JP2005108158A - Document retrieval apparatus and method, and program for same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document retrieval apparatus capable of retrieving documents which cannot be structured and using a new retrieval method different from the conventional one. <P>SOLUTION: Each of a plurality of representative documents to be retrieved which have been extracted at a specified rate from a plurality of documents to be retrieved is divided into paragraphs, and a document structure of each of the representative documents to be retrieved is classified. Also, by using a sample document, retrieval condition data are created. Then, the kind of the document structure of an arbitrary document to be retrieved out of the plurality of documents to be retrieved is detected, a paragraph for judging retrieval in the arbitrary document to be retrieval is specified on the basis of a position and the kind of paragraph stating retrieval contents to be held by the retrieval condition data, full text retrieval of the paragraph stating the retrieval contents is performed by using a key word held by the retrieval condition data, and it is judged on the basis of the result of full text retrieval whether or not the arbitrary document to be retrieved is output as a retrieval result. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a document search apparatus, a document search method, and a program for searching for a document similar to a sample document from a plurality of search target documents.

Conventionally, various document search methods have been developed and proposed. Then, as a technique for reducing malfunctions in document search (hereinafter referred to as search noise) such that search results that the user does not want are output, for example, in an apparatus that searches a document desired by a user from a large number of documents, When searching for a search target document that can be structured by document elements such as chapters, sections, and paragraphs, first accept the search conditions from the user, specify the document elements in the document based on the search conditions, and specify There is a technique for outputting a document including a document element as a search result (see, for example, Patent Document 1).
JP-A-6-309368

However, in the above-described technique, if a search target document cannot be structured, a search result cannot be obtained. In addition, because of the recent computerization, a large amount of documents are converted into data and accumulated in large quantities. Therefore, a document retrieval apparatus that can accurately retrieve a document desired by a user from among a large amount of documents to be retrieved has been developed. It is desired.
SUMMARY OF THE INVENTION An object of the present invention is to provide a document search apparatus that can search for a document that cannot be structured, and that uses a new search method that has not existed before.

  The present invention has been made to solve the above-described problem, and is a document search apparatus for searching for a search target document similar to a sample document from a plurality of search target documents. First paragraph dividing means for dividing each of the plurality of representative search target documents, and document structure classification means for classifying the document structure of each representative search target document based on the respective paragraph divisions of the representative search target documents Document structure storage means for storing a plurality of classified document structure information for each type of document structure, second paragraph dividing means for dividing the sample document, and paragraphing of the sample document Sample document structure detection for detecting the document structure type of the sample document based on the result and the document structure information stored in the document structure storage means A third paragraph dividing means for dividing an arbitrary search target document among the plurality of search target documents, a result of the paragraph division by the third paragraph dividing means, and the document structure storage means; A search target document structure detecting means for detecting a document structure type of the arbitrary search target document based on the document structure information, and the document structure type being the same as the document structure type of the sample document A document search apparatus comprising: a search target document determination unit that sets an arbitrary search target document as a search target document similar to the sample document.

  According to the present invention, the first paragraph dividing unit divides each of the plurality of representative search target documents extracted from the plurality of search target documents, and the document structure classification unit determines the paragraph based on each of the representative search target documents. The document structure of each representative search target document is classified. The second paragraph dividing unit divides the sample document into paragraphs, and the sample document structure detecting unit extracts the sample document based on the result of the paragraph division of the sample document and the document structure information stored in the document structure storage unit. Detect the type of document structure. Then, the third paragraph division means divides a paragraph for an arbitrary search target document among the plurality of search target documents, and the search target document structure detection means uses the result of the paragraph division by the third paragraph division means and the document structure storage means. Based on the stored document structure information, the type of the document structure of an arbitrary search target document is detected, and the search target document determination means has the same document structure type as that of the sample document. An arbitrary search target document is set as a search target document similar to the sample document. Thereby, the paragraph structure and the paragraph type are determined to grasp the document structure of the sample document or the search target document, so that the document structure can be specified for any document. In addition, a search target document having the same sentence structure as the sample document, and a paragraph containing the keyword described in the search content description paragraph and having the same paragraph position as the search content description paragraph in the sample document are retained. It is possible to provide a document search apparatus of a new technique in which a search target document is output as a search result. In this document search apparatus, if a plurality of sample documents are prepared, a plurality of search condition data are created. Therefore, a document search with high accuracy can be performed by using a plurality of combinations of document structure, search content description paragraphs, and keywords. .

  The present invention has been made to solve the above-described problem, and is a document search apparatus for searching for a search target document similar to a sample document from a plurality of search target documents. Based on the first paragraph division means for dividing each of the plurality of representative search target documents extracted at a predetermined ratio and the respective paragraph divisions of the representative search target documents, the document structure of each of the representative search target documents is classified. Document structure classification means, document structure storage means for storing a plurality of classified document structure information for each type of document structure, second paragraph division means for dividing the sample document into paragraphs, and the sample document For detecting the document structure type of the sample document based on the result of the paragraphing of the document and the document structure information stored in the document structure storage means A document structure detection unit, a paragraph position of a search content description paragraph specified by the user in the sample document, a keyword included in the search content description paragraph, a paragraph type indicating the description content in the search content description paragraph, and Search condition data creation means for creating search condition data that holds the type of document structure of the sample document, search condition data storage means for storing the search condition data, and any search target among the plurality of search target documents Based on the third paragraph division means for dividing the document into paragraphs, the result of the paragraph division by the third paragraph division means, and the information on the document structure stored in the document structure storage means, the arbitrary search Search target document structure detecting means for detecting the document structure type of the target document, and the search condition holding the document structure type of the arbitrary search target document When data is recorded in the search condition data storage means, search condition data reading means for reading the search condition data from the search condition data storage means, and the search contents held in the read search condition data Based on the position of the described paragraph and the paragraph type, a paragraph specifying means for specifying a search determination paragraph in the arbitrary search target document, and a keyword held in the search condition data read by the search condition data reading means And a search means for performing a full-text search of the paragraph containing the search content and determining whether to output the arbitrary search target document as a search result based on a result of the full-text search. A document retrieval apparatus.

  According to the present invention, each of the plurality of representative search target documents extracted from the plurality of search target documents at a predetermined ratio is divided into paragraphs, and based on this paragraph division, the document structure of each representative search target document is classified, Document structure information is recorded in the document structure storage means for each type of document structure. Here, the ratio of extraction from the plurality of search target documents of the representative search target document is such that one of the plurality of search target documents has the same content as the contents of any one of the plurality of search target documents. . Therefore, the document structure information recorded in the document structure storage means indicates one of the plurality of search target documents. Then, the sample document is divided into paragraphs, and the type of document structure of the sample document is detected. Also holds the paragraph position of the search content description paragraph specified by the user in the sample document, the keywords contained in the search content description paragraph, the paragraph type indicating the description content in the search content description paragraph, and the document structure type of the sample document Create search condition data. Further, in the search process, an arbitrary search target document among a plurality of search target documents is divided into paragraphs, and the type of document structure of the arbitrary search target document is detected. Then, when the search condition data holding the type of document structure of the search target document is recorded in the search condition data storage means, the search condition data is read from the search condition data storage means and read. Based on the position and paragraph type of the search content description paragraph held in the search condition data, the search determination paragraph in the arbitrary search target document is specified. In other words, if the paragraph in the search target document that matches the document structure of the sample document and the same paragraph position as the search content description paragraph of the sample document matches the paragraph type of the search content description paragraph, that search target document The paragraph inside is specified as the paragraph for search judgment. Then, a full text search is performed on the search content description paragraph using a keyword held in the search condition data, and a search target document is output as a search result based on the result of the full text search. Thereby, the paragraph structure and the paragraph type are determined to grasp the document structure of the sample document or the search target document, so that the document structure can be specified for any document. In addition, a search target document having the same sentence structure as the sample document, and a paragraph containing the keyword described in the search content description paragraph and having the same paragraph position as the search content description paragraph in the sample document are retained. It is possible to provide a document search apparatus of a new technique in which a search target document is output as a search result. In this document search apparatus, if a plurality of sample documents are prepared, a plurality of search condition data are created. Therefore, a document search with high accuracy can be performed by using a plurality of combinations of document structure, search content description paragraphs, and keywords. .

  According to the present invention, the keyword is a word having a predetermined part of speech obtained by morphological analysis of the sentence in the search content description paragraph. As a result, since the keyword is automatically created when the user specifies a search content description paragraph, the user is required to perform a full text search of a search target document having a paragraph similar to the description content of the search content description paragraph. There is no need to input a keyword, and the user's labor can be reduced.

  Further, the present invention determines a keyword inclusion paragraph including a keyword in the representative search target document based on a result of full text search of the representative search target document using the keyword in the search content description paragraph, The combination of the type of document structure and the paragraph position in the representative search target document of the keyword inclusion paragraph, the type of document structure held in the search condition data stored in the search condition data recording means, and the paragraph position of the paragraph describing the search content Based on the comparison of the combinations, the representative search target document holding the keyword inclusion paragraph is determined as a candidate for a new sample document. Then, a document designated by the user from the candidates is set as a new sample document. Accordingly, if the document search device creates search condition data for a new sample document other than the sample document designated in advance by the user, a more detailed search can be performed.

  Further, the present invention determines the line type of each line in either the sample document or the representative search target document based on the description content in the line, and determines whether the sample document or the representative search target document is based on the line type. Determine the paragraph start position. Then, in either the sample document or the representative search target document, the line from the line at the beginning of the paragraph to the line before the line at the beginning of the next paragraph is determined as one paragraph, and the paragraph type is the line included in the paragraph. Determine based on the row type. Thereby, the kind of each paragraph divided into paragraphs can be determined.

  The present invention is also a document search method of a document search apparatus for searching for a search target document similar to a sample document from a plurality of search target documents, wherein a plurality of documents extracted at a predetermined ratio from the plurality of search target documents. A first paragraph dividing process for dividing each of the representative search target documents, a document structure classification process for classifying the document structure of each of the representative search target documents based on the paragraph division of each of the representative search target documents, Document structure recording process for recording the classified document structure information in the document structure storage means for each document structure type, a second paragraph dividing process for dividing the sample document into paragraphs, and a paragraph division of the sample document Sample document structure for detecting the document structure type of the sample document based on the result of the document structure and the document structure information stored in the document structure storage means. The detection process, the paragraph position of the search content description paragraph specified by the user in the sample document, the keyword included in the search content description paragraph, the paragraph type indicating the description content in the search content description paragraph, and the sample document A search condition data creating process for creating search condition data for holding the document structure type, a search condition data recording process for recording the search condition data in a search condition data storage means, Based on a third paragraph dividing process for dividing an arbitrary document to be searched, a result of the paragraph dividing by the third paragraph dividing process, and information on the document structure stored in the document structure storage unit, A search target document structure detection process for detecting a document structure type of the arbitrary search target document, and a document structure type of the arbitrary search target document. When the search condition data held is recorded in the search condition data storage means, the search condition data reading process for reading the search condition data from the search condition data storage means, and holding in the read search condition data A paragraph specifying process for specifying a search determination paragraph in the arbitrary search target document based on the position of the search content description paragraph and the paragraph type, and the search condition read in the search condition data reading process A search step of performing a full-text search of the search content description paragraph using a keyword held in the data and determining whether or not to output the arbitrary search target document as a search result based on the result of the full-text search This is a document search method characterized by the above.

  The present invention also provides a program for causing a computer of a document search apparatus that searches for a search target document similar to a sample document from a plurality of search target documents to be extracted from the plurality of search target documents at a predetermined ratio. A first paragraph dividing process for dividing each of the plurality of representative search target documents, and a document structure classification process for classifying the document structure of each of the representative search target documents based on the paragraph division of each of the representative search target documents; A document structure recording process for recording the classified document structure information in a document structure storage unit for each document structure type, a second paragraph division process for dividing the sample document into paragraphs, Based on the result of the paragraphing and the information on the document structure stored in the document structure storage means, the document structure type of the sample document is determined. A sample document structure detection process to be performed, a paragraph position of a search content description paragraph specified by the user in the sample document, a keyword included in the search content description paragraph, and a paragraph indicating the description content in the search content description paragraph A search condition data creating process for creating search condition data that holds a type and a document structure type of the sample document; a search condition data recording process for recording the search condition data in a search condition data storage unit; A third paragraph dividing process for dividing an arbitrary search target document among the search target documents, a result of the paragraph division by the third paragraph dividing process, and information on the document structure stored in the document structure storage unit And a search target document structure detection process for detecting a document structure type of the arbitrary search target document, and the arbitrary search target A search condition data reading process for reading the search condition data from the search condition data storage means when the search condition data holding the document structure type of the document is recorded in the search condition data storage means; Paragraph specifying processing for specifying a search determination paragraph in the arbitrary search target document based on the position of the search content description paragraph and the paragraph type held in the read search condition data, and reading the search condition data Whether or not to perform a full-text search of the search content description paragraph using a keyword held in the search condition data read in the process, and whether to output the arbitrary search target document as a search result based on the result of the full-text search This is a program that causes a computer to execute search processing for determination.

  Hereinafter, a document search apparatus according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the document search apparatus according to the embodiment. In this figure, reference numeral 1 denotes a document search apparatus for searching for a document desired by a user from a large number of search response documents. Reference numeral 2 denotes a search target document database that stores a large number of search target documents such as 100,000 or 1 million. In this embodiment, the document search apparatus 1 searches for a document desired by the user from the search target documents recorded in the search target document database 2. The document search apparatus 1 and the search target document database 2 are connected via a communication network.

  In the document search apparatus 1, reference numeral 101 denotes a search condition learning document storage unit that stores a search condition learning document (representative search target document). Here, the search condition learning document is a document extracted at a predetermined rate from the search target documents recorded in the search target document database 2. The ratio of documents extracted from the search target document database 2 is similar to the content of at least one of the search target documents recorded in the search target document database 2 in the search condition learning document storage unit 101. Probably one of the content documents is a ratio that can be retained.

  A sample document storage unit 102 stores a sample document prepared by the user. The sample document is a document in which a part of the content of the document that the user desires to search is described, and the user specifies a specific range in which the content to be searched is described in the sample document. As a result, the document search device 1 searches the search target document database 2 for a search target document in which the content similar to the content of the text in the range specified by the user in the sample document is described. In the present embodiment, the range specified by the user in the sample document is a paragraph in the sample document.

  Reference numeral 103 denotes a paragraph analysis unit (first paragraph division means, second paragraph division means, third paragraph division means, line type determination means, paragraph head position determination means, paragraph determination means), which is used for search condition learning. The document, the sample document, and the search target document are analyzed to divide the paragraph. Reference numeral 104 denotes a document structure classification unit (document structure classification unit, sample document structure detection unit, search target document structure detection unit, paragraph type determination unit). Based on this, the document structure of each search condition learning document is classified. The document structure classification unit 104 records information on different types of document structures in the document structure storage unit (document structure storage unit) 105. The document structure is represented by the number of paragraphs in one document, the paragraph identifier of each paragraph, and the paragraph type indicating the description content of each paragraph. The document structure information holds the paragraph identifier of each paragraph and the paragraph type of each paragraph in one document.

  Reference numeral 106 denotes a search condition data creation unit (search condition data creation means), the position of a paragraph (hereinafter referred to as a search content description paragraph) specified by a user in a sample document, a keyword included in the search content description paragraph, Search condition data that holds the document structure number indicating the document structure of the sample document, the type of paragraph describing the search content, and the identification information of the sample document is created. This search condition data is used when searching a document desired by the user from a large number of search target documents recorded in the search target document database 2. Reference numeral 107 denotes a search condition data storage unit (search condition data storage means) that stores search condition data.

  Further, 108 performs a full-text search of the search condition learning document using a keyword (that is, a keyword in the search content description paragraph) held in the search condition data, and selects a paragraph in the search condition learning document including the keyword. A keyword inclusion paragraph determination unit (keyword inclusion paragraph determination means) that determines a keyword inclusion paragraph. Reference numeral 109 denotes a new sample document candidate determination unit (new sample document candidate determination means) that determines whether or not a search condition learning document holding a keyword inclusion paragraph is a candidate for a new sample document. Reference numeral 110 denotes a new sample document candidate storage unit that stores data of a search condition learning document that is a new sample document candidate. Reference numeral 111 designates a document specified by the user from the search condition learning document for the new sample document candidate recorded in the new sample document candidate storage unit 110 based on the user's instruction as a new sample document. Is a new sample document designation section to be recorded in

  A search target document reading unit 112 reads a large number of search target documents stored in the search target document database 2 one by one. Reference numeral 113 denotes the document structure of the search target document read by the search target document reading unit 112, the search condition data recorded in the search condition data storage unit 107, and the document structure information recorded in the document structure storage unit 105. A search processing unit (search condition data reading means, paragraph specifying means, search means, search target document determination means) for searching for a search target document that the user desires to search based on. A search result output unit 114 outputs the contents of the search target document obtained as a search result to a monitor or the like.

  The document search apparatus 1 uses the above-described processing units to search for a document desired by the user according to four processing flows. The first is a processing flow for classifying the document structure of the search condition learning document and recording information of a plurality of different types of document structures in the document structure storage unit 105. The second is a processing flow for creating search condition data based on the sample document and the document structure information. The third is a processing flow in which a candidate for a new sample document is determined and only a document designated by the user among the candidates for the new sample document is used as a new sample document. The fourth is a processing flow for searching a search target document desired by the user from the search condition database 2. Thereby, a search result desired by the user is obtained.

  FIG. 2 is a diagram showing an example of a search condition learning document. This figure shows three documents: a search condition learning document a, a search condition learning document b, and a search condition learning document c. The search condition learning document is probabilistically stored in the search condition learning document storage unit 101 as a document having the same content as at least one of the search target documents recorded in the search target document database 2. One records only the quantity that can be held.

FIG. 3 shows a sample document. The sample documents used in this embodiment are two sample documents, sample document A and sample document B. In the present embodiment, for convenience of explanation, a document having the same contents as the search condition learning documents a and b is used as a sample document. These sample documents A and B are recorded in the sample document storage unit 102.
Note that the search target document, the search condition learning document, and the sample document in this embodiment are documents configured by lines that fall into one of a normal sentence line, a blank line, a word line with a symbol, or a word line without a symbol. Suppose that Here, a normal sentence is a sentence including predicates and punctuation marks. A word line with a symbol means a line in which the beginning of the line starts with a symbol such as a numeral or alphabet, and the subsequent sentence is composed only of body words. Moreover, the line of a word without a symbol means a line in which a sentence is composed only of a body without a symbol such as a numeral or alphabet at the beginning of the line.

  FIG. 4 is a diagram showing document structure information. As shown in this figure, the document structure storage unit 105 associates, as document structure information, a document structure number, the number of a paragraph constituting the document structure indicated by the document structure number, and the type of each paragraph. I remember. The document structure data is created by the document structure classification unit 105 and recorded in the document structure storage unit 105.

  FIG. 5 shows an example of search condition data. As shown in this figure, the search condition data storage unit 107 includes a search condition data number, a sample document number, a sentence structure number, a position of the search content description paragraph, a paragraph type of the search content description paragraph, and a keyword. Correspondingly recorded. This search condition data is created by the search condition data creation unit 106 and recorded in the search condition data storage unit 107.

Next, a processing flow until the document structure of the search condition learning document is classified in the document search device 1 and the document structure information is recorded in the document structure storage unit 105 will be described. FIG. 6 is a diagram showing a processing flow when recording document structure classification and document structure information.
First, the paragraph analysis unit 103 reads one search condition learning document recorded in the search condition learning document storage unit 101 (step S101). It is assumed that the search condition learning document read here is a search condition learning document a. Next, the paragraph analysis unit 103 takes out the top line of the read search condition learning document a (step S102). Next, the paragraph analysis unit 103 analyzes which line type the line extracted in step S102 is a line of a normal sentence, a blank line, a line of a word with a symbol, or a line of a word without a symbol (step S103). Store it on a memory. Next, the paragraph analysis unit 103 checks whether or not there is a next line in the search condition learning document a after the lines processed in steps S101 to S104 (step S104). The processing from step S101 to step S103 is repeated for the next line, and this is performed for all lines in the search condition learning document a.

  FIG. 7 is a diagram showing data held in the memory of the paragraph analysis unit. In this figure, the line number indicates the line number of the search condition learning document a, and the line number, the line type of the analysis result, and the number of repetitions thereof are recorded. For example, the fourth line of the search condition learning document a is a normal sentence, and this normal sentence is repeated twice (the fifth line is also a normal sentence). In this case, the row type for the fifth row is not recorded, and the number of repetitions for the fourth row is changed from 1 to 2. In this way, the paragraph analysis unit 103 analyzes line types for all lines of the search condition learning document.

  Next, the paragraph analysis unit 103 extracts the line type and the number of repetitions corresponding to the line number 1 recorded in the data (FIG. 6) recorded on the memory (step S105). Then, based on the information of line number 1 extracted from the data on the memory shown in FIG. 6, it is determined whether or not the line of line number 1 is a line corresponding to the paragraph head position (step S106). Whether the line corresponds to the head position of this paragraph is determined by (1) the information of the line number is the line of the word with a symbol whose repetition count is 1, and (2) the information of the line number is the repetition count. The line of the unsigned word in which 1 is 1, (3) the information of the line number is a blank line, the next line is a normal sentence, or the line of the word having a repetition count of 2 or more, (4) the information of the line number However, the determination is made based on whether or not it applies to any one of the lines (1) to (4) of the unsigned word line having a plurality of repetitions. Then, the paragraph analysis unit 103 checks whether information corresponding to other line numbers is recorded in the memory (step S107), repeats steps S105 and S106, and repeats the line numbers recorded in the memory. For all, it is determined whether or not the line corresponds to the paragraph head position.

  Here, the line numbers corresponding to (1) in the search learning document a are 1, 3, 6, 8, 11, and 14. The line number corresponding to (2) is 16 in the search learning document a. The line number corresponding to (3) does not exist in the search learning document a. In the search learning document a, the line number corresponding to (4) is 18. The paragraph analysis unit 103 determines that the data corresponding to line number 1 extracted from the data shown in FIG. 6 is always data indicating a line corresponding to the paragraph head position. In this way, the paragraph analysis unit 103 determines which line of the search learning document a corresponds to the paragraph head position, and then from the line at the paragraph head position to the line before the line at the next paragraph head position. As a single paragraph, the paragraph in the search condition learning document a is determined (step S108). The paragraph analysis unit 103 includes the first and second lines as the first paragraph, the third and fifth lines as the second paragraph, the sixth and seventh lines as the third paragraph, and the eighth and tenth lines as the fourth paragraph, The 13th line is the 5th paragraph, the 14th to 15th lines are the 6th paragraph, the 16th to 17th lines are the 7th paragraph, the 18th to 20th lines are the 8th paragraphs, and each paragraph number and the line that each paragraph holds The number and the line type corresponding to each line number are output to the document structure classification unit 104 as a result of the paragraph analysis of the search learning document a.

Next, the document structure classification unit 104 determines the paragraph type of each paragraph based on each paragraph number of the search condition learning document a, the line number corresponding to each paragraph, and the line type corresponding to each line number ( Step S109). Here, when determining the paragraph type, the document structure classification unit 104 confirms whether the line type of each line in the paragraph applies in the order of a normal sentence, a word line without a symbol, a word line with a symbol, and a blank line. Go to the first line type and make it the paragraph type. For example, the lines included in the first paragraph of the search learning document a are the first and second lines, and since the line type of the second line is a normal sentence, the paragraph type of the first paragraph is a normal sentence. . In this way, the document structure classification unit 104 determines the paragraph type for each paragraph of the search condition learning document a.
The document structure classification unit 104 generates a paragraph identifier based on the line corresponding to the paragraph head position. In the present embodiment, the paragraph identifier is information indicating whether or not the first character of the line at the paragraph head position is a number. For example, in the first paragraph of the search condition learning document a, the paragraph identifier is “number”, and in the seventh paragraph of the search condition learning document a, the paragraph identifier is “none”. For example, the paragraph identifier may be a paragraph number or a paragraph title (first word in a line corresponding to the head position).

  Then, the document structure classification unit 104 associates the paragraph identifier of each paragraph of the search condition learning document a with the paragraph type of each paragraph, and records it in the document structure storage unit 105 as information of document structure number 1 (step S110). The information recorded in step S109 is information corresponding to the document structure number 1 shown in FIG. Further, after step S109, the document structure classification unit 104 notifies the paragraph analysis unit 103 that the recording of the document structure classification for the search condition learning document a has been completed. Then, the paragraph analysis unit 103 confirms whether another search condition learning document is recorded in the search condition learning document storage unit 101 (step S111), and another search condition learning document is recorded. In step S101, the processing of step S101 is started. Then, the processing from step S101 to step S110 is performed on the search condition learning documents recorded in all the search condition learning document storage units 101. Note that the processes in steps S101 to S111 described above are performed before the user issues a search start instruction, and data of a plurality of document structures is recorded in the document structure storage unit 105. In FIG. 4, information corresponding to the document structure number 2 indicates the document structure of the search condition learning document b.

Next, a process until the search condition data is created based on the sample document and the sentence structure information in the document search apparatus 1 will be described. FIG. 8 shows a flow of search condition data creation processing.
First, when a search start instruction is input from the user to the document search apparatus 1, the paragraph analysis unit 103 reads one sample document recorded in the sample document storage unit 102 (step S201). Here, the sample document read by the paragraph analysis unit 103 in step S201 is referred to as a sample document A. Next, the paragraph analysis unit 103 determines each paragraph of the sample document A by using the same processing technique as the processing from step S102 to step S108 described above (step S202). Next, the document structure classification unit 104 determines the paragraph type of each paragraph of the sample document A using the same processing method as the processing in step S109 described above (step S203). Then, the document structure classification unit 104 notifies the search condition data creation unit 106 of the paragraph identifier of each paragraph of the sample document A and the paragraph type of each paragraph (step S204). There are 1 to 8 paragraphs in the sample document A. The paragraph types of each paragraph are as follows: 1st paragraph: normal sentence, 2nd paragraph: normal sentence, 3rd paragraph: normal sentence, 4th paragraph: normal sentence, 5th paragraph: normal sentence, 6th paragraph: normal sentence, 7th paragraph: normal sentence, 8th paragraph: unsigned word.

  Next, the search condition data creation unit 106 determines the document structure of the sample document A from the paragraph identifier of each paragraph of the sample document A and the paragraph type of each paragraph. “Normal sentence”, the second paragraph has the paragraph identifier “number” and the paragraph type “normal sentence”, the third paragraph has the paragraph identifier “number”, the paragraph type “normal sentence”, and the fourth paragraph has the paragraph identifier “Number”, paragraph type “Normal Text”, paragraph 5 “Paragraph” “Number”, paragraph type “Normal Sentence”, paragraph 6 “paragraph identifier“ Number ”, paragraph type“ Normal Sentence ”, It is determined that the seventh paragraph has a structure with the paragraph identifier “None” and the paragraph type “Normal Text”, and the eighth paragraph has a structure with the paragraph identifier “None” and the paragraph type “Word without a symbol”, and stores the document structure. Information of the same document structure as that of the sample document A from the part 105 (FIG. 4) Reading the document structure number indicated (step S205). Here, the document structure number of the sample document A is “1”. Next, the search condition data creation unit 106 receives, from the user, designation of a paragraph in which search contents are described from among the paragraphs of the sample document A. Note that the paragraph specified by the user is the third paragraph of the sample document A. Then, the paragraph specified by the user is set as a search content description paragraph in the sample document A (step S206). Next, the search condition data creation unit 106 performs morphological analysis on the sentence in the search content description paragraph, and determines a word corresponding to a preset part of speech as a search keyword (step S207). In this embodiment, the part of speech of a word used as a keyword is a noun. The search condition data creation unit 106 uses the words “conventional”, “dictionary”, and “usage frequency” in the third paragraph of the sample document A as keywords.

  Next, the search condition data creation unit 106 detects the paragraph position in the sample document A of the search content description paragraph (step S208). At this time, the search condition data creation unit 106 first divides the number of paragraphs of the sample document A by 3, and if it is divisible, divides the sample document A by the number of paragraphs into the upper part, the middle part, and the lower part. In addition, the search condition data creation unit 106 divides the number of paragraphs of the sample document A by 3, and if the obtained value is a numerical value including a decimal point, the natural number is obtained by rounding off the first decimal place. The sample document A is set from the top to the top and the middle, and the remaining paragraphs are set to the bottom. In this way, the search condition data creation unit 106 detects whether the position in the sample document A of the search content description paragraph is in the upper, middle or lower paragraph position in the sample document A.

Then, the search condition data creation unit 106 assigns an arbitrary symbol “A” as the identification information of the sample document A, the identification information of the sample document A, the document structure number indicating the document structure of the sample document A, and the sample document A The search content description paragraph position information, the search content description paragraph type, and the keyword obtained from the search content description paragraph are associated with the data number “1” of the search condition data as one search condition data. And recorded in the search condition data storage unit 107 (step S209). After step 209, the search condition data creation unit 106 notifies the paragraph analysis unit 103 that the creation of the search condition data for the sample document A has been completed. Then, the paragraph analysis unit 103 checks whether or not another sample document is recorded in the sample document storage unit 102 (step S210). If another sample document is recorded, the process of step S201 is performed on the document. The subsequent processing is started. Then, the processes in steps S201 to S209 are performed for all the sample documents recorded in the sample document storage unit 102.
As described above, the search condition data is created by the processing from step S201 to step S209.

Next, a process flow of the document search apparatus 1 that determines a candidate for a new sample document and sets only a new sample document candidate designated by the user as a new sample document will be described. FIG. 9 is a diagram showing a processing flow for creating a new sample document.
When the creation of the search condition data described above is completed for all the sample documents recorded in the sample document storage unit 102, the search condition data creation unit 106 starts the process of determining a new sample document candidate in the keyword inclusion paragraph determination unit 108. Instruct. Then, the keyword inclusion paragraph determination unit 108 reads one arbitrary search condition learning document from the search condition learning document storage unit 101 (step S301). Here, the search condition learning document read in step S301 is set as a search condition learning document d. Next, the keyword inclusion stage paragraph determination unit 108 reads one search condition data recorded in the search condition data storage unit 107 (step S302). Here, the search condition data read in step S302 is the search condition data corresponding to the search condition data number “1” created using the sample document “A” in the search condition data shown in FIG. And Then, the keyword inclusion paragraph determination unit 108 performs a full-text search on the search condition learning document d read in step S301 based on the keyword included in the search condition data read in step S302 (step S303).

  Here, this full-text search method is, for example, a method for searching whether or not a word matching the keyword is included in the search condition learning document, or the keyword appears in the search condition learning document. There are various methods for detecting the frequency, and the keyword inclusion paragraph determination unit 108 performs a full-text search of the search condition learning document using the keywords using a known full-text search technique. As a full-text search document, Japanese Patent Laid-Open No. 8-44771 discloses a technique. Then, the keyword inclusion paragraph determination unit 108 stores the sentence line including the keyword in the search condition learning document d as a result of the full-text search, and then performs the same processing method as the processing from step S102 to step S109 described above. The determination of each paragraph of the search condition learning document d and the determination of the paragraph type of each paragraph are performed to detect the document structure of the search condition learning document d (step S304). Further, as a result of the full-text search, the keyword inclusion paragraph determination unit 108 determines the paragraph of the search condition learning document d including the sentence line including the keyword in the search condition learning document d as the keyword inclusion paragraph (step S305). Then, the keyword inclusion paragraph determination unit 108 reads the search condition learning document d read in step S301, the keyword inclusion paragraph in the search condition learning document d, the document structure information of the search condition learning document d, and step S302. The new sample document candidate determination unit 109 is notified of the document structure number held in the search condition data read in step.

  Next, the new sample document candidate determination unit 109 uses the information on the document structure of the search condition learning document d received from the keyword inclusion paragraph determination unit 108 and the keyword inclusion paragraph in the search condition learning document d. The paragraph position (upper, middle, lower) of the keyword-containing paragraph is detected (step S306). The detection of the paragraph position of the keyword-containing paragraph is the same as the processing in step S208 described above. Then, the new sample document candidate determination unit 109 holds the document structure number indicating the document structure of the search condition learning document d and the paragraph position of the keyword inclusion paragraph in the search condition learning document d in the search condition data storage unit 107. Whether there is search condition data to be searched is searched (step S307). In step S307, the new sample document candidate determination unit 109 stores search condition data that holds the document structure number indicating the document structure of the search condition learning document d and the paragraph position of the keyword-containing paragraph in the search condition learning document d. If not stored in the search condition data storage unit 107, the search condition learning document d including the keyword inclusion paragraph in the paragraph is recorded as a new sample document candidate in the new sample document candidate storage unit 110 (step S308). In this way, among the search condition learning documents recorded in the search condition learning document storage unit 101, a search condition that holds the same content as the content of the search content description paragraph of the sample document prepared by the user The learning document is selected as a new sample document and recorded in the new sample document candidate storage unit 110.

  In step S307, search condition data holding the document structure number indicating the document structure of the search condition learning document d and the paragraph position of the keyword-containing paragraph in the search condition learning document d is recorded in the search condition data storage unit 107. If so, the new sample document candidate determination unit 109 notifies the keyword inclusion paragraph determination unit 108 to process another search condition data. Then, the keyword inclusion paragraph determination unit 108 checks whether or not there is data other than the search condition data read in step S302 in the search condition data storage unit 107 (step S309). If another search condition data is recorded, the search condition data is used to repeat the processing from step S302 to step S306 for the search condition learning document d.

Next, when the data of the search condition learning document candidate new sample document candidate is recorded in the search condition learning document storage unit 101 in step S307, the keyword inclusion paragraph determination unit 108 stores the next search condition learning document. Instruct to perform processing. Then, the keyword inclusion paragraph determination unit 108 checks whether or not other search condition learning documents are recorded in the search condition learning document storage unit 101 (step S310), and other search condition learning documents are recorded. If yes, the processing after the processing in step S301 is started for the document. Then, the processing from step S301 to step S308 is performed for all sample documents recorded in the search condition learning document 101.
Next, the new sample document designating unit 111 displays the new sample document candidates recorded in the new sample document candidate storage unit 110 on a monitor, for example, and accepts selection of a document to be a new sample document from the user. The new sample document 111 registers a new sample document designated by the user in the sample document storage unit 102. Then, the document search apparatus 1 creates search condition data for the new sample document by using the new sample document by the processes in steps S201 to S209. Note that after the processing of steps S201 to S209 is performed on the new sample document to create the search condition data, the new condition data creation unit 106 starts the process of determining a new sample document candidate in the keyword inclusion paragraph determination unit 108. Is not instructed, and the process is terminated.

Next, a processing flow for searching for a search target document desired by the user from the search condition database will be described. FIG. 10 is a diagram showing a processing flow when searching for a search target document.
When the search condition data creating unit 106 creates search condition data for all sample documents and new sample documents, the search condition data creating unit 106 next instructs the search target document reading unit 112 to perform the search main process. Then, the search target document reading unit 112 reads one search target document data from the search target document database 2 via the communication network (step S 401), and transfers it to the paragraph analysis unit 103. Next, the paragraph analysis unit 103 determines each paragraph of the search target document and determines the paragraph type of each paragraph using the same processing technique as the processing from step S102 to step S109 described above. The document structure of the search target document is detected (step S402). Then, the paragraph analysis unit 103 notifies the search processing unit 113 of the document structure information of the search target document detected in step S402.

  Next, the search processing unit 113 reads a document structure number indicating the same document structure information as the document structure information of the search target document from the document structure storage unit 105 (FIG. 4) (step S403). Here, it is assumed that the document structure number indicating the document structure of the search target document is “1”. Next, the search processing unit 113 reads the search condition data holding the document structure number read in step S403 from the search condition data storage unit 107 (step S404). In addition, the search processing unit 113 detects in the upper, middle, and lower paragraph positions each paragraph of the search target document in the same manner as the processing in step S208 described above. Then, the search processing unit 113 extracts each paragraph of the search target document at the same paragraph position as the paragraph position held in the search condition data read in step S404 (step S405). In addition, the search processing unit 113 checks whether or not the paragraph type of the paragraph of the search target document read in step S405 is the same as the paragraph type held in the search condition data read in step S404 (step S406). In step S406, if the paragraph type of the paragraph to be searched read in step S405 is the same as the paragraph type held in the search condition data read in step S404, the search processing unit 113 selects the search reference document. The paragraph is determined as a search determination paragraph (step S407).

  Next, the search processing unit 113 performs a full text search using the keyword held in the search condition data read in step S404 for the search determination paragraph in the search target document (step S408). This full-text search method uses a conventional method as described in the description of step S303. Then, the search processing unit 113 performs a full text search of the search determination paragraph, and a word of a sentence described in the search determination paragraph matches a keyword, or a sentence having a high keyword appearance frequency is included in the search determination paragraph. It is confirmed whether it has been (step S409). The processing in step S409 differs depending on the full text search method. Then, as a result of the full-text search, when the word of the sentence described in the search determination paragraph matches the keyword, the search processing unit 113 searches the search target document holding the search determination paragraph as a search result. Is transferred to the search result output unit 114 (step S410), and the search target document transferred by the search result output unit 114 is output as a search result to, for example, a monitor provided in the document search apparatus 1 (step S411).

  In step S406, if the paragraph type of the search target paragraph read in step S405 is not the same as the paragraph type held in the search condition data read in step S404, the search processing unit 113 selects another search target document. The search target document reading unit 112 is notified to read from the search target document database 2. Further, in step S409, when the word of the sentence described in the search determination paragraph does not match the keyword, the search processing unit 113 reads the search target document so that the search target document is read from the search target document database 2. Notification to the unit 112. Even when the search result is output in step S410, the search processing unit 113 notifies the search target document reading unit 112 to read another search target document from the search target document database 2. Then, the document search apparatus 1 performs a search process for all the large numbers of search target documents recorded in the search target document database 2. As a result, the user can obtain a search target document having a desired content.

  1 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read and executed by a computer system, whereby the document search apparatus 1 is recorded. May perform each of the processes described above. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is a block diagram which shows the structure of the document search apparatus by embodiment of this invention. It is a figure which shows the example of the document for search condition learning. It is a figure which shows a sample document. It is a figure which shows the information of a document structure. It is a figure which shows the example of search condition data. It is a figure which shows the processing flow at the time of recording the classification | category of a document structure, and its information. It is a figure which shows the data hold | maintained at the memory of the paragraph analysis part. It is a figure which shows the flow of search condition data creation processing. It is a figure which shows the processing flow of new sample document preparation. It is a figure which shows the processing flow at the time of the search of a search object document.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Document search apparatus, 2 ... Search object document database, 101 ... Document storage part for search condition learning, 102 ... Sample document storage part, 103 ... Paragraph analysis part, 104 ... Document structure analysis unit 105 ... Document structure storage unit 106 ... Search condition data creation unit 107 ... Search condition data storage unit 108 ... Keyword inclusion paragraph determination unit 109 ... New sample Document determination unit, 110 ... New sample document candidate storage unit, 111 ... New sample document designation unit, 112 ... Search target document reading unit, 113 ... Search processing unit, 114 ... Output of search results Part

Claims (7)

A document search device for searching a search target document similar to a sample document from a plurality of search target documents,
First paragraph dividing means for dividing each of the plurality of representative search target documents extracted from the plurality of search target documents;
Document structure classification means for classifying the document structure of each of the representative search target documents based on the paragraph classification of each of the representative search target documents;
Document structure storage means for storing a plurality of information on the classified document structure for each type of the document structure;
Second paragraph dividing means for dividing the sample document into paragraphs;
Sample document structure detection means for detecting the document structure type of the sample document based on the result of the paragraph division of the sample document and the information of the document structure stored in the document structure storage means;
A third paragraph dividing means for dividing a paragraph about an arbitrary search target document among the plurality of search target documents;
A search target document for detecting a document structure type of the arbitrary search target document based on a result of the paragraph division by the third paragraph division unit and information on the document structure stored in the document structure storage unit Structure detection means;
A search target document determining unit that sets the arbitrary search target document having the same document structure type as the sample document as the search target document similar to the sample document;
A document search apparatus comprising: A document search device for searching a search target document similar to a sample document from a plurality of search target documents,
First paragraph dividing means for dividing each of a plurality of representative search target documents extracted from the plurality of search target documents at a predetermined ratio;
Document structure classification means for classifying the document structure of each of the representative search target documents based on the paragraph classification of each of the representative search target documents;
Document structure storage means for storing a plurality of information on the classified document structure for each type of the document structure;
Second paragraph dividing means for dividing the sample document into paragraphs;
Sample document structure detection means for detecting the document structure type of the sample document based on the result of the paragraph division of the sample document and the information of the document structure stored in the document structure storage means;
The paragraph position of the search content description paragraph specified by the user in the sample document, the keyword included in the search content description paragraph, the paragraph type indicating the description content in the search content description paragraph, and the document structure of the sample document Search condition data creating means for creating search condition data that holds the type,
Search condition data storage means for storing the search condition data;
A third paragraph dividing means for dividing a paragraph about an arbitrary search target document among the plurality of search target documents;
A search target document for detecting a document structure type of the arbitrary search target document based on a result of the paragraph division by the third paragraph division unit and information on the document structure stored in the document structure storage unit Structure detection means;
Search condition data for reading the search condition data from the search condition data storage means when the search condition data holding the type of document structure of the arbitrary search target document is recorded in the search condition data storage means Reading means;
Paragraph specifying means for specifying a search determination paragraph in the arbitrary search target document based on the position of the search content description paragraph held in the read search condition data and the paragraph type;
The full text search of the paragraph containing the search content is performed using a keyword held in the search condition data read by the search condition data reading means, and the arbitrary search target document is output as a search result based on the result of the full text search Search means for determining whether to do,
A document search apparatus comprising:   3. The document search apparatus according to claim 2, wherein the keyword is a word having a predetermined part of speech obtained by morphological analysis of a sentence in the search content description paragraph. A keyword inclusion paragraph determining means for determining a keyword inclusion paragraph in which the keyword is included in the representative search target document based on a result of a full text search of the representative search target document using the keyword in the search content description paragraph;
The combination of the type of document structure of the representative search target document and the paragraph position in the representative search target document of the keyword inclusion paragraph, and the document structure held by the search condition data stored in the search condition data recording means New sample document candidate determination means for determining the representative search target document holding the keyword-inclusive paragraph as a new sample document candidate based on a comparison of the combination of the type and the paragraph position of the search content description paragraph;
The document search apparatus according to claim 2, further comprising: Line type determining means for determining the line type of each line in the document of either the sample document or the representative search target document based on the description content in the line;
Paragraph start position determining means for determining a paragraph start position in either the sample document or the representative search target document based on the line type;
Paragraph determining means for determining, from the sample document or the representative search target document, one paragraph from the line at the paragraph head position to the line before the line at the next paragraph head position;
Paragraph type determining means for determining the paragraph type based on the line type of a line included in the paragraph;
5. The document search apparatus according to claim 2, further comprising: A document search method of a document search device for searching a search target document similar to a sample document from a plurality of search target documents,
A first paragraph dividing process of dividing each of the plurality of representative search target documents extracted from the plurality of search target documents at a predetermined ratio;
A document structure classification process for classifying the document structure of each of the representative search target documents based on the paragraph classification of each of the representative search target documents;
A document structure recording step of recording the classified document structure information in a document structure storage unit for each type of the document structure;
A second paragraphing process for dividing the sample document into paragraphs;
A sample document structure detection step of detecting a document structure type of the sample document based on the result of the paragraph division of the sample document and the information of the document structure stored in the document structure storage unit;
The paragraph position of the search content description paragraph specified by the user in the sample document, the keyword included in the search content description paragraph, the paragraph type indicating the description content in the search content description paragraph, and the document structure of the sample document Search condition data creation process for creating search condition data that holds the type,
A search condition data recording process for recording the search condition data in a search condition data storage means;
A third paragraph dividing process of dividing a plurality of search target documents into arbitrary search target documents;
A search target document for detecting a document structure type of the arbitrary search target document based on a result of the paragraph division by the third paragraph division process and information on the document structure stored in the document structure storage unit Structure detection process;
Search condition data for reading the search condition data from the search condition data storage means when the search condition data holding the type of document structure of the arbitrary search target document is recorded in the search condition data storage means Reading process,
A paragraph specifying process for specifying a search determination paragraph in the arbitrary search target document based on the position of the search content description paragraph held in the read search condition data and the paragraph type;
The full text search of the paragraph containing the search content is performed using a keyword held in the search condition data read in the search condition data reading process, and the arbitrary search target document is output as a search result based on the result of the full text search A search process for determining whether to
A document search method characterized by comprising: A program for causing a computer of a document search device to search for a search target document similar to a sample document from a plurality of search target documents,
A first paragraph dividing process for dividing each of the plurality of representative search target documents extracted from the plurality of search target documents at a predetermined ratio;
A document structure classification process for classifying the document structure of each of the representative search target documents based on the paragraph classification of each of the representative search target documents;
A document structure recording process for recording the classified document structure information in a document structure storage unit for each type of the document structure;
A second paragraph dividing process for dividing the sample document into paragraphs;
A sample document structure detection process for detecting a document structure type of the sample document based on the result of the paragraph division of the sample document and the information of the document structure stored in the document structure storage unit;
The paragraph position of the search content description paragraph specified by the user in the sample document, the keyword included in the search content description paragraph, the paragraph type indicating the description content in the search content description paragraph, and the document structure of the sample document Search condition data creation processing for creating search condition data that holds the type,
A search condition data recording process for recording the search condition data in a search condition data storage means;
A third paragraph-separating process for paragraph-separating an arbitrary search target document among the plurality of search target documents;
A search target document for detecting a document structure type of the arbitrary search target document based on the result of the paragraph division by the third paragraph division processing and the information of the document structure stored in the document structure storage unit Structure detection processing;
Search condition data for reading the search condition data from the search condition data storage means when the search condition data holding the type of document structure of the arbitrary search target document is recorded in the search condition data storage means Reading process,
Paragraph specifying processing for specifying a search determination paragraph in the arbitrary search target document based on the position and paragraph type of the search content description paragraph held in the read search condition data;
A full text search of the paragraph containing the search content is performed using a keyword held in the search condition data read in the search condition data reading process, and the arbitrary search target document is output as a search result based on the result of the full text search Search processing for determining whether or not to perform,
A program that causes a computer to execute.

Images