JPH08329116A - Method for retrieving structured document - Google Patents

Abstract

PURPOSE: To provide a character component table capable of reducing retrieval noise by practical capacity and to attain efficient document structure specification/retrieval in a large scale information retrieving system for a structured document. CONSTITUTION: A character component table describing the character appearance state of text data is prepared in each document to be registered, the structure of the document is recognized in accordance with a previously determined document structure name and text data are divided in each structure. In each appearing character, '1' is set up in a specific bit position corresponding to document structure indicating the appearance of the character, a structure bit string describing the appearance document structure position of each character is stored, and when a user specifies 'limit work' as a retrieving character string and 'the title invention', 'the claim', or 'merits', the character component table is searched based upon the 'limit work', documents 1, 7, 15, 38,... are obtained as the searched result and bit AND between a specified document structure bit string '100100001' based upon the specified document structure and the structure bit string of the retrieved document is found out to obtain documents 1, 7, 38,... as retrieved results.

Description

Detailed Description of the Invention

[0001]

This invention relates to a document database,
It relates to a document search method that searches only a user-specified document structure part in a document from a search string specified by a user and searches for a desired document. Particularly, a large number of documents are registered and high-speed search is performed. The information retrieval method suitable for the case is applied to a large-scale document database.

[0002]

2. Description of the Related Art First, a full-text search method that does not require adding a keyword when registering a document is described in "Patent Document 1".
3-174652 ”. This method equivalently increases the search speed by sieving out documents that are not related to the search word using a condensed text in which the document is compressed in word units and a character component table in which the characters used in the document are registered in single character units. The purpose is to perform full-text search at high speed at a practical level. In addition, a concatenated character component table system for improving this character component table and realizing a faster full-text search was proposed in "Japanese Patent Laid-Open No. 05-174064." The concatenated character component table used in this publicly known example extracts all concatenated character strings of a predetermined length contained in text without duplication, and corresponds to a number that identifies a document as identifier information of a document including these. It is described by a bit string in which the bit position to be set is 1. However, if the identifier information for all concatenated characters is described in bit strings, bit strings are required for the number of character combinations, and the concatenated character component table becomes enormous in capacity. Therefore, in this known example, a plurality of concatenated characters are assigned to one bit string by using a hash function so as to reduce the capacity.

[0003]

However, in the case where a plurality of concatenated characters are assigned to one bit string, that is, the document identifier information storing the document number in which the concatenated character appears, using the conventional hash function, , Document identifier information of completely different concatenated characters is also superimposed on the same bit string. Therefore, when a certain concatenated character is specified and the document identifier information is extracted from the corresponding bit string, a document including a completely different concatenated character may be obtained from the information. That is, the search result by the concatenated character component table using the hash function includes search noise. This means that in a large-scale document search system that registers a large number of documents, unnecessary documents that may not contain the search string may not be properly screened, that is, narrowed down. In that case, search performance is degraded.

It is possible to use one bit string for each concatenated character without using a hash function, but in that case, the data amount of the bit string becomes enormous, which is not practical. More specifically, there are currently about 8,000 character codes used in Japanese, so the type of concatenated character as a combination of two characters is 8,000 × 8,000 = 64 million types. Become. Assuming that the number of documents to be registered is 1 million, 1 million bi will be added to each of these 64 million types of concatenated characters.
Since the document identifier information of t is made to correspond as a bit string, a capacity of 64 million types × 1 million bits = 8 TBytes is required. Even if the size of the document body is 20 KB / case, the size of the character component table is 100
In all cases, 20 KB x 1 million = 20 GBytes, and the capacity of the character component table is overwhelmingly larger. In order to reduce the capacity of this character composition table,
In addition to storing the document identifier information in which a corresponding character appears in a fixed-length bit string, it is also possible to directly write the document number when the number of applicable documents is small. This is called an ID list storage format. Further, the conventional format for storing the document identifier information in the bit string is called a bit list storage format.
For example, to store the document identifier information in which each character appears in a bit list storage format in a database that stores 1 million items, the capacity of 1 million bits = 125 KB per character even if there is only one document that appears. Is always required, but in the ID list storage format that stores the document identifier information that appears in the document number, the document ID
4B and the number of stored document IDs is 4B, the capacity of 4B + 4B = 8B is sufficient.

