JP2006209399A - Device and method for retrieving document - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To retrieve a document of a keyword of KANA (Japanese syllabary) and KATAKANA (square form of kana) with an N-gram index of a small size. <P>SOLUTION: A character string extracting part 11 extracts a KANA character string and a KATAKANA character string in the document. A character string connecting part 12 connects the extracted character string so that a break can be discriminated and constructs it as a pseudo document. A bit vector generating part 13 generates a bit vector showing an appearing position of each N-garm realizing a KANA word and a KATAKANA word in the pseudo document by a flag bit. An index registering part 14 registers the bit vector in an index storage part 15 as an index for retrieving the document. A keyword input part 16 inputs a keyword of the KANA character string and the KATAKANA character string. A retrieval part 17 decomposes the keyword into N-garam, discriminates whether the keyword is included in the pseudo document or not from position information in the pseudo document of N-garam, and a retrieval result output part 18 outputs a retrieval result. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a document search technique using an N-gram method, and is particularly optimal in combination with a document search of a form book analysis method.

  A morphological analysis method and an N-gram method are known as document search methods. In the morphological analysis method, a document is decomposed into morphemes by morphological analysis and registered in an index. For each morpheme word entry, the identifier of the document including the morpheme word, or the document identifier and the appearance position of the morpheme word are registered in the index. By searching the index based on the input keyword, a document including the keyword can be selected at high speed. In this morpheme analysis method, there is a problem that a document to be originally selected may not be selected due to an erroneous determination of a morpheme segmentation due to an unknown word or the like.

  On the other hand, in the N-gram method, each N-gram in the document is registered in an index. N-gram is an expression in a language model, and is obtained by mechanically cutting two adjacent characters (bi-gram) or three characters (tri-gram) from a document. Usually, an appearance position in a document and an identifier of the document are also registered in an index, and a character string search of an arbitrary length can be performed by determining adjacency from the appearance position. For example, in N-gram of N = 2, the word “character string search processing” is decomposed every two characters and the appearance position is stored. In this case, “(1) Character”, “(2) Character String”, “(3) Column”, “(4) Search”, “(5) Search”, “(6) Search”, “(7) Processing” is decomposed and registered in the index (the number in parentheses is the appearance position). Here, when searching for the word “search”, a document including the N-gram “search” may be searched from the index. In the case of “search processing”, a document including both “search” and “processing” is searched, and the appearance position of “processing” is two more than the appearance position of “search” (in this case, ( Find 5) and (7)).

  However, in the search using the N-gram method, the word boundary is ignored, and the document is searched by simply matching the character strings. Therefore, the document that should not be hit is searched, and the original document is not searched. It may be buried in non-conforming documents. Also, the index size tends to be enlarged depending on the appearance position.

  The following documents are related to the present invention.

  Patent Document 1 assumes that the morpheme word set is the most compact in order to reduce the index of the morpheme analysis method (morpheme words including other morpheme words (extended words) are not included in the dictionary), and then compact. If a morpheme word included in the morpheme word set is used as an entry and a dictionary (extended word dictionary) that associates the morpheme word including the morpheme word as a part is used, and the input keyword is not included in the compact morpheme word set, A search without omission is performed using a constituent word and an extension word dictionary, and further, a search mode is limited to only a compact morpheme word set, or an omission search using an extension word dictionary is performed, Switching can be performed.

  Patent Document 2 cuts out a character string based on a character type, hiragana and katakana are used as feature words as they are, N-gram is extracted as a feature word for kanji, and a similar document search is performed based on the frequency information of the feature word. Is disclosed.

Patent Document 3 adds information on the word start position and word end position obtained from the morphological analysis result to the N-gram index so that a search such as a forward match or a backward match can be performed with the intention of a word boundary. Is disclosed.
Japanese Patent Laid-Open No. 11-73429 JP-A-11-143902 JP 2000-23156 A

  The present invention has been made in consideration of the above circumstances, and an object thereof is to make it possible to easily perform an N-gram search while suppressing an index size. Further, in a specific aspect, an object is to provide an analysis technique using an optimal N-gram technique in combination with a search for a morphological analysis technique.

  In a specific configuration example of the present invention, while using an index of a morphological analysis method as a basic of a search, typically, an index is constructed by an N-gram method limited to hiragana and katakana, whereby a morphological analysis is performed. To compensate for the problem of missing search methods.

  In addition, the index size is reduced by converting the appearance position in the pseudo document in which different words (words that are not the same) are separated by a blank character into a bit vector, thereby speeding up the adjacent calculation.

