including" Network "and described in Japanese or English" is obtained as a search result.

Example 2) Search formula (2): (UNIX or ti = (Network)) or _filter = (bold = (linux an
d solaris)) where or_filter represents 'or type filter' and ti
Is the ID of the reverse index for the document title portion, and bold is the ID of the forward index for the emphasized character. When the search expression (2) is given, the processing is as follows.

As in Example 1) above, 'UNIX is included in the default reverse index,
Retrieve documents that contain 'Network' in ti, and calculate the “normal search relevance” of each document based on the importance assigned to each word and the 'or' operator.

The documents are sorted in descending order of “ordinary search suitability”, and the following processing is performed on each of the top k documents j.

An index is set using the ID of the document j as a key.
Subtract bold and add 'linux' and 'solaris'
(Registered trademark) '. If so, the importance of the words 'linux' and 'solaris' and 'and'
'Calculate the degree of filter fit using the' operator. 's
Documents that do not include both olaris' and 'linux' in the index bold entry have a "filter fit" of zero.

The sum of the “normal search relevance” obtained above and the “filter relevance” is referred to as “the relevance of the entire document”.

Finally, the documents are arranged in descending order of “fit of the entire document” and output as a search result.

As described above, using the retrieval formula (2),
"Include 'UNIX' or add 'Networ
As a search result, a set of documents including “k”, in particular, a document in which both “solaris” and “linux” are emphasized in the search result is obtained.

[0091]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS As an embodiment, how to solve the above-mentioned problem will be described below.

(A) Solving the problem that it is impossible to calculate the degree of conformity reflecting the structure information of a document: In the present invention, by describing the structure information of a sentence in the format file 120, a specific partial structure of the structure of the document can be obtained. , An index file 150 can be created. In addition, by assigning an ID to such an index file 150 and describing this ID in a search formula, it is possible to select an appropriate index file at the time of search and calculate the fitness. Therefore, the structure included in the document can be reflected in the calculation of the degree of matching.

For example, "somewhere in the document, 'artificial intelligence'
I would like to especially increase the moderateness of documents that include 'learning' in the title. Can be answered as follows.

An index file 150 is created using the format file 120 shown in FIG. At this time, a title portion (a portion surrounded by a TITLE tag) is created as an index ti in addition to the text portion.

A search is executed using the following search formula.

[0096] Artificial intelligence or ti = (learning) (b) Solution of the problem that a filter using attributes attached to a document cannot be realized: In the present invention, a simple operation of describing attributes in the format file 120 is used.
A reverse index and a forward index corresponding to a specific part of the document structure can be created. By specifying 'and_filter' or 'or_filter' in a search expression, a search using a forward index can be performed in addition to a search using a reverse index. Therefore, a filter using the attributes attached to the document can be easily realized. For example, a search request that includes an 'and-type filter', "I want to retrieve only Japanese documents containing 'UNIX'" can be answered as follows.

An index file 150 is created using the format file 120 shown in FIG. At this time, the language used (part enclosed by <LANG>-</ LANG>) assigned to the document in advance is created as a forward index lang.

A search is executed using a search expression including the following 'and-type filter'.

UNIX and _filter = (lang = (japanese)) Similarly, when a document containing 'knowledge', which is a search request containing 'or type filter', and the author is 'Suzuki', Want to be higher (I want to rank higher)
Then you can answer as follows.

An index file 150 is created using the format file 120 shown in FIG. At this time, author information (portion enclosed by <AUTHOR> tags) assigned to the document in advance is created as a forward index auth.

A search is executed using a search expression including the following 'or type filter'.

Knowledge or _filter = (auth = (Suzuki)) In the above embodiment, the format file 120
Is created, and the index file 150 is created based on the description of the format file 120. In the structured document, which part of the structure of the document is to be indexed is described. You may do so.