On the other hand, in the document, as in the example of the patent publication,
It has a structure like "name of invention", "inventor", "applicant", "claim", "problem to be solved by the invention", and specific contents are contained in each structure. In many cases. Performing a full-text search on such a structured document by designating the document structure of the search target is called a structure-designated search. In order to realize this structure designation search, it is necessary to create a character component table for each structure and perform a search using a different character component table for each structure at the stage of character component table search. However, when a character component table is created for each structure of a document, there is a problem that the capacity becomes large because each structure unit has document identifier information with the same character component. For example, "Invention name", "Inventor",
When there are ten types of document structures such as “applicant” and “claim”, if a character composition table is created for each structure of the document, the capacity of the character composition table will be ten times as large. This offsets the effect of reducing the capacity of the character component table by storing the document identifier information in the ID list format. Further, when searching for a plurality of document structures, it is necessary to repeat the search of the character component table for the number of times, which causes a problem that the number of file input / output increases and it is not efficient. For example, "the title of the invention",
In the case of the condition "search for a document that has the character string" extreme work "in the document structure of claims or effect", after searching the character component table corresponding to each of the three types of structures, It is necessary to perform their OR operation. An object of the present invention is to provide a character component table with less search noise with a practical capacity and to realize an efficient document structure designation search in a large-scale information search system for storing documents having a structure. is there.

[0006]

To achieve the above object, according to the present invention, a document having a document structure is stored, a user designates a document structure name to be searched and a search character string, and the corresponding document is searched. In the document search system for searching, a step of creating a character component table that describes the appearance status of characters in the text data of the document for each registered document, and according to a predetermined document structure name for each registered document Recognizing the document structure, dividing the text data for each structure, and for each registered document, setting a specific bit value at a specific bit position corresponding to the document structure in which each character appears Then, the step of storing the structure bit string describing the appearance document structure position for each character and the document structure to be searched by the user. Inputting a name and a search character string; searching for a document in which all of the character components forming the search character string exist from the character component table for the search character string provided by the user; For each searched document, reading the structure bit string corresponding to each character of the search character string, and extracting the document in which the bit position of the document structure specified by the user is a specific bit value, Documents that include the search character string in the document structure specified by the user are searched. Further, a correspondence table including records that associate each name of the document structure with the bit position of the structure bit string is provided, and the document structure name and the bit position of the structure bit string are associated with each other based on the correspondence table. Further, the correspondence table is made up of records each consisting of a name of the document structure, a bit position of the structure bit string, and a structure identification tag which is a special character string indicating each name of the document structure. The structure identification tag corresponds to the text data. The structure identification tag is inserted into the document structure, and the inserted text data is stored. Further, based on the document structure name as the search target input by the user, a specified document structure bit string having a bit position corresponding to the document structure name as the search target of the structure bit string as a specific bit value is created, and the search is performed. A regarding the read structure bit string corresponding to each character of the character string and the bit value of each corresponding bit position of the specified document structure bit string A
An ND operation is performed, and an AND or OR condition between a plurality of document structure names designated as a search condition is processed based on the result of the operation. Furthermore, a document identifier information file that stores the document identifier information of the character component table and a structure bit string storage file that stores the structure bit string are created separately, and pointer information to the structure bit string storage file is added to each record of the document identifier information file. I am trying to store it. Furthermore, when the user inputs a document structure name to be searched, the read structure bit string corresponding to each character of the search character string and the bit value of each corresponding bit position of the specified document structure bit string AND operation is performed between a plurality of document structure names designated as search conditions based on the result of the operation, and when the user does not input the document structure name to be searched, Referring only to the character composition table,
The structure bit string is not read.

[0007]

Since the structure bit string is stored by the above means, only the document including the search character string in the search target document structure specified by the user can be searched by a simple process.
In particular, when there are a plurality of search target document structures specified by the user, the condition determination can be performed only by the bit AND process of the corresponding bit positions of the stored structure bit string and the specified document structure bit string, so high-speed search processing should be performed. You can In addition, in order to perform a structure-specified search, it was necessary to have a character component table for each structure in the past. By conducting an in-depth search, the character component table can be made single to save the capacity.