  In this configuration example, while using the morphological analysis method, a partial match search can be performed by the N-gram method, typically limited to hiragana and katakana.

  In addition, since a pseudo document consisting only of different words is created and the appearance position in the document is held by a bit vector, the N-gram index can be reduced.

  In addition, the bit vector can be folded back at a fixed length and degenerated to the fixed length.

  Further, the bit vector size can be reduced by dividing the bit vector into a plurality of sequences and omitting a sequence having no flag bit indicating the appearance position.

  The present invention will be further described. In the following description, for ease of understanding, the reference numerals of the respective parts of the embodiments may be attached and described, but this is not intended to limit the present invention to the embodiments.

  According to one aspect of the present invention, in order to achieve the above-described object, the document search apparatus (100): from each of the search target documents, a hiragana character string, a katakana character string, an alphabet character string, and a hiragana character, Pseudo document generation means (11, 12) for extracting and concatenating at least one kind of character string selected in advance from a mixed character string of katakana and alphabet, and generating a pseudo document; For each entry of N-gram, appearance position storage means (13, 14, 15) for storing appearance position information representing the appearance position of the N-gram in the pseudo-document; hiragana character string, katakana Of the character strings, alphabetic character strings, and hiragana, katakana and alphabetic mixed character strings, It is provided and the document identification means (17) for specifying the document that contains the search term against the above occurrence position information composed search keyword by also one string.

  According to this configuration, the search processing can be performed at high speed with a small number of computer resources by reducing the N-gram index.

  In this configuration, the appearance position storage means preferably represents the appearance position information representing the appearance position of the N-gram in the pseudo document with a bit vector that sets a flag bit at the bit position corresponding to the position.

  By using a bit vector, it is possible to easily determine the adjacency relationship by a shift operation and an AND operation, and as a result, a search process can be easily performed.

  The pseudo document is preferably generated by leaving one of the plurality of identical character strings in the same document and deleting the other. In this way, the index size can be further reduced. Of course, you may leave duplicates.

  A blank character can be used to represent a delimiter between adjacent character strings in the pseudo document, but is not limited thereto.

  The at least one kind of character string selected in advance is typically a hiragana character string and a katakana character string.

  Further, when the bit length of the bit vector exceeds a predetermined length, the bit vector may be folded back at the predetermined length position to perform OR logic of the flag bit, and the bit vector may be degenerated to a predetermined length. .

  If the bit length of the bit vector exceeds a predetermined length, the bit vector is divided into the predetermined length sequences, and sequences that do not include flag bits are omitted and registered, and complemented at the time of search. May be. The bit vector having such a format may be stored in, for example, a hard disk, and may be extracted in N-gram units when necessary and expanded on a memory in a normal format.

  According to another aspect of the present invention, the document search apparatus (100): Appearance position information representing the appearance position of the N-gram for each N-gram entry for each of the documents to be searched. And an appearance position storage means (15) for storing a flag vector at a bit position corresponding to the position, and specifying a document including the search keyword by comparing the search keyword with the appearance position information Means (17); further, when the search keyword is composed of two or more N-grams, the flag bit position of each bit vector of the N-gram is in a corresponding adjacent relationship. The document including the search keyword is specified by discrimination.

  In this configuration, since the appearance position of the N-gram is expressed by a bit vector, the adjacent relationship of the N-gram can be processed by a shift operation and an AND operation.

  According to another aspect of the present invention, the document search device (200) includes: a first search means for searching a document using a search dictionary generated from a morphological analysis result; and a document using an N-gram dictionary. A second search means for performing a search; and further to the second search means: from each of the documents to be searched, a hiragana character string, a katakana character string, an alphabet character string, and a mixed character string of hiragana, katakana, and alphabet A pseudo-document generation unit that extracts and concatenates at least one character string selected in advance to generate a pseudo-document; and for each N-gram entry for each of the search target documents Appearance position storage means for storing appearance position information representing the appearance position of the N-gram in the pseudo document; hiragana character string, katakana character string, al A document including the search keyword by collating a search keyword composed of at least one type of character string selected from among the alphabet character strings and a mixed character string of hiragana, katakana and alphabet with the appearance position information And a document specifying means for specifying.

  According to this configuration, the search of the N-gram method is combined with the search of the hiragana character string, the katakana character string, the alphabet character string, by using the search of the morpheme method and the search of the N-gram method, respectively. In addition, the character string can be limited to at least one kind of character string selected in advance from a mixed character string of hiragana, katakana, and alphabet. As a result, the N-gram index can be reduced. Moreover, it is possible to reliably search for character strings that are not in morpheme words.

