JP3376996B2 - Full text search method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION [0001] The present invention relates to a document database.
To search for the full text of the document by specifying a character string.
Text search method, especially for searching
Equivalently speed up full-text search processing using files
The present invention relates to a search method and an apparatus for searching. [0002] In a conventional document search system, registration is performed.
Index words (referred to as keywords) that represent the contents of the document
Is adopted. However, in this method
Indexers, called indexers, write documents
Read each word, understand the content, and assign appropriate keywords
Needed. Avoid the troublesome work of registering
For example, the text such as “Japanese Patent Laid-Open No. 63-198124”
Index words that appear in all words as keywords
A method of registering in a file has also been proposed. But above
Method is significant when creating an index file
Difficult to determine the smallest unit word, there is a word dictionary
Otherwise, the parsing of the sentence failed due to incomplete grammar rules.
Important words are not extracted as keywords
There's a problem. [0003] In order to solve this problem, documents are searched at the time of retrieval.
Register it directly in the computer as character-coded text,
When searching, the contents of all documents in the text database
And read the given keyword (conventional system
In order to distinguish from control keywords, search terms
Full text search to find documents containing
Is being planned. This full-text search method uses the information
Processing Society of Japan vol. 89, no. 66 Informatics
Basic 14-7 Text Database Management System SI
GMA and its Applications (July 27, 1989) "
As mentioned in the head, the entire text file
A major feature is that scanning is performed one character at a time from the beginning. This
By doing so, the document identifier etc. corresponding to the keyword
Even if there is no index file that describes
Database text as a clue
It becomes possible. That is, the text in a given search term
Searches the entire data string for a search term
Only those documents that exist can be output as search results.
However, this full-text search method does not
To scan the entire file one character at a time from the beginning.
It takes time and cannot be applied to large databases
There was a problem. As seen in Section 2 of the same document
Even if you have a general-purpose large computer, it is about 2MB / s
Only the search processing speed can be realized. Even at this speed,
Search time is practical for a gigabyte database
Enter the area. However, databases on a practical scale such as offices
Source requires hundreds of megabytes of storage,
Will not get a sufficient search response. [0004] To solve the present invention
The challenge is to target a practical-scale text database.
Search with sufficient search time that is practically acceptable
Fast full search for the full text of the resulting document
A text search method and apparatus are provided. [0005] In order to solve the above problems,
Has a full-text server consisting of the following processing steps:
An apparatus for implementing the method is constructed using the approach method. (1) Step of storing the text itself (2) Decomposition of the stored text into partial character strings at word level
And the mutual inclusion of the character strings between
Checks and excludes strings included in other substrings
Step of creating a condensed text composed of a set of character strings (3) A sentence in which characters used in the text are collected without duplication
(4) Decomposing the given search term at the character level,
Extract only documents that contain all the characters that make up the search terms
(5) Condensed text corresponding to the document extracted from the character component table
And extract documents that contain a given search term
Condensed text search step (6) If the given search condition expression is a book between multiple search terms
If you specify the positional relationship in the sentence,
Refer to the text data corresponding to the extracted document, and
Search terms that have been added and between search terms
Only those that satisfy the search conditions such as the specified positional relationship
Steps for text search. Hereinafter, a first embodiment of the present invention will be described.
This will be described with reference to FIG. This device has a display 1
00, keyboard 101, central control unit CPU102,
Character component table 105, condensed text 104, and text 103
Delivery file 110, floppy disk driver 10
6. It is composed of the main memory 200. The main memory 200 has a text registration program
201, condensed text creation registration program 202, character component
Table creation registration program 203, character component table search program
Ram 204, condensed text search program 205, text
Program 206, hierarchical pre-search control program
While the data area 207 is stored.
Is secured. These programs are executed by the CPU 102.
Be executed. When registering a document, use the keyboard 101
The CPU 102 causes the floppy disk
Is the floppy disk inserted into the disk driver 106
Read the document data and execute the body registration program 201
Document data that has been read
File 110. Next, the CPU 102
Executing the registration program 202, the text 103
Split into substrings at the level, and between the divided substrings
Examine the containment relationship between strings and include them in other substrings
Condensed book consisting of a set of substrings excluding the string
A sentence is created, and this is used as the condensed text 104 in the file 11
Store to 0. Finally, the CPU 102 creates a character component table.
Execute the registration program 203 and use it in the text 103
Create a character composition table that collects the characters that are not duplicated,
This is stored in the file 110 as the character component table 105.
You. When searching, input from the keyboard 101
The retrieved search condition expression is sent to the CPU 102. CPU
At 102, the hierarchical search control program 207 is executed first.
And a character component table search program 2 based on the control.
04, Condensed text search program 205, Text search program
The program 206 is sequentially executed. That is, the character component table
In the search, the search terms in the entered search condition
Decompose at the character level and remove all characters that make up the search term
Extract only the containing documents. And extracted in the character component table
Given condensed text corresponding to the given document and given search
Extract documents containing terms. If the given search terms
A single search term or multiple search terms in the expression
Only the logical relationship between the programs is specified.
If the positional relationship is not specified, search here.
The search is terminated and the results of the condensed text search are output as search results.
Power. Otherwise, ie given search condition
Specify the positional relationship in the text between multiple search terms in the expression
Documents that have been extracted by the condensed text search
Refers to the text data corresponding to, given search terms
And the positional relationship between search terms, etc.
Only those that satisfy the search conditions of
Output. The above is the full text of the first embodiment of the present invention.
3 is a schematic diagram of a search device. Hereinafter, a character component table server which is a feature of the present invention will be described.
H, search for condensed text, and search and narrow text
Outline of registration and search method of hierarchical pre-search method
The abbreviation will be described. First of all, the "condensed text" and "character composition table"
This is done automatically when the document is registered. This processing content,
As shown in FIG. In this figure, when the document to be registered is entered,
First, it is stored as "text" as it is. Next, this "book
Create a "condensed text" from the sentence. "Condensed text"
Character types (Kanji, Hiragana, Katakana,
Character string) and split the string
Created without duplicates. If the text is "Fuzzy search
In the case of Document 1, "Search technology for ...
"Search" is truncated as a duplicate word and "Fuzzy"
“Search technology” and “for” remain as “condensed text”
Will be. In addition, a "character composition table" is created from "text".
To achieve. Here, 1-bit characters appearing in the "body"
Of information. In the example of Document 1, there are "A" and "I"
'1' for each, and '0' because there is no 'U'
Set. “Inspection” and “Cable” are similarly set to '1' respectively.