[0008]

EXAMPLES Examples of the present invention will be described in detail below. FIG. 1 is a diagram showing the configuration of this embodiment. In this embodiment, the terminals 101, 102 ,. ． ． 110,
It consists of a network 200 and a document server 1000. Document server 1000 includes LAN adapter 1010 and CP
U1020, work memory 1030, character table 11
00, file pointer table 1110, document structure identification tag correspondence table 1200, memory 1050, character component table creation program 1310, structure recognition program 1
320, structure bit string storage program 1330, search condition input program 1340, character component table search program 1350, memory 1300 storing structure bit string AND program 1360, files 1401, 1402 ,. ． ． , Files 1411, 1412 ,. ． ． ． , Text data 1420.

First, an outline of a structure designation search using a structure bit string will be described. FIG. 2 shows an outline of a structure designation search method using a character component table in which two concatenated characters are character components and a structure bit string. In this figure, as the condition specified by the user, "any of" invention name "," claim "and" effect "" is specified in the search target document structure, and "extreme work" is specified as the search character string. It shows the situation. First, as the first step of the search, a document containing the search character string "extreme work" is searched using the character component table. In the example of the figure, the character component table stores the identifier information of a document including each character component in the form of an ID list, with two concatenated characters as the character component. In the character component table search, a search for a document that includes all three character components of the search character string "extreme work", two concatenated characters "" extreme "," limit work ", and" work "" is performed for each character component. This is done by taking the product set of the document identifier information corresponding to. In the example of FIG. 2, the document ID string “1, 7, 15, 38, ...” Is shown as the search result of the character component table thus obtained. The second step of the search is a designated document structure bit string in which the position of the structure bit string corresponding to the document structure to be searched is 1 (“100100001” in FIG. 2, and the first “1” is the invention name 2). The second "1" indicates the scope of claims and the last "1" indicates that the effect is designated) and the structured bit string corresponding to the character component of the search character string is subjected to bit AND processing. As shown in the figure, the structural bit string is stored such that, for each character component, the structural bit string of the document in which the character component appears is aligned. From this arrangement of structural bit strings, the structural bit string corresponding to the document searched in the character component table is read, and the bit AND process is performed with the specified document structural bit string,
The document in which the result is non-zero for all the character components forming the search character string is the final search result. In the example of FIG.
Since the bit AND process of the structured bit string corresponding to the document number 15 of the character component "Limited" and the specified document structured bit string is "0", the document number 15 is omitted from the search result of the character component table, and the character component "extreme limit" is obtained. Since the bit AND processing of the structure bit string corresponding to the document number 1 of "," Limited work ", and" Work "and the specified document structure bit string is" 1 "in the document structure of" Invention title "and" Effect ". The document number 1 is the search result, and the document number 7 is "1" in the "claim", so the document number 7 is the search result. In this way, the document ID string “1, 7, 38, ...” Further narrowed down from the search result by the character component table is obtained as the final result.

The outline of the structure designation search has been described above. Next, the structure of the character component table and the structure bit string used in this embodiment will be described. In this embodiment, a character table and a file pointer table are used to manage the document identifier information corresponding to the concatenated character. FIG. 3 is a diagram showing an outline of the character table and the file pointer table. For example, when retrieving a document including a character string “structure”, first, the record corresponding to the character “structure” in the character table is referred to obtain pointer information 580 to the file pointer table. Next, each record from the 580th byte from the head of the file pointer table is referenced to search for a record in which the second character is "success". In the file pointer table, a record in which the second character is 0 is stored at the beginning for each first character of each concatenated character. In the record in which the second character is 0, pointers to the document identifier information of all the documents including the first character of the first character are stored. That is, in the record in which the second character is 0, a file identifier (hereinafter also referred to as a file ID) and a byte position in the file (hereinafter also referred to as an offset) for accessing the document identifier information corresponding to a single character including only the first character To store. Therefore, since there is always a record with the second character being 0 for each concatenated character, for example, when searching for a concatenated character of “composition”, the 580th byte from the beginning of the file pointer table corresponding to “structure” The search starts from the record
Continue the search until the second character becomes 0 again,
If the character is not found, it can be determined that there is no corresponding concatenated character. In the example of FIG. 3, since there is a record of “completion”, the file ID is 1 and the offset is 1 from here.
Information for accessing the document identifier information 034 can be obtained.