  The present invention can be realized not only as an apparatus or a system but also as a method. Of course, a part of the invention can be configured as software. Of course, software products used for causing a computer to execute such software are also included in the technical scope of the present invention.

  These and other aspects of the invention are set forth in the appended claims and will be described in detail below with reference to examples.

  According to the present invention, it is possible to easily perform an N-gram search while suppressing the index size.

  Examples of the present invention will be described below.

  First, the document search apparatus 100 according to the first embodiment in which the basic configuration of the present invention is implemented will be described. In this embodiment, hiragana and katakana words are received as keywords and a document is searched by the N-gram technique. The document search apparatus 100 is realized by installing software on a computer, for example, a personal computer 1000 using, for example, a recording medium 1001. As is well known, the personal computer 1000 includes a CPU, a main memory, an external memory, a bus, and various I / O devices, and the hardware resources and software resources of the personal computer 1000 cooperate with each other in the document search device 100. Each part, that is, each functional block is configured.

  FIG. 1 shows a document search apparatus 100 according to the first embodiment. In this figure, the document search apparatus 100 includes a document input unit 10, a character string extraction unit 11, a character string connection unit 12, a bit vector generation unit 13, The index registration unit 14, the index storage unit 15, the keyword input unit 16, the search unit 17, the search result output unit 18 and the like are configured. The document input unit 10 inputs a search target document (electronic data). Document input can be performed in various ways. One or a group of document files on the file management system may be designated and input, or may be input by file transfer or message transfer. The input document may be selected by its attribute, word vector, or the like. The character string extraction unit 11 extracts hiragana character strings and katakana character strings in the document. You may make it exclude the thing which comprises one morpheme word with a Chinese character etc. using morpheme information. In this example, only the hiragana and katakana character strings are targeted. -You may make it extract a katakana / alphabet mixed character string. The character string concatenation unit 12 concatenates the extracted character strings so that the delimiters can be discriminated and constitutes a pseudo document. The extracted character string may be divided into a plurality of pseudo documents for each type. Even if it is included in the original document, words that are not to be extracted can be omitted from the pseudo document and the size thereof can be made compact. The bit vector generation unit 13 generates a bit vector indicating the appearance position in each pseudo document with a flag bit for each N-garm capable of realizing a hiragana character string (Hiragana word) and a katakana character string (Katakana word). is there. The bit vector will be described later with an example. The index registration unit 14 registers the bit vector in the index storage unit 15 as a document search index. The keyword input unit 16 inputs a keyword composed of a hiragana character string or a katakana character string. The keyword may be input directly by the search user, or the search condition input by the user may be processed by a predetermined front end, and the part corresponding to the hiragana and katakana words may be input from the keyword input unit 16. . The search unit 17 decomposes the input keyword into N-garam, and determines whether the pseudo document includes the keyword from the position information in the N-garam pseudo document. This point will also be described later with an example. The search result output unit 18 outputs whether or not a keyword is included in the pseudo document, that is, in the input document. A list of applicable documents may be output.

  FIG. 2 shows the flow of index registration processing for input documents, and FIG. 6 shows the flow of search processing.

  First, an index registration process for an input document will be described with reference to FIG. Here, the document shown in FIG. 3 will be described as an example. The document in FIG. 3 is composed of sentences 1 to 5, sentence 1 has a hiragana character string “aiueo”, sentence 2 has a hiragana character string “a”, and sentence 3 The sentence 4 has the hiragana character string, the sentence 4 has the character string “Ayu”, and the sentence 5 has the hiragana character string “Eokaki”. In this example, an example of a hiragana character string is shown, but it goes without saying that a katakana character string may be included.

  An example of the index registration process in FIG. 2 is as follows.

[Step S10]: The document to be searched is input by the document input unit 10. The document is as shown in FIG.

[Step S11]: The character string extraction unit 11 extracts hiragana character strings and katakana character strings from the document. This character string can be extracted using the character type.

[Step S12]: The extracted character string is sequentially connected by the character string connecting unit 12 as shown in FIG. 4 to generate a pseudo document. Insert a delimiter between strings. In this example, a space character is used. Further, when the same character string appears (for example, “Ayan” in sentence 2 and “Ayan” in sentence 4), the connection is omitted. This is because if only one identical character string is registered, its occurrence can be searched. Of course, although the size of the pseudo document increases, the same character string may be repeatedly registered. Conversely, when a character string becomes a partial character string of another character string, concatenation may be omitted to further reduce the size of the pseudo document.