In the above description, the document index unit 100 and the search execution unit 200 have been described. These operations are constructed as a program, and a disk device connected to a computer used as a full-text search device, a floppy The present invention can be easily realized by storing it in a portable storage medium such as a (registered trademark) disk or CD-ROM and installing it when implementing the present invention.

The present invention is not limited to the above-described embodiments, but can be variously modified and applied within the scope of the claims.

[0105]

As described above, according to the present invention, the document structure information and the filter information can be obtained by a simple operation of describing in the format file which index of the structure of each document is to be created. , The appropriate moderateness calculation can be performed, and the convenience of the full-text search system can be improved.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of a full-text search device of the present invention.

FIG. 4 is an example of a structural tree used in the present invention.

FIG. 5 is an example of a format file according to the present invention.

FIG. 6 is an example of a forward index according to the present invention.

FIG. 7 is a configuration diagram of a conventional full-text search device.

FIG. 8 is an example of a conventional reverse index.

FIG. 9 is an example of a structured document.

[Explanation of symbols]

REFERENCE SIGNS LIST 100 document index creation means, document index section 110 document database to be searched 120 format file 150 index file group (index file) 151 reverse lookup file 152 forward lookup file 200 search execution means, search execution section 300 format file creation means

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Yoshihiko Hayashi 2-3-1 Otemachi, Chiyoda-ku, Tokyo F-term in Nippon Telegraph and Telephone Corporation (reference) 5B075 ND03 ND35 PP23 PQ02 PQ74 UU06

Claims (18)