The document identifier information is divided and stored in a plurality of files as shown in FIG. The file ID information of the file pointer table identifies which file stores the document identifier information. Further, it is predetermined that the specific file ID has the document identifier information in the bit list format. In the example of FIG. 4, the file 1 is a file having document identifier information in the bit list format. Further, the offset information of the file that stores the structural bit string is stored at the beginning of each document identifier information. In the example of FIG. 3, the file ID of 1 and the offset of 1,034 are obtained as access information to the document identifier information regarding the concatenated character “composition”. Therefore, by reading the data from the 1,034th byte in the file 1, the offset information 6,734 of the file storing the structural bit string and the bit string “0111” indicating the document identifier information are read.
.. ”is obtained. This bit string is“ 1 ”corresponding to the document number from the first bit.
Indicates a document containing the concatenation character "composition". Therefore, in this example, the document numbers of the documents including the "configuration", that is, the ID lists are 1, 2, 3, 5, 7, 9, ...
Can be converted mechanically. The other files (file 2 and file 3) in FIG. 4 store the document identifier information in the ID list format. The beginning of each ID list is
Similar to the file stored in the bit list format, this is the offset information of the file storing the structural bit string.
Also, the head of the ID list linked to the offset information indicates the number of stored document numbers. For example, in the case of the concatenation character “structure”, in the example of FIG. 3, the file ID is 2,
Since the offset is 340, by referring to the 340th byte from the beginning of the file 2, there are 56 documents including the offset information 684 and the concatenation character “structure”, and the document numbers are 562, 1038, ... ID list information is acquired.

Similarly to the document identifier information, the structural bit string is divided into a plurality of files according to the file ID stored in the file pointer table and stored. FIG. 5 shows a state of a storage file of a structural bit string. For example, in the case of the concatenation character “structure”, in the example of FIG. 3, the file ID is 2 and the offset is 340. Therefore, by referring to the 340th byte from the beginning of the document identifier information storage file 2, the offset of the structure bit string Information 684 is obtained. Therefore, by referring to the 684th byte from the beginning of the storage file 2 of the structure bit string, the structure bit string "0100111001111" indicating the position of the document structure including the corresponding character component "structure".
111 "is obtained. This structured bit string is arranged in order from the position of the offset information in the document identifier information storage file by the number of documents in which the character component appears. That is, the structured bit string storage file 2 The structure bit string from is the structure bit string of the document 562, that is, the structure bit string of the document 562 for the number of documents including the concatenation character “structure”, that is, 56
Are arranged in order according to the document ID in the structure bit string and the document identifier information. In the present embodiment, 16 bits are assigned to each structure bit string so that 16 document structures can be managed for one document. However, by increasing the number of bits, it is possible to manage 16 or more document structures. It's easy to do.

Thus, since only concatenated characters existing in the database are registered in the file pointer table, there is an advantage that all combinations of characters not existing in the database can be eliminated. Therefore, it is possible to significantly reduce the file amount and the memory amount for storing the management information of concatenated characters realized by the character table and the file pointer table. Also, the document identifier information is stored in the form of a bit string or an ID list, and is managed as a bit string when storing a large number of documents, and is managed as an ID list when storing a small number of documents, thereby significantly reducing the file size. be able to. Specifically,
To store the document identifier information in the bit list format, the number of bits for all cases registered in the database is always required. However, when storing the document identifier information in the ID list format, the bits representing the document identifier are stored. The number x the number of registered documents will suffice. For example, the total number of registrations in the database is 1
Assuming that 32 bits are allocated to represent one document identifier information in one million documents, if 10 documents including the concatenation character “structure” are registered, in the case of a bit list format, 1 million bits = 125 KB A storage area is required, but in the case of the ID list format, a storage area of 32 bits x 10 cases = 40B will suffice. On the other hand, for example, if there are 900,000 documents containing the concatenation character “composition” in 1 million, the bit list format can be stored in the storage area of 1 million bits = 125 KB, whereas the ID list format can be used. In this case, an area of 32 bits × 900,000 = 3.6 MB is required. Therefore, this 1 million
When the document identifier is stored in 32 bits, the document identifier information is stored in a bit list format when the number of registrations is larger than this, and in the ID list format when the number of registrations is smaller than 1 million bits / 32 bits = 31,250 cases. Storing is the most effective way to use the storage area.

Further, by storing the structural bit string in association with each character component, it is not necessary to prepare a plurality of character component tables each including a character table, a file pointer table and a document identifier information file for each document structure. There is. This has a great effect on reducing the number of files in the database.

The outline of the structure designation search and the structures of the character component table and the structure bit string have been described above. Now, a method of registering document data will be described. FIG.
It is a PAD figure which shows the flow of registration. When registering data, the character component table creating program 1310 registers character components in the character table and the file pointer table as needed for each character component of the registered document, and stores the document ID in the document identifier information file for each character component. (6
020). Further, the structure recognition program 1320 divides the text data for each structure of each document (6030), and the structure bit string storage program 1330 creates and stores a structure bit string for each character used in each structure (6040).

In the character component table registration step (6020), for each character component used in the document, the character table and the file pointer table are referenced to check whether or not the character component is registered. In this case, the character components are registered in the character table and the file pointer table. When the corresponding character is not registered in this character table or file pointer table, an empty area of the file storing the document identifier information is secured, and the file ID and offset value are stored in the file pointer table. In this way, for each character component in the document, the file ID that stores the document identifier information and the offset value are acquired from the file pointer table, and the document ID is stored in the file that stores the corresponding document identifier information.

In the structure dividing step (6030), the text data between the identification tags is extracted according to the document structure identification tag correspondence table shown in FIG. For example, as shown in FIG. 8, text data is converted into “# BIJ-Title”,
It is separated by an identification tag such as "# BIJ-Inventor" and sent as text data of the document structure of "invention name" and "inventor" to the next structure bit string storing step. The document structure identification tag correspondence table also stores the bit position corresponding to each document structure of the structure bit string used in the structure bit string storing step of the next processing step.

In the structure bit string storing step (6040), a structure bit string is created and stored for each extracted text data of each document structure in each character unit used therein. In the storage of this structure bit string, the data of the character component existing in the text data of each document structure is registered at the bit position stored in the document structure identification tag correspondence table shown in FIG. For example, in the example of FIG. 8, document 1
Since there is a character component "extreme" in the document structure "Invention name" of, the first bit of the structure bit string representing "Invention name" is set to 1 as in the example shown in FIG. The storage position of the structural bit list corresponding to each character component is determined by referring to the character table, the file pointer table and the document identifier information file, as already described in this embodiment. For example, in the case of the character component "extreme", the character table is referred to by the first character "extreme" to obtain the offset value 870 to the file pointer table. Next, a file ID for accessing the document identifier information storage file is searched for a record in which the second character is “limit” from the 870th byte from the beginning of the file pointer table.
And get the offset value. In this way, the document identifier information corresponding to the character component "extreme" is read from the offset 1084 with the file ID of 3, and the offset value 8692 of the file storing the structural bit string of the first 4 bytes is obtained. In this embodiment, a 16-bit structured bit string is stored for each document including the character component "extreme" from this offset value 8692. Therefore, the structural bit string whose document number is 1 has the structural bit string storage file ID 3
It can be seen that 16 bits correspond from the 8692th byte from the beginning. In this way, the character table, the file pointer table, and the document identifier information are stored for the character components of each document to be registered, and for each character component of each document, the position of the document structure in which the character component exists is determined by the structure bit string. Will be stored as.

The search process is performed according to the procedure shown in FIG. First, the character component is cut out from the search word (9010). Next, regarding each extracted character component (902
0), the character table is searched (9030). And
For each record in the corresponding file pointer table, the second character is searched (9040) to obtain the corresponding file ID and offset value. In this way, the file storing the document identifier information and the document identifier information corresponding to each concatenated character corresponding to the offset value are acquired (9050). If the corresponding concatenated character is not registered in the character component table in the process of acquiring this document identifier information (9
(060, 9070), that is, if any one of the character components forming the search word is not registered in the character component table, there is no corresponding document including the search word, and the result of the search result is 0. The document identifier information search program 1350 returns it to the search terminal via the LAN adapter 1010.

When the corresponding document identifier information is obtained for all the character components forming the search word, the structural bit string corresponding to each character component is read next. This reading can be performed from the file ID obtained from the file pointer table, the offset value of the structured bit string storage file obtained from the document identifier information, and the number of documents including the character component. In this way, the structure bit string is read for all the character components, and only the document having the bit position 1 indicating the document structure designated as the search condition is extracted (9080). For example, when searching for a document that includes the character string "extreme work" in "invention title",
For each of the character components “extreme”, “limited work”, and “work” of the search character string, the structural bit strings for the number of cases described in the document identifier information are read, and “name of invention” is read.
Only the document corresponding to the bit position corresponding to the above, that is, the first bit is 1, is extracted. At this time, if there is no corresponding document for all the character components forming the search character string, it can be determined that the number of documents including the search character string in the specified document structure is 0 (9090). In all other cases, with respect to all the documents including the character component of the search character string in the designated portion thus obtained, the document including all the character components is set as the search result (9100). This is the product intersection of the obtained documents including the respective character components (for example, in the case of FIG. 2,
It is performed by taking a product set of document IDs 1, 7, 38, ...

In reading the structure bit string and determining whether the designated document structure includes a character component, it is possible to make a determination for a plurality of document structures at once.
When each document structure corresponds to the bit position shown in the document structure identification tag correspondence table of FIG. 7, if a document including a search character string in the “invention title” or the “claim” is searched, , A specified document structure bit string in which the bit positions corresponding to these structures are set to "1001000000000"
000 "and the structural bit string are bit-ANDed, and the document whose result is non-zero may be used as the extraction result. Further, the document including the search character string in the" invention name "and the" claim "is searched. In this case, a document whose bit AND result of the designated document structure bit string and the structured bit string is equal to the designated document structure bit string is set as the extraction result.

Since the search result of the character component table thus obtained has very little search noise, it can be sufficiently used even if the search result of the character component table is displayed. In the above explanation, the search in which the document structure name is specified has been described.
If no document structure name is specified, only the document identifier file is referenced and the structure bit string storage file is not accessed.

Of course, based on the search result of the character component table, the text of the document may be searched and narrowed down to only the documents actually including the search word, or the document satisfying the positional relationship between the plurality of search words may be searched. It is possible. Further, the search result of the character component table may be displayed once on the search terminal, and whether or not to search the text may be determined by the user's designation.

Further, in the character component table used in the present embodiment, two concatenated characters are character components, but it is also easy to implement the present invention with three or more concatenated characters as character components. Constructing a character component table with three or more concatenated characters as character components is effective in improving the search accuracy of the character component table search. In particular, this is effective when searching for a search character string composed of a small number of character types such as katakana. Similarly, in order to improve the accuracy of the character component table search, it is possible to use a combination of two characters for every one character as the character component. Such a combination of a few characters, that is, a skip character string as a character component is suitable for searching an English word or the like having a strong correlation between front and rear characters. Furthermore, both normal concatenated characters and this skip character string can be used together. In this case, the file capacity of the character component table is required rather than the normal concatenated character as the character component, but the character component table search with a smaller file capacity and higher accuracy than the case of taking three concatenated characters. Can be realized.

As described above, according to the present embodiment, there is an advantage that the condition determination designating the document structure can be performed at high speed only by the bit AND processing of the structure bit string and the designated document structure bit string. In addition, the structure bit string is stored separately from the conventional character component table in order to perform the structure-specified search, so that the capacity of the character component table is minimized, and the conventional high-speed processing is maintained even for ordinary search processing that does not specify the structure. It becomes possible to add the structure designation search function.

[0026]

According to the present invention, by storing the structure bit string, it is possible to search only the document including the search character string in the search target document structure designated by the user by a simple process. In particular, when there are a plurality of search target document structures specified by the user, the condition determination can be performed only by the bit AND process of the corresponding bit positions of the stored structure bit string and the specified document structure bit string, so that high-speed search processing can be performed. There are advantages. In addition, in order to perform a structure-specified search, it was necessary to have a character component table for each structure in the past. By performing an in-depth search, there is an advantage that the character component table can be made single and the file and memory capacity can be saved.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a first embodiment.

FIG. 2 is a diagram showing an outline of a structure designation search method.

FIG. 3 is a diagram showing a table configuration of a character component table.

FIG. 4 is a diagram showing an outline of a document identifier information storage file.

FIG. 5 is a diagram showing an outline of a structural bit string storage file.

FIG. 6 is a PAD diagram showing a document registration process.

FIG. 7 is a diagram showing an example of a document structure identification tag correspondence table.

FIG. 8 is a diagram showing an example of clipping text data by document structure.

FIG. 9 is a PAD diagram showing a search process.

[Explanation of symbols]

 101, 102, ..., 110 Terminal 200 Network 1000 Document Server 1010 LAN Adapter 1020 CPU 1030, 1050, 1300 Memory 1100 Character Table 1110 File Pointer Table 1200 Document Structure Identification Tag Correspondence Table 1310 Character Component Table Creation Program 1320 Structure Recognition Program 1330 Structure bit string storage program 1340 Search condition input program 1350 Character component table search program 1360 Structure bit string AND program 1401 Document identifier information file 1 1402 Document identifier information file 2 1403 Document identifier information file n 1411 Structure bit string storage file 1 1412 Structure bit string storage file 2 1413 Structure bit string storage file n 1420 Text data

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Satoshi Asagawa 5030 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Hitachi Ltd. Software Development Division

Claims (6)

[Claims] 1. A document search system that stores a document having a document structure, and a user specifies a document structure name to be searched and a search character string to search for a corresponding document Step of creating a character composition table describing the appearance status of characters in the text data, and recognizing the document structure according to a predetermined document structure name for each registered document, and dividing the text data for each structure Then, for each of the registered documents, by setting a specific bit value at a specific bit position corresponding to the document structure in which each character appears, a structure bit string describing the appearance document structure position for each character is set. The step of storing, the step of receiving the document structure name to be searched by the user, and the input of the search character string, For the search character string given by the user, a step of searching the character component table for a document in which all of the character components forming the search character string are present; The step of reading the structure bit string corresponding to each character and extracting the document in which the bit position of the document structure specified by the user has a specific bit value, and the search character string is included in the document structure specified by the user. Structured document search method characterized by searching for existing documents. 2. The structured document search method according to claim 1, further comprising a correspondence table composed of records that associate each name of the document structure with the bit position of the structure bit string, and based on the correspondence table, the document structure name and the structure bit string A structured document search method characterized by taking correspondence between bit positions. 3. The structured document search method according to claim 2, wherein the correspondence table is a structure identification tag that is a special character string indicating each name of the document structure, the bit position of the structure bit string, and each name of the document structure. A structured document search method comprising: inserting the structure identification tag into a corresponding document structure of text data, and storing the inserted text data having the structure identification tag. 4. The structured document search method according to claim 1, wherein a bit position corresponding to the search target document structure name of the structure bit string is set based on a search target document structure name input by a user. A specified document structure bit string having a specific bit value is created, and the read structure bit string corresponding to each character of the search character string and the bit value of each corresponding bit position of the specified document structure bit string are ANDed.
A structured document search method comprising performing an operation, and processing an AND or OR condition between a plurality of document structure names designated as a search condition based on the result of the operation. 5. The structured document search method according to claim 1, wherein a document identifier information file storing document identifier information of a character component table and a structure bit string storage file storing a structure bit string are separately created, and the document identifier is created. A structured document search method characterized by storing pointer information to a structured bit string storage file in each record of an information file. 6. The structured document search method according to claim 4, wherein when the user inputs a document structure name to be searched, a read structure bit string corresponding to each character of the search character string is stored. An AND operation is performed on the bit values at corresponding bit positions of the specified document structure bit string, and an AND or OR condition between a plurality of document structure names specified as search conditions is processed based on the result of the operation. When the document structure name to be searched is not input from, the structured document search method is characterized in that only the character component table is referenced and the structured bit string is not read.