[Step S13]: Each character position of the pseudo document is represented by a bit, and a flag bit (for example, “1”) is set at the head position of each N-gram. Here, N is set to 3. In the example of FIG. 4, the “Any” N-gram is in the 0th bit and the 6th bit, so a bit vector in which flag bits (shown in black) are set in the 0th bit and the 6th bit is generated. . Similarly, for the “Iue” N-gram, a bit vector in which a flag bit is set in the second bit is generated. Similarly, a bit vector is generated for a feasible N-gram.

[Step S14]: The index registration unit 14 registers the bit vector of the input document in the index storage unit 15 as an index. In this way, for each N-gram entry, a document ID including the entry and its bit vector are generated and registered as an index record.

  An example of the search process in FIG. 6 is as follows.

[Step S20]: A keyword of a hiragana character string or a katakana character string is input by the keyword input unit 16.

[Step S21]: The search unit 17 decomposes the character string of the input keyword into N-grams. In this example, N is 3. For example, if the input keyword is “Okaki Kukeko”, it is broken down into “Okaki” and “Kukeko”.

[Step S22]: The retrieval unit 17 extracts the document ID and bit vector corresponding to the N-gram from the index storage unit 15. In this example, as shown in FIG. 7, the “Okaki” bit vector and the “Kukeko” bit vector are extracted.

[Step S23]: The search unit 17 determines the adjacent interval of the bit vectors and determines whether there is a corresponding character string “Okaki Kukeko”. In this example, as shown in FIG. 7, the bit position difference is “3”, and it is determined that “Okaki Kokeko” exists. Details of this determination processing will be described later. On the other hand, when “Ueokaki” is used as the keyword and “Ueokki” N-gram and “Kakikuku” N-gram are used, it is found that there is no adjacent relationship as shown in FIG. You can see that there is no.

  This search result is output by the search result output unit 18.

  Here, a detailed example of the N-gram adjacency determination process using a bit vector will be described.

  FIG. 9 shows a flow of a detailed processing example of the adjacency determination processing (keyword search processing), and the processing is as follows.

[Step S30]: Indexes are searched for all N-grams (also referred to as search words) constituting the keyword, and if the document ID is the same, each bit vector is extracted. Here, an index is searched for N-grams “Okaki”, “Kukeko”, and “Sashissu” for the keyword “Okaki Kokesashisu”, and a predetermined document ID as shown in FIG. A case where a bit vector is taken out will be described as an example.

[Step S31]: The i-th bit vector is shifted 3 bits to the right, ANDed with the i + 1-th bit vector, and the result is used as the i + 1-th bit vector. (The initial value of i is 0) Further, add 1 to i.

[Step S32]: It is determined whether there is a next bit vector. If there is, return to step S31 and repeat the process. If there is no next bit vector, the process proceeds to step S33.

[Step S33]: It is determined whether or not “1” stands in the bit vector. If “1” stands, there is a corresponding adjacency relationship between N-grams in the document, indicating that a search keyword exists (hit), and if there is no correspondence, it corresponds between N-grams in the document This indicates that there is no adjacent relationship and no search keyword exists.

  This example is further shown in FIG. As is apparent from this figure, it can be easily processed by a bit vector shift operation and an AND operation.

  Next, a modification of the first embodiment will be described.

  FIG. 12 suppresses the size of the bit vector. In the example of FIG. 12, a bit vector exceeding 1000 bits is folded every 1000 bits to be reduced to a 1000-bit bit vector. An OR operation is performed for overlapping bits. In this case, by performing an OR operation, it may be erroneously determined that there is a corresponding adjacency even among N-grams that are not originally in the corresponding adjacency, but no search omission occurs. Of course, it is possible to set the bit vector to an arbitrary size instead of 1000 bits.

  In FIG. 13, the bit vector is divided into a plurality of sequences, and the sequence of flag bits is omitted. In the example of FIG. 13, the bit vector (FIG. 13A) is divided into a sequence of a predetermined length, for example, 1 byte (FIG. 13B), and can be managed in sequence units using the sequence number. When there is no bit, the sequence itself is also omitted. That is, as shown in FIG. 13C, a sequence number is used in addition to an identifier (KEY) indicating N-gram, and a sequence number not including a flag bit is omitted.

  Actually, the index data in the format of FIG. 13C is stored in the compressed index storage unit 15b (for example, a hard disk) as shown in FIG. 14, and at the stage where the N-gram is determined by decomposing the keyword, The corresponding N-gram index data (format in FIG. 13C) is extracted from the compression index storage unit 15b and expanded into the main memory 15a, and the above-described shift processing and AND operation are performed to determine the adjacent relationship.

  Next, a second embodiment in which the present invention is applied to a document search apparatus using both an index of a morphological analysis technique and an index of an N-gram technique will be described.

  FIG. 15 shows a document search apparatus 200 according to the second embodiment. In this figure, the document search apparatus 200 includes a search condition input unit 20, a search front end 21, an N-gram document search unit 22, and an N-gram index storage. A unit 23, a morpheme word document search unit 24, a morpheme word index storage unit 25, a search result synthesis unit 26, a synthesis search result output unit 27, and the like. This example can also be realized by installing software in the computer as in the first embodiment.

  The search condition input unit 20 is for inputting a search condition. The search condition may be a natural sentence input. The search front end 21 outputs a hiragana character string keyword or a katakana character string keyword to the N-gram document search unit 22 according to the search condition, and outputs a morpheme word as a keyword to the morpheme word search unit 24. The N-gram document search unit 22 and the N-gram index storage unit 23 correspond to the N-gram document search device 100 in FIG. The N-gram index storage unit 23 corresponds to the index storage unit 15. The morpheme word document search unit 24 performs a document search with reference to the morpheme word unit index stored in the morpheme word index storage unit 25. The morpheme unit index consists of an index record including the ID of the document in which the morpheme word appears with the morpheme word as an entry. Appearance position information in the document may be held.

  The search result combining unit 26 combines the search results of the N-gram document search unit 22 and the morpheme document search unit 24. The same keyword may be supplied to the N-gram document search unit 22 and the morpheme document search unit 24 so as to perform a search without omission by ORing the search results. A keyword of a hiragana character string or similar katakana character string that is not included is supplied to the N-gram document search unit 22, and a keyword corresponding to the morpheme word in the search condition is supplied to the morpheme word document search unit 24. The corresponding processing may be performed by the search result combining unit 26 in accordance with the AND or OR condition. The combined search result output unit 27 outputs the combined search result as, for example, a document list.

  The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the spirit of the invention. For example, although the above example has been described as a stand-alone device, the search method of the present invention may be realized using a plurality of computer systems. For example, it may be realized using a document search server device and an arbitrary client device (including a personal computer, a portable information terminal, etc.).

It is a block diagram explaining the structure of the document search device of Example 1 of this invention. It is a flowchart explaining the example of the index registration process of the said Example 1. FIG. It is a figure for demonstrating the operation | movement of an index registration process, and explaining the example of a document. It is a figure for demonstrating the operation | movement of an index registration process, and explaining the example of a pseudo document. It is a figure for demonstrating the operation | movement of an index registration process, and is a figure explaining the example of a bit vector. It is a flowchart explaining the example of the search process of the said Example 1. FIG. It is a figure explaining the example with the N-gram neighborhood relationship corresponding to a search keyword. It is a figure explaining the example which does not have the N-gram neighborhood relationship corresponding to a search keyword. It is a flowchart explaining the example of the process which discriminate | determines from the bit vector whether there exists N-gram neighborhood relation corresponding to a search keyword. It is a figure explaining the example of N-gram in the flowchart of FIG. It is a figure explaining the operation example of the flowchart of FIG. It is a figure explaining the modification of the above-mentioned Example. It is a figure explaining the other modification of the above-mentioned Example. It is a figure explaining the above-mentioned other modification. It is a block diagram explaining the structure of the document search device of Example 2 of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Document input part 11 Character string extraction part 12 Character string connection part 13 Bit vector production | generation part 14 Index registration part 15 Index storage part 15 Index storage part 15a Main memory 15b Compression index storage part 16 Keyword input part 17 Search part 18 Search result output Unit 20 search condition input unit 21 search front end 22 N-gram document search unit 23 N-gram index storage unit 24 morpheme word document search unit 25 morpheme word index storage unit 26 search result synthesis unit 27 synthesis search result output unit 100 document search Apparatus 200 document search apparatus 1000 personal computer 1001 recording medium

Claims (13)

First search means for performing a document search using a search dictionary generated from a morphological analysis result;
Second search means for searching for a document using an N-gram dictionary,
The second search means includes
Extract and concatenate at least one character string selected from hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana, and alphabet from each search target document. And pseudo document generation means for generating a pseudo document;
Appearance position storage means for storing appearance position information representing an appearance position of the N-gram in the pseudo document for each entry of the N-gram for each of the search target documents;
A search keyword composed of at least one character string selected from the hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana and alphabet is collated with the appearance position information. And a document specifying means for specifying a document including the search keyword. Extract and concatenate at least one character string selected from hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana, and alphabet from each search target document. And pseudo document generation means for generating a pseudo document;
Appearance position storage means for storing appearance position information representing an appearance position of the N-gram in the pseudo document for each entry of the N-gram for each of the search target documents;
A search keyword composed of at least one character string selected from the hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana and alphabet is collated with the appearance position information. And a document specifying means for specifying a document including the search keyword. For each of the documents to be searched, the appearance position information indicating the appearance position of the N-gram for each N-gram entry is stored as a bit vector for setting a flag bit at the bit position corresponding to the position. Position storage means;
Document specifying means for checking a search keyword against the appearance position information and specifying a document including the search keyword;
When the search keyword is composed of two or more N-grams, it is determined that the flag bit position of each bit vector of the N-gram has a corresponding adjacency, and a document including the search keyword is obtained. A document search apparatus characterized by specifying. The pseudo-document generating means selects at least one kind of character string selected from hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana, and alphabet from each of the documents to be searched. A pseudo document generation step of extracting and concatenating and generating a pseudo document;
An appearance position storage step of storing, for each of the search target documents, appearance position information representing an appearance position of the N-gram in the pseudo document for each entry of the N-gram;
The document specifying means selects a search keyword composed of at least one kind of character string selected in advance from the hiragana character string, katakana character string, alphabet character string, and a mixed character string of hiragana, katakana and alphabet. A document search method comprising: a document specifying step of specifying a document including the search keyword by collating with appearance position information. The appearance position storage means sets, for each of the search target documents, appearance position information indicating the appearance position of the N-gram for each entry of the N-gram, and sets a flag bit at the bit position corresponding to the position. An appearance position storing step of storing as a bit vector;
A document specifying means for checking a search keyword against the appearance position information and specifying a document including the search keyword;
When the search keyword is composed of two or more N-grams, it is determined that the flag bit position of each bit vector of the N-gram has a corresponding adjacency, and a document including the search keyword is specified. A document retrieval method characterized by: Extract and concatenate at least one character string selected from hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana, and alphabet from each search target document. And pseudo document generation means for generating a pseudo document;
Appearance position storage means for storing appearance position information representing an appearance position of the N-gram in the pseudo document for each entry of the N-gram for each of the search target documents;
A search keyword composed of at least one character string selected from the hiragana character string, katakana character string, alphabet character string, and mixed character string of hiragana, katakana and alphabet is collated with the appearance position information. A computer program for searching for a document, which is used for realizing a document specifying means for specifying a document including the search keyword.   7. The document retrieval use according to claim 6, wherein the appearance position storage means represents the appearance position information representing the appearance position of the N-gram in the pseudo document by a bit vector that sets a flag bit at a bit position corresponding to the position. Computer program.   8. The computer program for searching a document according to claim 6, wherein the pseudo document is generated by degenerating a plurality of identical character strings in the same document into one.   9. The computer program for searching a document according to claim 6, 7 or 8, wherein a blank character is used to represent a break between adjacent character strings in the pseudo document.   The computer program for document search according to any one of claims 6 to 9, wherein the at least one character string selected in advance is a hiragana character string or a katakana character string.   8. The computer program for searching a document according to claim 7, wherein when the bit length of the bit vector exceeds a predetermined length, the bit vector is folded at the predetermined length position and ORed with the flag bit.   8. The computer for document search according to claim 7, wherein when the bit length of the bit vector exceeds a predetermined length, the bit vector is divided into the sequence of the predetermined length, and the sequence not including the flag bit is omitted and complemented. program. For each of the documents to be searched, the appearance position information indicating the appearance position of the N-gram for each N-gram entry is stored as a bit vector for setting a flag bit at the bit position corresponding to the position. Position storage means;
And a document specifying means for specifying a document including the search keyword by comparing the search keyword with the appearance position information, and
When the search keyword is composed of two or more N-grams, it is determined that the flag bit position of each bit vector of the N-gram has a corresponding adjacency, and a document including the search keyword is obtained. A computer program for document search characterized by specifying.

Images