[Claims] 1. A "fitness" that indicates to what degree each document (structured document) having a structure to be searched matches a word or a search expression formed by combining Boolean operators of words. In a full-text search method that calculates documents and sorts documents in descending order of relevance, and determines the index corresponding to each part of the structure of each document in the structured document, or in an independent format file And a plurality of index files are created based on the description of the format file, and one or more of the indexes of the created index file are designated by a search expression. A full-text search method characterized by using structural information of a document for calculating a degree of relevance. 2. When describing a partial structure of a document to be indexed in the structured document or in an independent format file, if the structure of the target document is nested, By creating a tree structure with the elements representing the respective structures as nodes and the elements representing the entire document as root nodes, and describing the path from the root node in the format file. 2. The full-text search method according to claim 1, wherein a portion of the index to be indexed is specified. 3. When the index file is created, a reverse index using a word as a key, and a set of a pair of a document ID in which the word appears and the importance of the word as a value,
A document ID is used as a key, and one or both of a forward index and a value of a set of a pair of a word appearing in the document and the importance of the word are created based on the description of the format. 2. The full-text search method according to claim 1, wherein an index ID is assigned to each. 4. The full-text search method according to claim 1, wherein a search expression is specified by using an index ID assigned to each index to determine which of the plurality of created indexes is to be used. 5. When calculating the degree of relevance between a search expression described by a word, a Boolean operator combination of words and a filter expression using the index file, and each document to be searched, The full-text search method according to claim 1, wherein a search is performed using the reverse index, and a search is performed using the forward index. 6. When performing a search using the forward index, a condition specified by the search formula is selected from a predetermined number of documents having a high degree of relevance among acquired search results. 'An' to retrieve only matching documents and use them as search results
A request using a 'd-type filter' and an 'or-type filter' that increases the relevance of documents that match the conditions specified by the search expression, sorts the documents in descending order of the corrected relevance again, and sets the results as search results Item 5. The full-text search method described in Item 5. 7. A “fitness” that indicates to what degree each document (structured document) having a structure to be searched matches a word or a search expression formed by combining Boolean operators of words. Is a full-text search device that calculates documents and arranges documents in descending order of relevance to obtain search results, and determines which part of the structure of each document is to be indexed in a structured document or independent. Format file creating means to be described in a format file; document index creating means to create a plurality of index files based on the description of the format file; one or more of the indexes of the created index file A search expression specifies whether to use an index, and uses the document structure information to calculate relevance. A full-text search device comprising: 8. The format file creating means, when describing a partial structure of a document to be indexed in the structured document or in an independent format file, nests the structure of the document to be indexed. Means for analyzing the document, creating a tree structure having elements representing the respective structures as nodes and elements representing the entire document as root nodes; and a path from the root node in the format file. Means for designating which part of the structure of the document is to be indexed by describing the document. 9. The document index creating means includes: a reverse index using a word as a key, and a set of a pair of a document ID in which the word appears and the importance of the word as a value;
Using a document ID as a key, one or both of a forward index and a value of a set of pairs of words appearing in the document and importance of the words are created based on the description of the format file. 8. The full-text search device according to claim 7, further comprising means for assigning an index ID to each of the indexes. 10. The search execution means includes means for designating a search expression from among the plurality of created indexes by using an index ID assigned to each index. Full text search device. 11. The retrieving means, using the index file, is a relevance between a search expression described by a word, a Boolean operator combination of words and a filter expression, and each document to be searched. 8. A fitness calculation means for calculating a search index, wherein the fitness calculation means includes a normal search means for performing a search using a reverse index, and a filter search means for performing a search using a forward index. Full-text search device described. 12. The filter search means retrieves only documents that match a condition specified by the search expression from a predetermined number of documents having a high degree of relevance among the obtained search results. 'An the result
Means using a 'd-type filter' and an 'or-type filter' that increases the relevance of documents that match the conditions specified by the search expression, sorts the documents again in descending order of the corrected relevance, and sets the results as search results The full-text search device according to claim 11, further comprising: 13. A “fitness” that indicates to what degree each document (structured document) having a structure to be searched matches a word or a search expression composed of word Boolean operators. Is a storage medium that stores a full-text search program that arranges documents in descending order of relevance and obtains search results, and determines which part of the structure of each document is to be indexed in the structured document. Or a format file creation process described in an independent format file; a document index creation process for creating a plurality of index files based on the description of the format file; and one of the indexes of the created index file. Alternatively, specify which index to use in the search expression, and A storage medium storing a full-text search program, comprising: a search execution process that uses structural information for calculating a degree of relevance. 14. The format file creation process according to claim 1, wherein when describing a partial structure of the document to be indexed in the structured document or in an independent format file, the structure of the target document is nested. A process of analyzing the document and creating a tree structure with elements representing respective structures as nodes and elements representing the entire document as root nodes; and a path from the root node to the format file. 14. A storage medium storing a full-text search program according to claim 13, further comprising: a process of specifying which part of the structure of the document is to be indexed by describing the document. 15. The document index creation process includes: a reverse index using a word as a key and a set of a pair of a document ID in which the word appears and the importance of the word as a value;
Using a document ID as a key, one or both of a forward index and a value of a set of pairs of words appearing in the document and importance of the word are created based on the description of the format file, and the index is created. 14. The storage medium storing the full-text search program according to claim 13, including a process of assigning an index ID to each of the first and second programs. 16. The search execution process includes a process of designating which of the plurality of created indexes to use by using an index ID assigned to each index. A storage medium storing the full-text search program of the present invention. 17. The retrieving execution process, using the index file, a relevance between a search expression described by a word, a Boolean operator combination of words and a filter expression, and each document to be searched. 14. A fitness calculation process for calculating a fitness index, wherein the fitness calculation process includes a normal search process for performing a search using a reverse index and a filter search process for performing a search using a forward index. A storage medium that stores the described full-text search program. 18. The filter search process retrieves only documents that match a condition specified by the search expression from among a predetermined number of documents having high relevance in the obtained search results. 'An the result
A process using a 'd-type filter' and an 'or-type filter' which increases the relevance of documents matching the conditions specified by the search expression, sorts the documents again in descending order of the corrected relevance, and sets the results as search results. The full-text search device according to claim 17, further comprising: