JP2009157458A - Index creation device, its method, program, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To retrieve a document with high ranking precision even when link information between pages is little, and even when the quantity of content written in a page is poor. <P>SOLUTION: An index creation device is configured to; detect a query word belonging to one word class by a query word detection means 6 from each correct answer document in a correct answer document group DB1 and each incorrect answer document in an incorrect answer document group DB2; measure the appearance frequency of an optional word appearing in the neighborhood of the detected query word in each document by a neighboring word frequency measuring part 7; calculate the significance of the optional word in one word class to which the query word is belonging by a neighboring word significance calculation means 8; calculate scores for the word belonging to one word class included in each retrieval object document in a retrieval object document DB3 by using the significance of the optional word; and generate an index with the scores corresponding to the combination of the respective words belonging to each retrieval object document and one word class by an index generation means 10. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a technique for searching a document that matches a query (a processing request represented by a character string) from a set of documents and creating an index for ranking in the order of matching.

Conventionally, various methods have been proposed as a method of searching documents that match a query from a set of documents and ranking (ranking) them in the order of matching. For example, a vector space model is typical as a method for expressing the contents of a document (see, for example, Non-Patent Documents 1 and 2), and a page rank is famous as a method for considering link information between pages (for example, non-patent documents). (See Patent Document 3).
Takenobu Tokunaga, “Language and Computation 5 Information Retrieval and Language Processing”, The University of Tokyo Press, 1999, p. 154-159 Kita Kenji, Tsuda Kazuhiko, Sasabori Masatomi, “Information Retrieval Algorithm”, Kyoritsu Publishing Co., Ltd., 2002, p. 60-65 Hayato Yamana, Hidekazu Kondo, Commentary Search Engine Google, “Information Processing”, Information Processing Society of Japan, 2001, Vol. 42, No. 8, p. 775-780

  However, with the conventional document search method, high-accuracy ranking can be performed when there is a lot of link information between pages or when the amount of content described on the page is abundant. There is a problem that the ranking accuracy is not good when the content described in the above is insufficient.

  An object of the present invention is to determine whether a document to be searched is based on a set of documents that have been determined whether or not they conform to a certain kind of pre-sampled query, in other words, whether or not to be positioned higher in the search ranking. By creating an index for a set, it is possible to improve the accuracy of ranking when searching for the same kind of queries, especially when there is little link information between pages or the amount of content described on the page is scarce Even in such a case, it is to enable ranking of search with high accuracy.

  An index creation apparatus according to the present invention is an apparatus for retrieving documents that match a query from a set of search target documents, and creating an index for ranking in the order of matching. A correct document set database that stores a set of query words that belong to one word class and a set of correct documents that match each query word in the set; a set of query words that belong to the one word class; Stored in the incorrect document set database that stores a set of incorrect documents that do not match each query word in the set, the search target document database that stores the set of search target documents, and the correct document set database From the correct answer documents stored in the correct answer document and incorrect answer document set database, the correct answer document set database is obtained. Query word detecting means for detecting a word that matches the query word stored in the incorrect document set database, each correct document stored in the correct document set database, and each stored in the incorrect document set database A neighboring word frequency measuring unit that measures, for each arbitrary word, an appearance frequency of an arbitrary word that appears in the vicinity of a word that matches the query word in each document detected by the query word detecting unit in an incorrect answer document; , The frequency of occurrence of any word that appears in the vicinity of a word that matches the query word in the correct document, and any word that appears in the vicinity of the word that matches the query word in the incorrect answer document, measured by the neighborhood word frequency measurement means The importance of the arbitrary word in the one word class to which the query word belongs is calculated for each arbitrary word from the appearance frequency of A query word calculated by the neighborhood word importance calculating means and a word belonging to the one word class from the search target document stored in the search target document database together with words in the vicinity thereof, and calculated by the neighborhood word importance calculating means Of the importance of the arbitrary word in the one word class to which the word belongs, the score of the search target document with respect to the word belonging to the one word class using the importance of the arbitrary word that matches the neighboring word This is repeated for all the words belonging to the one word class included in each search target document, and the score corresponding to the combination of each search target document and each word belonging to the one word class And an index generating means for generating an index.

  The index creation method of the present invention is a method of creating an index for searching documents that match a query from a set of search target documents and ranking them in the order of matching, and includes a word class that represents some classification related to words. A correct document set database that stores a set of query words belonging to one of the word classes and a set of correct answer documents that match each query word in the set, and a set of query words that belong to the one word class Query word detecting means using a wrong document set database that stores a set of incorrect documents that do not match each query word in the set and a search target document database that stores a set of search target documents. , Each correct document stored in the correct document set database and each incorrect stored in the incorrect document set database Detecting a word that matches a query word stored in the correct document set database and the incorrect document set database from the document, and a neighborhood word frequency measuring means, each correct document stored in the correct document set database, In each incorrect answer document stored in the incorrect answer document set database, the appearance frequency of an arbitrary word appearing in the vicinity of a word that matches the query word in each document detected in the query word detection step is The step of measuring for each word, and the neighborhood word importance calculation means measure the appearance frequency and neighborhood word frequency of any word that appears in the vicinity of the word that matches the query word in the correct word measured in the neighborhood word frequency measurement step. Frequency of occurrence of any word that appears in the vicinity of a word that matches the query word in the incorrect document measured in step Calculating the importance of the arbitrary word in the one word class to which the query word belongs, for each arbitrary word, and an index generating means from the search target document stored in the search target document database, A word belonging to one word class is extracted together with its neighboring words, and among the importance of the arbitrary word in the one word class to which the query word calculated in the neighborhood word importance calculating step belongs, the neighboring word The score of the document for the word belonging to the one word class is calculated using the importance of the arbitrary word that matches the word, and this is used for all the words belonging to the one word class included in each search target document The index is formed with a score corresponding to a combination of each search target document and each word belonging to the one word class as an element. Generating a process.

  According to the present invention, the importance of neighboring words of query words existing in correct and incorrect documents is learned, and an instance of the same word class as the query word having learned the importance of neighboring words for the search target document ( By storing the score of the word) in advance, when an instance of the word class is searched as a query word, a highly accurate ranking can be realized even when the link information is small or the content of the document is small. Is possible. For example, when a query word belongs to a person name class, the importance level of the neighboring word of the person name class is learned from the correct answer document prepared when the query word is a person name instance, the person name instance of the incorrect answer document, and the neighbor word. In addition, for the set of search target documents, the learned importance level is reflected in the score of the personal name instance for the neighborhood words of the personal name class existing in the document, and a personal name document matrix storing this is created, When a query word is an instance having a personal name class as a search request, a document is output in the order from the highest score of the personal name instance, whereby a highly accurate search ranking can be realized. The present invention is particularly effective for a search request when there is only one query word.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows an example of an embodiment of an index creation apparatus according to the present invention. The index creation apparatus according to the present embodiment includes a correct document database (DB) 1, an incorrect document database (DB) 2, and a search. Target document database (DB) 3, morphological analysis unit 4, word class identification unit 5, query word detection unit 6, neighborhood word frequency measurement unit 7, neighborhood word importance calculation unit 8, neighborhood word importance It consists of a table 9 and an index generation unit 10.

  The correct document DB 1 is a word (hereinafter referred to as a query word) for searching a document that matches a query for various queries (hereinafter referred to as a query word), such as a specific expression such as a person name or a place name, a specific field, A set of query words belonging to one of the word classes representing a certain classification of words, such as a genre name indicating a purpose, and a document that matches each query word in the set of query words (any one And a set of correct documents that are at least one query word and are appropriate documents as a search result of the query word.

  The incorrect answer document DB 2 includes a set of query words belonging to the one word class and a document that does not match each query word in the set (including at least one of the query words. As a result, it stores a set of incorrect documents that are inappropriate documents).

  Note that the correct answer document and the incorrect answer document described above are acquired by manually determining in advance whether the document is included in an arbitrary large-scale text corpus. Further, instead of the above-mentioned incorrect answer document, a document other than the correct answer document may be stored in the incorrect answer document DB 2 as an incorrect answer document.

  The search target document DB 3 stores a set of search target documents.

  The morpheme analysis unit 4 divides the input document into morphemes (words), and gives word information such as notation, reading, part of speech, etc. to each word, and a word string with word information (hereinafter simply referred to as a word string). ) Is output.

  The word class identification unit 5 discriminates (identifies) the word class of each word in the input word string from a set of prepared word classes, here, a set of word classes to which each query word belongs. , An identifier representing the word class is assigned and output. As a specific word class identification method, a method based on a statistical model previously learned from data, which is used for well-known specific expression extraction (for example, Japanese Patent No. 4005477 (Japanese Patent Application No. 2002-317435)). Reference) can be used.

  The query word detection unit 6 stores the correct answer document stored in the correct answer document set DB1 and the incorrect answer document stored in the incorrect answer document set DB2 and stored in the correct answer document set DB1 and the incorrect answer document set DB2. Find the words that match the query words you have.

  The neighborhood word frequency measurement unit 7 is configured to determine whether each of the correct documents stored in the correct document set DB1 and the incorrect answer documents stored in the incorrect answer document set DB2 includes each document The frequency of appearance of an arbitrary word that appears in the vicinity of a word that matches the query word is measured for each arbitrary word.

  The neighborhood word importance calculation unit 8 matches the appearance frequency of an arbitrary word appearing in the vicinity of a word that matches the query word in the correct document and the query word in the incorrect answer document measured by the neighborhood word frequency measurement unit 7. The importance of the arbitrary word in the one word class to which the query word belongs is calculated for each arbitrary word from the appearance frequency of the arbitrary word appearing in the vicinity of the word.

  The neighborhood word importance level table 9 includes the appearance frequency of an arbitrary word that appears in the vicinity of a word that matches the query word in the correct word measured by the neighborhood word frequency measurement unit 7 and the error measured by the neighborhood word frequency measurement unit 7. Stores the appearance frequency of an arbitrary word that appears in the vicinity of a word that matches the query word in the correct answer document, and also stores the arbitrary word in the one word class to which the query word calculated by the neighborhood word importance calculation unit 8 belongs. Stores the importance of.

  The index generation unit 10 extracts the words belonging to the one word class from the search target document stored in the search target document DB 3 together with the words in the vicinity thereof, and calculates the query word calculated by the neighborhood word importance calculation unit 8. Of the importance of the arbitrary word in the one word class to which it belongs, the score of the search target document for the word belonging to the one word class is calculated using the importance of the arbitrary word that matches the neighboring word. Calculating, repeating this for all the words belonging to the one word class included in each search target document, and taking the score corresponding to the combination of each search target document and each word belonging to the one word class as an element Generate an index to

  2 to 4 show an overview of index creation processing according to the present invention, and here, an overview of processing when a query is an instance (query word) belonging to a personal name class.

  That is, first, among documents including various query words Qi (i = 1, 2,...) (Person names), a query is performed for a set of correct documents A that are preferably ranked higher and a set of incorrect documents B that are not desired to be ranked higher. Prepared for each word Qi, and the occurrence frequency of arbitrary words W1, W2,... Wn appearing in the vicinity of one query word Qi in the document is counted for the correct document A and the incorrect document B, and this is counted for each query word. Repeat similarly for Qi, and further add this appearance frequency for each arbitrary word W1, W2,... Wn to learn (calculate) the importance of words W1, W2,... Wn with respect to words belonging to the personal name class (Step 1). .

  Next, the words P1, P2,... Pm belonging to the personal name class are extracted from the search target document together with the neighboring words, and based on the importance of the learned words W1, W2,. The score of the search target document for the words P1, P2,... Pm is calculated, and this is repeated for each search target document D1, D2,... Do to generate a personal name document matrix (index) (Step 2).

  The above processing is performed offline in advance, and when a query to an arbitrary search engine executed online is an instance of a person name class, the score of that instance in the person name document matrix is reflected in the search ranking. Can do. In particular, since (1) information of words in the vicinity of the query is used, it is effective even for a document set with little link information. (2) Since only a nearby word is required, it is effective even for a document with a small content.

  Hereinafter, the details of the processing in each unit described above will be described with examples.

[Query word detection unit 6]
The query word detection unit 6 detects, for each correct document stored in the correct document DB1, the appearance position of a word that matches the query word stored in the correct document DB1 in the document. For each incorrect answer document stored in the incorrect answer document DB 2, the appearance position of the word that matches the query word also stored in the incorrect answer document DB 2 is detected.

  At this time, the morphological analysis unit 4 and the word class identification unit 5 are used to detect a word that matches the query word. For example, when the information that the query word stored in the correct document DB1 or the incorrect document DB2 belongs to the personal name class is given in advance, the query word detection unit 3 selects the correct document stored in the correct document DB1 or The word class of each word of the incorrect document stored in the incorrect document DB 2 is identified using the morphological analysis unit 4 and the word class identification unit 5, and an identifier of the personal name class is given from the correct document or the incorrect document. The extracted word (person name instance) is extracted, and the appearance position in the document of the word that matches the query word is detected from the set of the extracted person name instances.

  FIG. 5 shows an example of a learning model in the word class identification unit 5, and here shows an example in which the appearance frequencies of two consecutive morphemes are stored.

  FIG. 6 shows an example of a word string after performing morphological analysis and word class identification on the document. Here, for the document “Mr. Hasegawa, Yokosuka Bank President, is really a good person”. Perform morphological analysis to add notation, reading, and part of speech to each word, and further perform word class identification (here, specific expression extraction as word class identification in a narrow sense) to identify an identifier (here, a specific expression tag) An example is given. At this time, if the information that the query word belongs to the personal name class is given, the word “Hasegawa” whose personal expression tag is the personal name class (<PSN>) is extracted.

  It should be noted that detection of a word that matches the query word may take a method such as simply using collation with a character string, and it is not always necessary to use the morpheme analysis unit 4 and the word class identification unit 5. .

[Neighboring word frequency measurement unit 7]
The neighborhood word frequency measurement unit 7 is a neighborhood of the appearance position in the document of the word that matches the query word in each correct document detected by the query word detection unit 6 in each correct document stored in the correct answer DB 1. Any word that appears in a window size defined as a pre-neighbor before and after the appearance position, for example (if the window size defined as a neighborhood is 5, for example, matches the query word) Occurrence frequency of 5 arbitrary words before and after the word) is counted for each arbitrary word, and each incorrect word stored in the incorrect word document DB 2 is detected by the query word detection unit 6. The frequency of occurrence of any word that appears in the same vicinity of the appearance position of the word that matches the query word in the incorrect document in the document is counted for each arbitrary word, and these are converted into the correct document. And stores in the vicinity word importance table 9 with kicking the total number of said any word in the arbitrary total number of words and incorrect document.

  It should be noted that the words for which the appearance frequency is counted are selected using information such as part of speech provided by the morphological analysis unit 4, for example, words whose part of speech corresponds to a functional word such as a particle or auxiliary verb may be excluded. At this time, if the morphological analysis unit 4 is not used, a list storing a set of words whose importance is to be calculated is prepared in advance, and the list is placed near a word that matches the query word detected by the character string matching. In the same manner as described above, words corresponding to function words such as particles and auxiliary verbs may be excluded by checking whether the stored word is included by character string matching.

[Neighboring word importance calculator 8]
The neighborhood word importance degree calculation unit 8 measures the appearance frequency of an arbitrary word appearing in the vicinity of a word that matches the query word in the correct document, which is measured by the neighborhood word frequency measuring means 7 and stored in the neighborhood word importance degree table 9 ( Correct neighborhood frequency), the appearance frequency of any word that appears in the vicinity of a word that matches the query word in the incorrect document (incorrect neighborhood frequency), the total number of the arbitrary word in the correct document (total number of correct neighborhood words) and The importance level of the arbitrary word in the one word class to which the query word belongs is calculated for each arbitrary word from the total number of arbitrary words in the correct answer document (total number of incorrect neighboring words), and the neighborhood word importance level table 9 is stored.

  FIG. 7 shows an example of the neighborhood word importance degree table, in which correct answer frequencies, incorrect answer neighborhood frequencies, correct answer neighborhood words total number, incorrect answer neighborhood words total number, and importance levels for arbitrary words W1, W2,. ing.

Here, the importance, for example, C is the correct neighborhood frequency (E), incorrect answer neighborhood frequency (F), correct answer neighborhood word count (G), and incorrect answer neighborhood word count (for each arbitrary word W1, W2,. H)
C = ln [(E / G) / {(E + F) / (G + H)}] (1)
Can be obtained. Note that the importance may take a negative value. Also, the above formula (1) is an example, and as a result of comparing the correct answer document and the incorrect answer document, if the word that has a high ratio of frequent occurrence in the correct answer document has a large importance, It is not limited.

[Index generation unit 10]
The index generation unit 10 selects a word belonging to the one word class from a search target document stored in the search target document DB 3, and more precisely, a word in the vicinity thereof, more precisely, from a position of a word belonging to the one word class. Extracted with words that appear within a predefined window size (for example, if the window size defined as the neighborhood is 3, any three words before and after a word belonging to one word class) Of the importance of the arbitrary word in the one word class to which the query word belongs, which is calculated by the degree calculation unit 8 and stored in the neighborhood word importance table 9, of the arbitrary word that matches the neighboring word The score of the search target document for the words belonging to the one word class is calculated using the importance, and this is used for all the one words included in each search target document. Repeating the words belonging to the class, to generate an index to a score corresponding to the combination of each word belonging to said one word class of each search target document element.

  At this time, the morphological analysis unit 4 and the word class identification unit 5 are used to extract words belonging to one word class. That is, the index generation unit 10 identifies the word class of each word of the search target document stored in the search target document DB 3 by using the morphological analysis unit 4 and the word class identification unit 5, and This is done by extracting words to which an identifier of one word class is assigned.

Here, the score (S (d, q)) of the document (d) including the instance (q) that can be a word (query word) belonging to one word class is the neighborhood word (Wj) of each instance (qi). Add the in-document frequency (TF (q)) of the instance of the one word class with a weight to the sum of importance (C (qi, Wj)),
S (d, q) = αΣΣC (qi, Wj) + βTF (q) (2)
Can be obtained as

  The calculated score is stored as a matrix element between each instance of the one word class and each search target document. For example, when the word class is a personal name class, a personal name document matrix is created around the personal name instance and the search target document, and the score of the personal name instance included in each search target document is stored in each element.

  FIG. 8 shows an example of an index, here an example of a personal name document matrix. For the search target documents D1 to Do, the scores of the person names P1 to Pm are stored as person name instances included in the respective documents. If there is no personal name instance in a certain search target document, the score of the personal name instance in the document is 0.

  FIG. 9 shows an example of a detailed processing flow in the index generation unit 10, morphological analysis and specific expression extraction are applied to the search target document, and a specific expression tag representing a specific expression category is given to the word string. It shall be.

  First, in step s1, a pointer is set at the head of the document to be searched, and in step s2, step s12, and step s13, the word pointed to by the pointer is sequentially acquired until the pointer reaches the end of the sentence.

  In step s3, it is determined whether or not the acquired word is a word constituting a personal name instance Pk (k = 1, 2,... M), and if so, to reflect that the personal name instance Pk has appeared, In step s4, 1 is added to the score of the search target document for the person name instance Pk. Note that step s4 may be omitted.

  In steps s5 to s11, for the N words before and after the personal name instance Pk, if there is a word that matches an arbitrary word whose importance is calculated by the neighborhood word importance calculator 8, the importance is assigned to the word It adds to the score of the said search object document with respect to person name instance Pk.

  Finally, in step s14, the score of the search target document is output for all the personal name instances P1 to Pm in the search target document.

  By performing the processing shown in FIG. 9, the score of the document for each personal name instance Pm in the search target document can be obtained, and as a result, a personal name document matrix can be obtained.

  In the above-described embodiment, a set of query words belonging to another word class and a set of correct documents that match each query word in the set are stored in the correct document DB1, and the incorrect document DB2 stores the above-described correct answer document DB2. If a set of query words belonging to another word class and a set of incorrect documents that do not match each query word in the set are stored, each search target document and each word belonging to the other word class Needless to say, an index having a score corresponding to the combination as an element can be generated.

  The present invention can also be realized by installing a program for realizing the functions shown in the configuration diagram of FIG. 1 via a recording medium or a communication line in a known computer.

The block diagram which shows an example of embodiment of the index production apparatus of this invention Explanatory drawing which shows the outline | summary of the index creation process of this invention Explanatory drawing which shows the detail of Step1 in FIG. Explanatory drawing which shows the detail of Step2 in FIG. Explanatory drawing which shows an example of the learning model in a word class identification part Explanatory drawing which shows an example of the word string after performing morphological analysis and word class identification Explanatory drawing which shows an example of a neighborhood word importance degree table Explanatory drawing which shows an example of an index The flowchart which shows an example of the flow of a process of an index production | generation part

Explanation of symbols

  1: correct document database, 2: incorrect document database, 3: search object document database, 4: morpheme analysis unit, 5: word class identification unit, 6: query word detection unit, 7: neighborhood word frequency measurement unit, 8: Neighborhood word importance calculator 9: Neighborhood word importance table 10: Index generator

Claims (5)

A device that searches a document that matches a query from a set of search target documents and creates an index for ranking in order of matching,
A correct document set database that stores a set of query words belonging to one of the word classes representing a certain class of words and a set of correct documents that match each query word in the set;
An incorrect answer document set database for storing a set of query words belonging to the one word class and an incorrect answer document set that does not match each query word in the set;
A search target document database for storing a set of search target documents;
A word that matches a query word stored in the correct document set database and the incorrect document set database from each correct document stored in the correct document set database and each incorrect document stored in the incorrect document set database Query word detection means for detecting
In each correct answer document stored in the correct answer document set database and each incorrect answer document stored in the incorrect answer document set database, the vicinity of a word that matches the query word in each document detected by the query word detecting means Near word frequency measuring means for measuring the frequency of appearance of any word appearing in each arbitrary word;
The frequency of occurrence of an arbitrary word that appears in the vicinity of a word that matches the query word in the correct answer document and the word that appears in the vicinity of the word that matches the query word in the incorrect answer document, measured by the neighborhood word frequency measuring means. From the appearance frequency, neighborhood word importance calculating means for calculating the importance of the arbitrary word in the one word class to which the query word belongs for each arbitrary word;
A word belonging to the one word class is extracted from the search target document stored in the search target document database together with words in the vicinity thereof, and the query word calculated by the neighborhood word importance calculating unit is included in the one word class to which the query word belongs. Of the importance levels of the arbitrary words, the score of the search target document for the words belonging to the one word class is calculated using the importance levels of the arbitrary words that match the neighboring words. Index generation that repeats for all the words belonging to the one word class included in the target document and generates an index having a score corresponding to a combination of each search target document and each word belonging to the one word class as an element Means for creating an index. A method for searching documents that match a query from a set of search target documents and creating an index for ranking in the order of matching,
A correct document set database that stores a set of query words belonging to one of the word classes representing a certain class of words and a set of correct documents that match each query word in the set;
An incorrect answer document set database for storing a set of query words belonging to the one word class and an incorrect answer document set that does not match each query word in the set;
Using a search target document database that stores a set of search target documents,
The query word detecting means is stored in the correct document set database and the incorrect document set database from each correct document stored in the correct document set database and each incorrect document stored in the incorrect document set database. Detecting a word that matches the query word;
The query in each document detected by the query word detection step in each correct answer document stored in the correct answer document set database and each incorrect answer document stored in the incorrect answer document set database Measuring the appearance frequency of an arbitrary word appearing in the vicinity of a word that matches the word for each arbitrary word;
Incorrect document measured by the neighborhood word importance calculation means by the appearance frequency of an arbitrary word appearing in the vicinity of a word that matches the query word in the correct word document measured by the neighborhood word frequency measurement step and the neighborhood word frequency measurement step Calculating the importance of the arbitrary word in the one word class to which the query word belongs for each arbitrary word from the appearance frequency of the arbitrary word appearing in the vicinity of the word that matches the query word in
The index generation means extracts the words belonging to the one word class from the search target documents stored in the search target document database together with the words in the vicinity thereof, and the query words to which the query words calculated in the neighborhood word importance calculating step belong Calculating the score of the document for the words belonging to the one word class using the importance of the arbitrary word that matches the neighboring word among the importance of the arbitrary word in one word class, Is repeated for all the words belonging to the one word class included in each search target document, and an index having a score corresponding to a combination of each search target document and each word belonging to the one word class as an element is generated. And an indexing method comprising the steps of: The index generation step
Sequentially acquiring words in the search target document from the beginning to the end of the sentence;
It is determined whether or not the acquired word belongs to one word class. If so, the importance level is calculated in the neighborhood word importance calculation step for N words before and after the acquired word belonging to the one word class. If there is a word that matches the calculated arbitrary word, adding the importance to the score of the search target document for the word belonging to the acquired one word class;
The index creation method according to claim 2, further comprising: outputting scores of the search target document for all words belonging to one word class acquired from the search target document.   The program for functioning a computer as each means of the index production apparatus of Claim 1.   The computer-readable recording medium which recorded the program of Claim 4.

Images