Set. In the same manner, the corresponding character part of the character component table
If there is the character in the "body", "1"
If not, set '0'. In this way, the statement
"Condensed text" and "character composition table" automatically when registering
Create and prepare for hierarchical presearch. At the time of retrieval, as shown in FIG.
Refer to these auxiliary files in the following order. First,
In the character component table search, the character component table is referred to and searched.
Corresponding character in the character composition table corresponding to all characters in the term
Select the one with '1' set in the part. Second
In the condensed text search, the sentence selected in the character component table
Search term given to the conditional expression by referring to the condensed text of the book
Select documents containing. Finally, in the text search, the search
Only those that appear in the body of the
Choose out. In the example of this figure, search [4C] understanding, that is,
"" Search "and" Understanding "are within 4 characters
Example of searching with the conditional expression "Search for what appears"
I have. As a result, "Search" and "Understanding" of Document 4 are in the text
Extracts documents that are four characters apart. Hereinafter, character type division / duplication used in this embodiment will be described.
Creation of word exclusion type condensed text and character code dependent type character component table
Method and hierarchical presearch control method using the same
Will be specifically described. First used in this example
On how to create condensed text
Will be explained. As shown in FIG.
Split a character string by character type. The character type at this time is
Kanji, Hiragana, Katakana, English letters, numbers, symbols, etc.
is there. For each character string consisting of a sequence of these single character types,
Split the body text string. Next, that of the split string
For each, the substring in the same document
If a substring is included in its entirety, the string
Is excluded from the target of the condensed text as a duplicate character string. example
For example, if the substring "search"
Is completely included in the substring "intelligent search technology"
Therefore, this “search” is not registered in the condensed text. Only
Then, in the condensed text search, even if the character string "search"
Is "intelligent search technology" even if it is not registered in the condensed text
Will be hit as a substring. like this
In the sub-string obtained by eliminating duplicate registration of sub-strings
As shown in FIG. 5, a character string is separated between character strings for each document.
Insert a translator. In this figure,
The numbers ',' are used. This separator is shown in FIGS.
Is represented by the symbol '|'.
It does not need to be represented as a letter and is not assigned to a letter.
You can also use special codes. Next, the character code dependent type sentence used in this embodiment
A method of creating a character component table will be described. As shown in FIG.
The character code dependent character component table is
Bit position to set 1 as bit information indicating presence
Is a character component table for determining. In this figure, the shift JIS
This is explained using the mode as an example. In the figure, (XXXX) H is a character
The code is displayed in hexadecimal. For example, "search"
Indicates that the character string exists in the body of document 1.
In the bit list of document 1, (8C9F) H, (8DF
5) Set 1 to H-th. This character in the bitlist
The bit position corresponding to is called the entry number of the character component table.
I will do it. For example, the entry number of “check” is (8C9
F) 35999 if H or decimal is displayed. Hierarchy using the above character component table and condensed text
The control and search operation of the pattern pre-search will be described.
First, the search terms in the search condition expression should be
Decompose and perform a character component table search. Given here
Bit list corresponding to the character code that constitutes the search term
List in which all entry number positions in the list are 1
You will need a document with For example, "search"
Is given as a search term,
(8C9F) of the bit list corresponding to 'Detect' and 'Search'
H, (8DF5) Document where the H-th bit is all 1s
1, 2, 3, 4,. . . The character component table search results
And That is, the “test” is indicated as shown in FIG.
C9F) A bit list 701 of H entry numbers,
Bit search of the entry number of (8DF5) H that indicates 'search'
Performs an AND operation for each bit with the
The AND operation result 703 is obtained. This bit AND operation
Bit that is 1 in the bit list of result 703
The document number corresponding to the position is the search result of the character component table search
As a hit document. That is,
Documents containing all of the search and search will be extracted.
You. Also, the search term is only one character like “lake”
If it is composed of
Is output to terminate the search. Next, of the document extracted by the character component table search,
Search for the condensed text. Here, as shown in FIG.
Scan the contents of the condensed text registered for each document
Extract documents that contain the obtained search terms as words.
In other words, the two characters “search” and “search” are consecutive with “search”
Extract only those documents that appear. In other words, 'test' and 'chord'
Even if is included, like "detection" and "search",
Anything that appears as another word is truncated here
I will. For this purpose, it is narrowed down by the character component table search.
Same as text data for condensed text for each document
Search while scanning one character at a time. this
At the time, it corresponds to the document number of the result obtained by the character component table search
Only scan the condensed text. For example, a character composition table
If the search results are document numbers 1, 2, 3, 4,. . . That
For example, in the condensed text search, document numbers 1, 2, 3,
4,. . . Scan the condensed text of. And actually
Documents whose search terms exist in the condensed text
Output as search results. As described above, in the "hierarchical pre-search method",
Is a two-stage pre-
Perform a search in advance to find the "character level" and "word
Text search that takes the longest time
Narrow down the number of documents targeted for
As a result, the search document size can be reduced.
And a very fast full-text search is equivalently realized.
Will be able to. That is, in the character component table search, the sentence
The character component table indicates the presence of a character with 1-bit information.
Therefore, the amount of data to be searched can be extremely small.
As a result, search time can be reduced in a short time.
You. In addition, a bitlist for each character that makes up the keyword
By taking the logical conjunction of
Documents are greatly truncated, and subsequent documents are markedly narrowed down.
Can be included. In the condensed text search,
Search because the amount of data is smaller than scanning the sentence directly
Processing time can be reduced. Next, a second embodiment of the present invention will be described.
In this embodiment, even when a plurality of search terms are specified,
Full-text for efficient hierarchical pre-search
It provides a strike search method. For example, ""
When the conditional expression "search" AND "understand" is given
First searches the character component table as the first step
You. Here all the sentences for each given search term
Search for documents containing characters, then given between search terms
Output a document that satisfies the conditions. "" Search "AND
In the case of the conditional expression "understand", two characters "search"
And a document that includes the two characters “understanding”. sand
In other words, "('test' AND 'search') AND ('rea' AND
“Solution”) ”
AND 'solution', that is, a document containing the above four characters simultaneously
Search for. Next, narrow down the result of this character component table search
Search the condensed text of the placed document. Here,
Extract only documents where the specified keyword appears as a word
You. That is, a sentence that includes both “search” and “understanding” at the same time
Search for books. As in this example, the relationship between search terms
If only the logical conditions such as "AND" and "OR"
Specified the condition that defines the positional relationship between keywords.
If not, end the search here, and search for condensed text
Is output as the final search result. If the location condition
If is specified, extracted by condensed text search
Search the text of the document and extract those that match the specified conditions.
And outputs this as the final search condition. That's the truth
5 is a description of a search operation in the embodiment. Thus, the character
Conjunction between search terms in component table search and condensed text search
To search for multiple search terms.
Even when the hierarchical pre-search is performed efficiently,
A text search can be realized. A third embodiment will now be described.
The search control of hierarchical pre-search is explained in detail.
You. FIG. 8 shows the procedure for controlling the hierarchical presearch at this time.
This will be described with reference to a PAD diagram. Here, “computer” and “intelligent
Find Documents Containing Either of the Interfaces "
The search for “computer” or “intelligent interface”
This will be described using a cable type as an example. First, step 80
At 00, a character component table search is performed. Here the given inspection
For each search term, search for a document that contains all its characters,
Sentence that satisfies the compound condition given between post-search terms
Output the certificate. In this example, as shown in FIG.
Character components for each of the three characters that make up the machine
Performs a bit AND operation between the corresponding entry numbers in the table.
Next, 9 items that make up the “intelligent interface” in the same way
Corresponding entry in the character composition table for each of the characters
Performs a bit AND operation between the serial numbers, and creates
AND operation result for "computer"
The bit string is ORed. That is, "('Total'
AND 'calculation' AND 'machine') OR ('knowledge'AND'like')
AND'i'AND'n'AND'ta'AND'h'A
ND'e'AND '-'AND's')"
Will be executed. This constitutes a "computer"
Or a document that contains all three characters
Document that contains all nine characters that make up the
Is extracted. No result in the character component table search
Then, as shown in FIG.
Output the result and end the search. Also, like 'lake'
If the search term consists of just one character,
To output the result of the character component table search and terminate the search. If the search term is composed of a plurality of characters,
And the number of results of the character component table search must be 0
Then, a condensed text search is performed next. Registered in the condensation text
The content is a character string divided for each character type.
For example, characters such as "intelligent interface"
If the species is different, the condensed text is decomposed into substrings,
Separators are inserted at the division points, such as
You. Therefore, different “intelligent interfaces”
For search terms composed of different character types,
Does not find a matching character string when searching for condensed text
Will be. Therefore, before entering the condensed text search,
Search terms and search terms composed of different character types
System divides this for each character type. In this way,
The search term that has been split
This is called a split search term to distinguish it from the term. And condensed text
The search is performed, for example, on “computer”, “intelligent”, “interface”.
Search terms, such as “
For split search terms, between terms that have the same split source
To perform a search under AND conditions. For example, "Computer" OR
In the case of the conditional expression “Intelligent Interface”, the condensed book
In the sentence search, "" computer "OR (" intelligent "AND" in "
Interface)), that is, "intelligent" and "interface"
Source is in the same document, or
Search for existing documents "as a conditional expression.
And If the result of the condensed text search is 0,
Here, a search result of 0 is output and the search is terminated. Ma
If there is a neighborhood condition or context condition specified, or
Is a searcher that is divided like an "intelligent interface"
Search terms and split search terms are different.
Text search is performed only when it becomes necessary. If not, this
This ends hierarchical pre-search and searches the results of condensed text.
Output as result. Here, the context condition is, for example,
Like "computer" [S] "intelligent interface"
In the conditional expression shown, this is “computer” and “intelligent
Search for a face that is in the same sentence
"Means". Or the neighborhood condition
For example, “Computer” [10C] “Intelligent Interface
This is described as ""
Machine and intelligent interface are within 10 characters
Find the document that appears. " That is, the sentence
Search terms that appear in the document for both pulse and neighborhood conditions
This is a search condition that specifies the placement relationship. [0021] The order of search terms appearing in such texts
If a search condition that specifies the placement relationship is given, or
In the condensed text, the character type
If a changing search term is given,
Refer to the text data corresponding to the result of the
Search for only those terms that appear in the text exactly as they appear
The result is output, and the search ends. This
If the search terms consist of different character types,
Or the occurrence position in the text between search terms
Even when there is a condition specification, the hierarchical
A high-speed full-text search
Can be. Next, a fourth embodiment of the present invention will be described.
I do. This embodiment is based on the character component table in the first embodiment.
The capacity has been reduced to make it more compact. First fruit
The character code-dependent character component table used in the examples is simple to process.
Simple, but a bit list per document in the character composition table
, There is a problem that the character component table becomes large.
Also, although the corresponding character code does not exist, the entry number
There is a problem that there are many useless parts because
You. For example, in the case of shift JIS, from (0000) H
During (8140) H and from (A000) H to (E04
0) During H, that is, from 0th to 33087th and 4
For entry numbers from 0960th to 57408th
Has no corresponding character code. Nevertheless, the sentence
This part is used to determine the entry number by
It is necessary to have all minutes as table entries. This
Statement to eliminate useless parts in the
Convert character code and set bit positions from zero
Create a character composition table for use. This character code
The details of the embodiment using the character conversion table of the character conversion type are as follows.
explain. To create a character code conversion type character component table
The following formula is given as an example of the conversion formula to the character code. Ma
The corresponding PAD diagram is shown in FIG. if SJIS <(A000) H then SCODE = SJIS− (8040) Helse SCODE = SJIS− (C040) H SCODE = SCODE− (SCODE / 256) × 64 (64-1) Formula (4-1) However, the part of the small value of the normal character code is the control code
In this formula, (8140)
Leave some margin as (8040) H instead of H
ing. In addition, the small calculation result of (SCODE / 256)
Truncate below a few points, multiply the result of truncation by 64
Do. ) In the formula, SJIS indicates the original shift JIS code, and SJIS
CODE indicates the character code after conversion. KEIS code
And other code systems are also compatible with Shift JIS code.
Conversion to SCODE is possible with the same formula
Noh. When the expression (4-1) is represented in a character code table, FIG.
It means a conversion like 1. That is, (000
0) From (H) to (FFFF) H, (8140) H to
(9FFC) H and (E040) H ~ (FEFC)
H and the character codes that are distributed
0) Change the character code so that it is located
Will be replaced. Using this equation (4-1), code conversion
In this way, as shown in FIG.
The length can be very short, and the whole
The capacity can be reduced. The control of the hierarchical presearch is performed in the first embodiment.
Is the same as That is, the control procedure of FIG.
First, use the characters in the search term
Secondly, a condensed text search is performed using a search term.
Do. Output search results here if no context condition is specified
And terminate the search. Third, if the context condition is specified
Performs a sentence search and outputs the result. However, character composition table
At the time of search, all of the input search terms are (4-
1) Using character code conversion based on the formula
Become. As described above, the fourth case using the character code conversion type character component table
The embodiment has been described. According to this embodiment, the character
Code and convert the bit positions from the 0th
By creating a well-arranged character component table,
Can eliminate entries that are not assigned
The size of the character composition table can be very small.
You. Next, a fifth embodiment of the present invention will be described.
I do. This embodiment is based on the character component table of the fourth embodiment.
The capacity has been further reduced using hashing techniques.
You. The capacity of the character component table of the fourth embodiment is further reduced.
Therefore, in this embodiment, one entry in the bit list is used.
Assign multiple characters to the serial number. Ie hash
You can use functions to search for characters in search terms and
The method of associating the cut positions is adopted. This hash function and
Then, for example, the following equation can be used. In the equation, SCODE is shifted according to equation (4-1) according to JIS.
This is the character code converted from. mod takes the first argument
This function outputs the remainder after dividing by the second argument. N is optional
Is an integer value. For example, 512 is used as N
"A" is entry number 480, "Ma" is entry number
No. 193. An example of the character component table created in this way is
As shown in FIG. In this case, N was set to 512,
Only 512 bits are required to register one document
I understand. When searching, each sentence of the given search term
In the same way as when registering characters,
The entry number is calculated using the
Refer to the bit position in the character component table. For example, "Aima
In the case of the character string "i", the entry number 4 as shown in FIG.
A document in which all bits at positions 80, 482, 193 are 1
Is the search result of the character component table search. In this way
Next, the condensed book of the documents found by the search
Search for sentences. Hereinafter, the condensed text search and text search system will be described.
The control procedure will be described with reference to FIG. First implementation
In the example, the search term starts from one character after the character component table search.
The search result is the character component table search result.
To output the hierarchical pre-search. But,
In the character component table search of the character component table used in this embodiment,
Is a condensed text because of possible search noise
It is necessary to continue the hierarchical presearch until the search. An example
For example, hiragana 'wa' (shift JIS code (82C
D) H) is the entry number 13 in the equation (5-1).
But the kanji 'ship' (shift JIS code (8ACD)
H) has the same entry number 13. This means that the search
If the term “ship” is given, a sentence containing “wa”
All books are also searched as a result of the character component table search.
Will be. Therefore further scan the condensed body
To actually extract the document that contains the kanji “ship” and inspect it.
It will be output as a search result. As described above, the fifth embodiment
Was explained. In this embodiment, a hash function is used.
To assign multiple characters to one entry in the character composition table
By doing so, the capacity of the character component table can be significantly reduced
The effect is obtained. Next, a sixth embodiment will be described. Fifth
In case of simple hashing as in the example of
Characters that easily appear in documents such as
Characters that rarely appear like the Kanji
The possibility of becoming a re-number comes out. For example, hiragana
'Wa' and kanji 'ship' have the same entry number 13.
If “ship” is given as a search term,
All included documents are hit as a result of character component table search
Will do. Hiragana 'ha' is a Japanese document
Almost all documents are text because of the very frequently used text
Hits in the ingredient list search. Thus, the character composition table server
When the rate of narrowing down in the
Search time increases due to the number of documents to be searched
Will be. In order to prevent such a reduction in the narrowing rate,
Must determine the hash function taking into account the frequency of character use.
It is necessary. The character component table used in this embodiment is
It is called a single-type character component table. Character type hashing type character
To create a composition table, for example, as shown in FIG.
Allocate the character component table entry area for each character type,
A hash function that wraps in the area by character code
make. To implement such a hash function, the character
After judging the character type by code, fold by mod function
May be returned, or a correspondence table between character codes and entry numbers
Can also be realized. An example of this hash function
FIG. 16 is a PAD diagram. In this embodiment, hiragana,
The number of katakana and alphabetic entries is assumed to be 20 and the symbol
The number of entries is 10, the number of numeric entries is 10, JIS
The number of first-level entries is 370, and the JIS second-level entries are
The number of birds is 61. First, the entered search term
To determine the character type based on the character code.
Of the assigned entry in the character composition table for each character type
The part is folded back using the mod function. That is, whether SCODE is (01DF) H
If it is in the range of (0231) H, it is a hiragana character string.
So, take that SCODE and mod it with 20
Let it be an entry number. SCODE is from (0240) H
(0296) If it is in the range of H, it is a katakana character string
So, take the SCODE mod 20 and
Value obtained by adding 20 which is the head of the katakana hashing area
Is the entry number. SCODE is (01A0) H
If it is in the range of (01DA) H, it is an alphabetic character string.
So, take the SCODE and mod it at 20,
Enter the value obtained by adding 40, which is the beginning of the character hashing area.
It is a bird number. SCODE changes from (018F) H to (0
198) If it is in the range of H, it is a numeric character string.
Take the SCODE of 10 as the mod, and add
The entry number is the value obtained by adding 70, which is the head of the
No. SCODE changes from (065F) H to (123
2) If it is in the range of H, use a JIS first-level kanji character string.
So, take the SCODE mod at 370,
In addition to this, the hashing area for JIS 1st level kanji character strings
The value obtained by adding the leading 80 is set as the entry number. SC
ODE is in the range of (125F) H to (1FDE) H
Is a JIS second-level kanji character string.
MOD was taken at 61 and JIS 2nd water
450 which is the beginning of the hashing area of the quasi-kanji character string
The added value is used as the entry number. Other than the above SCODE
In the case of, it is regarded as a character string with symbols and other character types
Then, modulate the SCODE by 10 and write it down.
The value obtained by adding 60, which is the head of the hashing area of the
It is a bird number. Using this character type hashing type character component table
The control procedure of the hierarchical pre-search was the same as in the fifth embodiment.
The same. That is, first, the characters in the search term
Second, a character component table search is performed using a search term.
Perform a condensed text search. Context conditions etc. are not specified
If so, end the search here, otherwise
Thirdly, performs a text search and outputs the result. Explanation above
As described above, according to the present embodiment, the sentence is
Characters that correspond to the character component table entry numbers for each character type
By using the type hashing type character component table,
Condensed text because noise in component table search can be reduced
Scans the document in the
Kist search becomes possible. Next, as a seventh embodiment, a character component
Improve the rate of narrowing down in table search,
Frequency information hashing type that can reduce the amount of scanning
Explains the control method of hierarchical pre-search using character component table
I do. How to create a frequency information hashing type character composition table
Is the frequency of use of characters in documents registered in the database
And determine a hash function based on the frequency information. frequency
For characters with large
Characters, and infrequent characters
Hash function so that multiple characters are included in the same entry
To adjust. By doing so, an average stable aperture
The insertion rate can be obtained by the character component table search. Ingredient
Specifically, as shown in FIG. 17, it can be obtained by equation (4-1).
Once in the database based on SCODE
Check the number of documents using, and sort by frequency. Next
From the most frequent to the number of entries in the character component table
Take only t. And the top of frequency distribution within Nt
Other entries, leaving only the entry with the highest frequency.
Entry numbers equal to or greater than Nt are sequentially assigned to the addresses. This
No. (Nt + 1) is assigned to the entry number of Nt or more
The first entry is an entry of Nt, and the (Nt + 2) th entry
To be the (Nt-1) th entry
Entries with higher frequency are sequentially assigned. allocation
Process, the entry with the highest frequency is always
Should not be assigned other entries. Assignment
The digit entry is in the form of a table as shown in FIG.
Remember and refer to this table to construct a hash function.
To achieve. That is, the character of SCODE is (095F) H
It can be seen that “check” is entry number 231. The control procedure of the hierarchical presearch is based on the fifth embodiment.
It is the same as the embodiment. That is, the control procedure of FIG.
First, use the characters in the search term to form characters.
Secondly, perform a table search, and use the search term to condense the text.
Perform a search. When no context condition is specified
Ends the search here, but otherwise,
3. Perform a text search and output the result. Thus, the book
According to the embodiment, statements actually used in the database
By creating a character component table based on the frequency distribution of characters,
In the character component table search, a consistently high narrowing rate
As a result, stable and short-term
The search processing time can be obtained. As described above, the embodiments having different character component tables are described.
Five embodiments have been described. The actual content of the condensed text is different
An example will be described. Condensation used in the first example
The text is easy to create, but can be seen in Figure 4.
So that it is not originally used for searches like "for"
Even a character string will remain in the condensed text. This is
This leads to a reduction in the compression rate of the compressed text. That is, scan on search
Search processing time increases because the amount of condensed text
Would. Decreasing the compression rate of such condensed text
The main factor is a series of adjuncts such as “for”
Register a character string that has no meaning by itself in the condensed text
It is in the place to do. Therefore, as an eighth embodiment, this search
Hierarchy using condensed body text from which unnecessary attachments are removed
The type presearch will be described. This condensed text is divided into character types
This is called a deduplication / adjunct removal type condensed text. This condensed text
As shown in FIG. 19, the method of creating
Split character type from column to substring, then duplicate
After removing the words, the attached words are removed. Character type division and weight
Deduplication is not different from the first embodiment. Attachment removal is heavy
This is performed on the de-duplicated hiragana character string. This attachment
The analysis for the word removal is performed as shown in FIG.
It is performed based on the connection rules between books and words. The basic word dictionary contains
Verbs and instructions composed of only hiragana as in Fig. 21
Pronoun, adjective, adjective verb, adverb, conjunction, particle, auxiliary
Part-of-speech and the conjugation ending of these parts-of-speech are registered together with part-of-speech information.
Has been recorded. In the example of this figure, the verbs <a>, <
<Naru>, <Uta>, etc. are registered with their endings.
Have been. The connection rules include each word registered in the basic word dictionary.
Register which words can be connected to. For example, FIG.
As shown in <verb-having adjunct form> and <noun-thing>
Are connected, and <noun-thing> is connected to <noun-
It is registered that it can continue. Such basic words
Hiragana substrings using appendix and connection rules
Judge whether the sentence is composed of
Determines whether to register a character string. For example, for
Is a substring <particle-of><noun-for>
Can be analyzed as a connected character string such as particle->
And remove it as a character string consisting only of attached words
You. On the other hand, the character string “ambiguity” is
To the condensed text as it is without exclusion
You. As described above, the attached words are analyzed and the hiragana sentence is
Eliminate character strings and useless information that is not used for search
Increase the compression rate of the condensed text by deleting it
Becomes possible. In addition, the basic word dictionary used for analysis and connection rules
The rule is that the traditional key word that the number of registered words increases with the times
Basically, it is a universal dictionary.
There is an advantage that it does not need to be updated once it is completed.
To exclude only those that can be parsed as adjuncts,
Even if a new word composed of hiragana that does not exist in the book appears
It will always remain in the condensed text. Next, character type division / duplication elimination / adjunct word removal
Control of Hierarchical Presearch Method Using Type Condensed Text
Will be explained. Character type division, deduplication, adjunct removal type condensation
In the text, the Hiragana character string is analyzed as ancillary words and converted to a condensed text.
May not register. Therefore, certain hiragana characters
If you try to search by column, the condensed text search is missed
It may be. For example, the character string “vertigo”
The verb conjugation ending "me" and the adjective "mai"
Can be analyzed as connected. A specific example is “Accepted
"Mai" is a noun.
Even if it is used, the condensed book as a result of attached word removal processing
It is deleted from the sentence. Therefore such a place
Search for condensed text with "vertigo"
The possibility comes out. Therefore, the search term is condensed text
Is it an original word in the middle, or too much condensed text
Check if the word may have been removed
It is necessary to search from. Search terms appear in condensed text
Checking whether a word should be recorded is done by creating a condensed text.
The adjunct removal algorithm used when
Well apply. In this example, the search term "vertigo"
Is given, judge this as a sequence of adjuncts
be able to. The above search control procedure will be described with reference to FIG.
You. First, a character component table search is performed. If the number of results is 0
If so, output 0 as a search result and end the search process
You. As described in the first embodiment, do not use a hash function.
If the search term is a single character,
The search result of the distribution table can be output as the final search result. You
That is, the character component table described in the first and fourth embodiments is
If used, determine whether the search term is a single character.
In addition, if it is one character, the result of the character component table search result
Is output, and the process ends. Fifth, sixth, seventh fruit
When using the character component table by the hash function described in the embodiment
Check if this search term is a single character
Is not performed, the next condensed text search is always performed. After this,
As in the case of the first embodiment, a divided search term is generated. Next, for each of the divided search terms
Perform attached word analysis. Includes one of the split search terms
If it is determined that the term is a condensed text,
May have been deleted.
Without performing, directly based on the results of the character component table search
Search. On the other hand, as a result of adjunct analysis,
If all are not adjuncts, the first embodiment
A condensed text search is performed in the same manner as. Neighborhood condition or statement
If no pulse condition is specified, or if the split search
If the search term is the same as
The result is output as the final search result, and the search ends. Also
If a neighborhood condition or context condition is specified,
Or if the split search term and the original search term are different
Then performs a full text search and the results are the final search
Output the result. Thus, according to this embodiment,
Analyze kana character strings and condense unnecessary adjunct strings
Type separation / duplication elimination / adjunct removal type condensation
By using the text, the compression rate of the condensed text is improved.
Thus, the search processing time can be reduced. Next, as a ninth embodiment, a hiragana character
Character type division, duplicate elimination, Hiragana sentence with all columns removed
Explain Hierarchical Presearch Using Condensed Text with String Removal
You. The condensed text explained in the eighth embodiment does not
Can be parsed incorrectly when parsing words
There is. For example, "vertigo" used in the eighth embodiment
In addition to the example of the character string,
In rare cases, it is not possible to correctly determine which is an adjunct word.
For example, in the document "Working in this application ..."
The substring "Toshino" is
Is used in the meaning of "~"
Judge whether it means a machine lever like "~"
Difficult. When used in the latter sense,
When you specify the search term “Lever”, “Lever”
Search for condensed text because it is not determined as an adjunct
Will be. On the other hand, in the creation of the condensed text,
Condensed because it is analyzed as an adjunct and deleted from the condensed text
Text search is missing. This adjunct analysis
To correct incompleteness of
To achieve hierarchical pre-search with a simple
This is the ninth embodiment. How to create this condensed text
Is shown in FIG. In this method, after character type division,
And remove duplicate registrations. This character type division / duplication elimination / Hiragana character
Control procedure of hierarchical presearch using condensed text with column elimination
Will be described with reference to FIG. First, the eighth embodiment
Similarly, a character component table search is performed. After this, the split search
Generate a system. Next, for each of the split search terms
Check whether the character string is a hiragana character string. Split searcher
If there is a hiragana character string in at least one of the
Performs a search based on the results of the character component table search without performing a search.
Search sentences directly. On the other hand, during split search terms
If there is no unique character string, the condensed text
If the neighborhood and context conditions are specified, or
If the split search term is different from the original search term,
Continue the search process until the text search. Thus, the real
According to the example, the condensed text that excludes all the hiragana character strings
Search for hiragana character strings by using
An accurate full-text search without leakage can be realized. Next, a tenth embodiment of the present invention will be described.
I will tell. In the ninth embodiment, the search term of Hiragana
, You need to refer to the text directly. I
As a result, more search time is required. There
So, even if Hiragana search terms are given,
As a method that allows full-text search, the tenth embodiment
Give an explanation. In this embodiment, the condensation used in the ninth embodiment
In the ninth embodiment, the removed hiragana character strings
Create the registered condensed text separately. As shown in FIG.
The remaining character string after character type division and duplicate registration elimination
Judge whether it is a Hiragana character string
Is registered as condensed text A, and the hiragana character string is condensed text.
Register as B. This will allow you to search only Hiragana
When given, the user can search for the condensed text B.
Will reduce search time.
You. FIG. 27 shows the procedure of actual hierarchical presearch search control.
Shown in First, a character component table search is performed as in the eighth embodiment.
Do. If the search result is 0, end the search here
You. Thereafter, a divided search term is generated. Next, the split inspection
String terms that are hiragana for search terms and other strings
Classified into terms consisting of After that, sentences other than Hiragana
If there is a split search term consisting of character strings,
Search. Next, if there is a Hiragana split search term
, Search for condensation B. After that, the first embodiment
Similarly, if there are neighbors, context conditions specified, or
If the search term is different from the original search term,
The search process is continued until the sentence search. In this way, hiragana
Separate from the condensed text of Nanano and the condensed text other than Hiragana
By storing, search term of any character type can be input
Even if it is done, the condensed text can be used effectively and it is always fast full
Text search can be realized. Next, an eleventh embodiment will be described.
In this embodiment, in order to increase the compression rate of the
It is based on a method that uses a separate condensed text for each
You. Character type division / duplication elimination for the condensed text used in this embodiment
-It is called a character type registration type condensed text. This character type split / heavy
Figure 2 shows how to create a deduplication / character type registration type condensed text.
As shown in Fig. 8, character type division and duplicate registration elimination were performed.
After that, determine the character type of the remaining substring and condense Hiragana
Text H, Katakana Condensed Text I, Kanji Condensed Text J, English Text
Condensed text K, number condensed text L, symbols and other character type condensed books
Classify and register as sentence M. By doing this, for example,
When searching with the kanji search term, the kanji character type
Since only the contracted body J needs to be searched,
The time can be further reduced. Specific hierarchical type
The control procedure of the research will be described with reference to FIG. First,
A character component table search is performed as in the eighth embodiment. Search result
If the number of results is 0, the search ends here. After this,
Generate a split search term. Next, split search terms into characters
Classify by species. Then, Hiragana split search terms
In some cases, the condensed H is
In the decomposition search term.
Select the condensed text to search according to the character type. That
Thereafter, as in the first embodiment, the designation of the neighborhood and the context condition is not performed.
If any, or if the split search term is the original search term
If not, continue the search process until the text search
You. In this way, separate the condensed text file for each character type
By reducing the volume of each condensed text,
Only, katakana only, or hiragana only
Fast full-text search for single-character search terms
The effect that can be performed is obtained. Next, a twelfth embodiment will be described with reference to FIGS.
This will be described with reference to FIG. This embodiment is described in Japanese Patent Application No.
Using the document search device proposed in 193015, the present invention
It has been realized. The main configuration of this device is a keyboard
3001, search expression analysis program 3002, bit server
Multi-processor 3007a, string search engine 3
006, Complex condition determination microprocessor 3045
a, search result storage memory 3046, display 302
0, semiconductor memory device 3010a, RAM disk device
3010b, collective magnetic disk 3010c, and search
An execution control program 3008 is provided. Semiconductor memory device
The character component table is stored in the storage 3010a.
Reference numeral 010b denotes a condensed text, a collective magnetic disk device 301
0c stores the text. However,
The distribution table and the condensed text are stored on the collective magnetic disk 3010c.
Are stored, and are semi-conductive at the start of operation of this equipment.
Body memory device 3010a and RAM disk device 301
0b. The procedure of the hierarchical pre-search control has been implemented up to now.
It is not different from what was explained in the example. Implementation up to now
The difference from the example is that the character composition table
Is stored on a RAM disk and the text is stored on a collective magnetic disk.
And a microprocessor dedicated to character component table search
String search, condensed text search and text search
That is, the use of a search engine. Search procedure
Will be described below. Search condition input from keyboard 3001
The formula is a search machine control microprocessor MPU03.
Analyzed by the search expression analysis program 3002 on 050
It is. That is, the search expression analysis program 3002 searches for
Keyword parts that make up search condition expressions and their inclusion conditions
And a decoding condition description section that describes the arrangement condition. Parcel
Containment conditions are described as logical conditions, and placement conditions are neighborhood conditions.
And context conditions. After separation and extraction,
The keyword part is also a synonym exhibition on MPU03050
The program is passed to the open program 3003.
It is passed to the condition analysis program 3041. Synonym expansion
In program 3003, the built-in synonym dictionary is
Refers to a synonym for the entered keyword
You. And the keyword group expanded here is synonymous.
It is passed to the notation development program 3004. In the case of this example
In the case of "computer", "computer", "computer",
“COMPUTER” or the like is generated. Different notation expansion
In the program 3004, the keyword entered here
A different notation development process is performed on the C group. In the case of this example
If "computer" is changed from "computer" to "C
"Computer" generated from "OMPUTER"
Is done. Keywords developed in this way as synonyms and variants
Then, the automaton generation microprocessor M
Automaton generation program 3 on PU13005a
005. Automaton generation program 30
05, sent from the different notation development program 3004
For all keywords that have been
Key to generate tomato and match with state transition table
As the identification code information of the word, the search engine 300
Set to 6. Search engine 3006 is a finite automatic
This is a high-speed multi-character matching circuit based on the ton method. Also,
Key expanded in different notation by the different notation expansion program 3004
The word group, along with the corresponding keyword,
Bit search on microprocessor MPU33007a
The program is passed to the program 3007. On the other hand, neighborhood conditions, context conditions, AND, O
Logical conditions such as R are obtained from the search expression analysis program 3002.
Compound condition analysis program 3041, neighborhood condition analysis program
RAM 3042, context condition analysis program 3043, logic
Compound condition judgment program via condition analysis program 3044
It is sent to the ram 3045. The required search information is
Approach program 3007, string search engine 3
006, sent to the complex condition determination program 3045
Then, the search control execution program 3008 first
Of the program 3007. Bit search
The program 3007 is stored in the semiconductor memory device 3010a.
Read the stored character component table and search for the character component table
I do. The results of the character component table search are stored in the search result storage memo.
Stored in the folder 3046. After the character component table search is completed, a search execution system
The control program 3008 has a search result storage memory 3046.
And if the search result is 0, 0 search results
Output and interrupt the search process. Search results must be 0
If so, start the string search engine 3006
Is stored in the search result storage memory 3046 at the same time.
Condensed text of documents that were hit in the results of the character component table search
Is read from the RAM disk device 2910b, and the
Sent to the search engine 3006
Let it run. As a result, the condition judgment whether the number of cases is 0
The setting is performed by the search execution control program 3008. String
Search engine 3006, the RAM disk device 3
Split search term for condensed text read from 010b
Search by. The comparison result is a compound condition determination program 30
45. Complex condition determination program 3045
Now, determine the logical condition given between the search terms,
The search result storage memory 30 stores the document number of the document that matches the search condition.
The data is sequentially stored in 46. After the condensed text search is completed, a search execution system
The control program 3008 stores the search result storage memory once again.
3046, if the number of results is 0, 0
Output as a search result and end the search. If not 0
The neighborhood, context conditions are set, or split
Search only if the search term is different from the search term
Read the search result document number from the search result storage memory,
The corresponding text is stored in the collective magnetic disk drive 3010.
read from c and to string search engine 3006
Send it, and execute a text search this time. Neighborhood, context conditions
Is not set and the split search term is
If they are equal, they are stored in the search result storage memory
Outputs the number of search results and ends the search. In the string search engine 3006,
Book read from collective magnetic disk drive 3010c
Scan the sentence and perform a text search. The result is a complex conditional
It is sequentially sent to the fixed program 3045. Compound condition judgment
In the program 3045, the logic assigned between the search terms
A sentence that determines the condition, neighborhood and context conditions, and matches the condition
The document numbers of the documents are sequentially stored in the search result storage memory 3046.
I do. If the search is performed up to the text search,
After completion, the search execution control program 3008
The number of search results is output with reference to the delivery memory 3046 for searching.
To end. In this way, large volume text data is
A small character composition table and condensed text
By storing in semiconductor memory or RAM disk,
Fast full-text support for large databases
Can be realized. Next, the condensed text is stored in the magnetic disk.
Thirteenth embodiments will be described. Condensed text on magnetic disc
When storing in a hierarchical
By optimizing, normal hierarchy using the same configuration
Can be faster than performing type pre-search.
Wear. Hereinafter, the procedure of this control will be described. Magnetic de
Disks usually have a magnetic head that moves mechanically.
For this reason, the information on the disk is read out at intervals (scans).
Rather than collective information)
And read (called sequential access) is faster
There is a feature that. Now, read skip access
Speed Vskip MB / s, sequential access
Assuming that the data read speed is Vseq MB / s,
The total number of documents in the database is Na,
If the number of results is Nc and the document size is
Nc> (Vskip / Vseq) · Na …………………………………………………………………………………………… (12-1)
The person who searches is based on the results of the character component table search.
The processing time is shorter than with the Kip access. But
Therefore, after the character component table search as shown in FIG.
Determine the number of results in the pre-search control program,
When the number of hits satisfying the expression (12-1) is reached,
Condensed text data is ignored, ignoring the results of the character component table search
Search for all bases. Using the above method,
Large amount of RAM to store condensed text on disk
Eliminates the need for discs and provides relatively fast full-text
To be able to implement strike search with a low-cost document search device
Become. Next, the condensed text is stored in the magnetic disk.
Fourteenth embodiments will be described. Specify neighborhood and context conditions
The search results are very low.
If there is no search, the text component table
Searching the text directly based on the search results
Be shorter. Now, the search speed of the condensed text is Vsr MB /
s, the search speed of the text is VtxMB / s,
Nc is the number of results in the distribution table, and Ns is the number of results in the condensed text.
r, the data volume per case of condensed text Qsr, book
Assuming that the data capacity per sentence is Qtx, when NcQsr / Vsr + NsrQtx / Vtx> NcQtx / Vtx... (13-1), the text search is directly performed without performing the condensation text search.
Search time is shorter. Nsr writes the condensed text
I don't know until I actually search,
To determine whether to perform a condensed text search.
You. For example, let Na be the number of documents in the entire database.
Nsr = αNa (0 <α <1)... (13-2), by transforming equation (13-1), Nc <αNa (Qtx / Vtx) / (Qtx / Vtx−Qsr / Vsr) In the case of Expression (13-3), the text search is directly performed. α threshold
Set a value in advance as a value before searching, and
After the table search, the condensed text search is performed using equation (13-3).
Decide whether to do it. By performing this control,
High-speed full-text search under specified context conditions
Can be realized. The cheap version of the twelfth embodiment is described above.
Tenth to realize full-text search with system configuration
Third and Fourteenth Embodiments have been described. In addition, use the condensed text at all.
Omitting the step of condensed text search
Control by executing a text search directly from the
Layered presearch can be realized. By this method
Search time is longer because the amount of text to be scanned increases.
Small, but not expensive RAM disks
And there is no need for magnetic disk capacity to store the condensed text.
Therefore, a lower-priced document search device can be realized.
And Also, directly use RAM data without using a character component table.
Search all texts on disk or magnetic disk
Search for positional relationships between search terms such as neighborhood and context conditions.
Control method to search text only when search condition is specified
Therefore, the hierarchical pre-search can be realized. This
Method increases the amount of search for condensed text,
It takes some time, but semiconductor memory to store character composition table
Is not necessary, so a low-priced document search device is realized.
You can do it. Alternatively, the video used in the previous embodiments may be used.
As shown in FIG.
A form in which characters appearing in the book are arranged in a row, that is, one character
The character code itself is not represented as one bit.
A character component table stored as a body can also be used.
Alternatively, at this time, the fifth embodiment, the sixth embodiment, and
One sentence using the hash function described in the seventh embodiment
Character entry table capacity by associating multiple characters with character entries
Can also be reduced. Store character code like this
Character component table search using the extracted character component table
Data from a file one character at a time, similar to text and text search
To determine whether the corresponding character exists
Can be realized. Thus, the sentence used in the text
By using a character component table that collects only characters,
Condensed text, which can simplify the structure and without bit operation,
The same scan type search as the text search can be used.
The effect is that the search processing method can be simplified.
Can be Further, a character component table is also stored on the magnetic disk.
Can achieve a hierarchical pre-search
You. When a character component table is stored on this magnetic disk
Is used in the search term in the character component table search.
Bit list of characters
While performing bit operation processing. Or on
If the character code of
While reading the character component table sequentially,
Out documents. This character composition table is stored on a magnetic disk.
According to the method of delivery, it is not necessary to use semiconductor memory
In addition, it is possible to realize a lower-priced document search device
It becomes. According to the present invention, a character component table and a condensed book
Input hierarchically at the character level and word level using sentences
To filter out documents that are not relevant to the search term
Since the useless text search can be omitted,
Is a means of realizing high-speed full-text search
Full response with practical response speed even for a simple document database
It becomes possible to do a kist search.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing a configuration of a first exemplary embodiment of the present invention. FIG. 2 is a diagram showing a registration process for hierarchical presearch, which is a feature of the present invention. FIG. 3 is a diagram illustrating a search process of a hierarchical presearch, which is a feature of the present invention. FIG. 4 is a diagram showing an example of creating a condensed text. FIG. 5 is a diagram showing a storage form of a condensed text. FIG. 6 is a diagram showing an outline of a character component table. FIG. 7 is a diagram showing an outline of a character component table search. FIG. 8 is a diagram showing a processing procedure of a hierarchical pre-search. FIG. 9 is a diagram showing a character component table search process according to the third embodiment. FIG. 10 is a PAD diagram showing a code conversion process of a character component table used in the fourth embodiment. FIG. 11 is a diagram illustrating an outline of code conversion of a character component table used in the fourth embodiment. FIG. 12 is a diagram showing an outline of a character component table used in a fourth embodiment. FIG. 13 is a diagram showing an outline of a character component table used in the fifth embodiment. FIG. 14 is a diagram showing a processing procedure of a hierarchical presearch used in the fifth embodiment. FIG. 15 is a diagram showing an outline of a character component table used in the sixth embodiment. FIG. 16 is a diagram showing a processing procedure of a hierarchical presearch used in the sixth embodiment. FIG. 17 is a diagram showing an outline of a method of creating a character component table used in the seventh embodiment. FIG. 18 is a diagram showing an outline of a character code-entry number correspondence table used in a hash function for a character component table used in the seventh embodiment. FIG. 19 is a diagram showing a method of creating a condensed text used in the eighth embodiment. FIG. 20 is a diagram illustrating a method of processing a hiragana character string for a condensed text used in the eighth embodiment. FIG. 21 is a diagram showing a basic word dictionary for attached word analysis used in the eighth embodiment. FIG. 22 is a diagram showing connection rules for attached word analysis used in the eighth embodiment. FIG. 23 is a diagram showing a processing procedure of a hierarchical presearch used in the eighth embodiment. FIG. 24 is a diagram showing a method of creating a condensed text used in the ninth embodiment. FIG. 25 is a diagram showing a processing procedure of hierarchical presearch used in the ninth embodiment. FIG. 26 is a diagram illustrating a method of creating a condensed text used in the tenth embodiment. FIG. 27 is a diagram illustrating a processing procedure of hierarchical presearch used in the tenth embodiment. FIG. 28 is a diagram illustrating a method of creating a condensed text used in the eleventh embodiment. FIG. 29 is a diagram illustrating a processing procedure of hierarchical presearch used in the eleventh embodiment. FIG. 30 is a diagram showing a part of the configuration of a twelfth embodiment. FIG. 31 is a diagram showing the remaining part of the configuration of the twelfth embodiment. FIG. 32 is a diagram illustrating a processing procedure of hierarchical presearch used in the twelfth embodiment. FIG. 33 is a diagram showing an outline of a character component table stored as characters.

   ────────────────────────────────────────────────── ─── Continuation of front page    (72) Inventor Kanji Kato               1-280 Higashi Koigakubo, Kokubunji-shi, Tokyo                 Central Research Laboratory, Hitachi, Ltd. (72) Inventor Hisamitsu Kawaguchi               1-280 Higashi Koigakubo, Kokubunji-shi, Tokyo                 Central Research Laboratory, Hitachi, Ltd. (72) Inventor Naoki Minegishi               890-12 Kashimada, Saiwai-ku, Kawasaki-shi, Kanagawa                 Hitachi, Ltd. Information System Engineer               Inside the hall                (56) References JP-A-59-112339 (JP, A)                 Kanji Kato et al. "Text Search for Full Text Search               Development of Chimashin "Research of IEICE               Report (DE-89-38) Vol. 89 N               o. 335 (Heisei 1-12-14) P. 17−24

Claims (1)

(57) [Claims] [Claim 1] Full text for searching a document including a keyword specified by a searcher with reference to the body content of a document database storing document information as character code data In a search method, when registering a document in the document database, a body character string of the registered document is divided for each character type such as hiragana, kanji, and alphanumeric characters, and a partial character string for each character type is divided. Investigate the mutual inclusion of character strings between each other, create a condensed text consisting of a set of partial character strings excluding character strings included in other character strings, and a character component table that registers the characters that appear in the text, At the time of retrieval, the character component table is used to examine candidates for documents that include all the characters used in the keyword, and the condensed text of the resulting candidate document is searched for the part that is divided into character types In a full-text search method using a hierarchical pre-search method in which a document including all character strings is searched and a document including a keyword is searched by searching the body of a candidate document obtained as a result, A full-text search method, wherein a plurality of characters are associated with entries in the character component table using a correspondence table that associates a preset character code with an entry in the character component table.