JP2006227823A - Information processor and its control method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve a problem of a search time prolonged when high level search is carried on a large amount of documents. <P>SOLUTION: A query is inputted (S71), and its query character string is classified by category (S73). Information having category information matching the classified category is extracted as a search object (S74), and search processing based on the query character string is carried out on the extracted information (S75). <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an information processing apparatus that searches for information based on information including a character string and a control method thereof.

  In recent years, with the increase in capacity of storage devices, databases that store a large amount of documents and the like have become common, and a technique for retrieving a desired document from these large amounts of documents has become important. This is because a simple keyword search (indicating a keyword and searching for a document based on whether or not it appears) does not sufficiently satisfy the user's request. Therefore, as a document search method, a method using a relationship with a keyword and a similarity of syntax information, a method using a document vector characterizing the content of a document, and the like have been newly proposed. Furthermore, a method has also been proposed in which the meaning, field, and word itself that characterize the contents of a document are taken as dimensions, the vectors are expressed as feature quantities, and the similarity between documents is obtained using values such as inner products between document vectors. .

  In addition, for the purpose of directory search and document management, a method of registering a search target document in a database by manually or automatically classifying the document into some categories has been proposed.

  Further, when searching by keyword, it is effective as a technique for reducing search omissions to search by expanding the keyword into words related to the keyword. The simplest technique for this is to expand the search keyword by referring to the expanded word dictionary for the target word and adding the expanded word stored in the dictionary as a new keyword. I was searching. For example, if the keyword “cop” is expanded by referring to the expansion word dictionary and referring to the heading “cop”, for example, at the expansion level 1, “police officer”, “inspector”, etc. can be found as search keywords. . Moreover, in the expansion level 2, “police” and “criminal” can be found as expansion words.

  Patent Document 1 describes a technique for expanding words into related words using a neural network. In this example, when a search condition of ““ mandarin orange ”and“ winter ”” is input, “mandarin orange” and “winter” are expanded into related words by a neural network for word expansion. Furthermore, in this example, “Shizuoka”, “Apple”, “Kotatsu” and “Cat” are obtained with a weak relevance degree as a strong relevance degree for “mandarin oranges”. As related words of “winter”, “kotatsu” and “cat” are extracted with strong relevance, and “apple” and “Shizuoka” are extracted with weak relevance. In the end, “kotatsu” having a high degree of association is obtained for both the words “mandarin orange” and “winter”, and the other words “cat”, “apple” and “Shizuoka” are obtained as weakly related words. At the time of search, a search using new keywords ("kotatsu", "cat", "apple", "Shizuoka") is performed in addition to the keyword ("mandarin orange", "winter") that was input first.

Patent Document 2 discloses a technique for changing an input search condition according to its meaning. For example, if you enter the keyword “Co., Ltd. XX Construction”, it will be divided into “Co., Ltd.”, “XX”, and “Construction” by morphological analysis, and “XX Construction Co., Ltd.” “ It describes that a search is performed using another keyword that expands the original keyword, such as “XX Construction Co., Ltd.”, “XX Construction Co., Ltd.”, and “XX Construction”.
JP-A-5-135110 JP-A-10-207896

  However, the conventional information retrieval apparatus has the following problems.

  When an advanced search is performed on a large number of documents, the search time becomes longer.

  Further, in the conventional keyword expansion process, a headline for a keyword is provided, and the expansion word for the headline is acquired and expanded. However, since there are words having different meanings even in the same expression, word expansion corresponding to such ambiguity of words cannot be performed. For example, since “DV” has meanings such as “digital video” and “domestic violence”, when “DV” is expanded into words, it is expanded into “digital video” and “domestic violence”. For this reason, even when it is originally desired to search for a document related to an audio device, a document including “domestic violence” is searched, and many unnecessary documents are included as a search result. For this reason, omission of search can be prevented by performing the word expansion described above, but it is insufficient to improve the search accuracy.

  The present invention is to solve the above-mentioned drawbacks of the prior art.

  Another feature of the present invention is to provide an information processing apparatus capable of obtaining a highly accurate search result in a short time and a control method thereof.

  The above features are achieved by combinations of the features described in the independent claims, and the dependent claims merely define advantageous embodiments of the invention.

An information processing apparatus according to one embodiment of the present invention has the following configuration. That is,
An information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input means for entering a query string;
Query classification means for classifying the query string input by the input means into one or more categories;
Extraction means for extracting information having category information corresponding to the category classified by the query classification means as a search target;
A search unit that executes a search process based on the query character string input by the input unit, using the information extracted by the extraction unit as a search target;
Display means for displaying a search result by the search means;
It is characterized by having.

An information processing apparatus according to one embodiment of the present invention has the following configuration. That is,
An information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input means for entering a query string;
Extraction means for extracting a character string as a keyword based on a word included in the query character string input by the input means;
An acquisition means for acquiring an expanded character string related to the character string serving as the keyword and a category of the expanded character string;
Candidate search means for searching for documents in the storage device using the character string as the keyword and the expanded character string as keywords and extracting search candidates;
Search means for making a search result a document having category information that matches the category of the expanded character string among the documents searched by the candidate search means;
It is characterized by having.

An information processing apparatus control method according to an aspect of the present invention includes the following steps. That is,
A method for controlling an information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input process for entering a query string;
A query classification step of classifying the query string input in the input step into one or more categories;
An extraction step of extracting information having category information corresponding to the category classified in the query classification step as a search target;
A search step for performing a search process based on the query character string input in the input step, using the information extracted in the extraction step as a search target;
A display step for displaying a search result in the search step;
It is characterized by having.

An information processing apparatus control method according to an aspect of the present invention includes the following steps. That is,
A method for controlling an information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input process for entering a query string;
An extraction step of extracting a character string as a keyword based on the word included in the query character string input in the input step;
An acquisition step for acquiring a development character string related to the character string to be the keyword and a category of the development character string;
A candidate search step of searching a document in the storage device using the character string as the keyword and the expanded character string as a keyword and extracting a search candidate;
A search step in which a document having category information that matches the category of the expanded character string among the documents searched in the candidate search step is a search result;
It is characterized by having.

  The outline of the present invention does not enumerate all necessary features, and therefore, a sub-combination of these feature groups can also be an invention.

  According to the present invention, it is possible to perform a high-speed and high-precision search from a query including a character string. As a result, the operability for the user is greatly improved, and the desired information can be efficiently extracted.

  DESCRIPTION OF EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The following embodiments do not limit the invention according to the claims, and all the combinations of features described in the embodiments are not necessarily essential to the solution means of the invention. .

[Embodiment 1]
FIG. 1 is a block diagram showing a schematic configuration of an information search apparatus (for example, a personal computer (PC)) according to an embodiment of the present invention.

  This information retrieval apparatus includes a central processing unit (CPU) 1 that executes processing according to a program, and a storage unit (RAM and RAM) that stores a program and a dictionary in addition to a document to be searched and an annotated image. (Including a mass storage device such as a hard disk) 2, an input unit 3 such as a keyboard and a mouse that are operated by a user to input various data and commands, a display unit 4 that displays search results, and an FD (floppy) (Registered trademark) disk, CD-ROM, DVD, and storage medium reading unit 5 that reads programs and data stored in a storage medium such as a magnetic tape and stores them in the storage unit 2.

<Overview of search operation>
Next, an outline of a search operation performed by the information search apparatus according to this embodiment will be described.

  FIG. 2 is a flowchart showing an automatic classification registration process for documents and the like by the information search apparatus according to the first embodiment of the present invention. A program for executing this process is stored in the storage unit 2 and is controlled by the central processing unit 1. Executed below. This process indicates a process for registering the classification information together with the document by determining which field the document is related to when registering the document.

  First, in step S1, registration target data such as a document to be registered is acquired. Next, in step S2, the language of the registration target data is analyzed, and analysis data such as words included in the registration target data is created. In step S3, the registration target data is classified based on the analysis data. Based on the classification result, the registration destination is determined (step S4). In step S5, the registration target data (here, a document) is registered in the registration destination determined in step S4.

  Next, details of the registration target data classification process in step S3 will be described. As an example of the classification of the registration target data, a case of automatic classification using vectors will be described. In classifying the registration target data, an effective word dictionary and a base word dictionary are created from documents in categories to be classified in advance.

  Next, the effective word extraction process will be described with reference to the flowchart of FIG. Here, in a state where a plurality of documents are stored in the database, valid words included in the documents are extracted from the plurality of documents.

  FIG. 3 is a flowchart for explaining processing for extracting a word from a plurality of documents stored in a database, determining whether or not it is a valid word, and registering it in the information search apparatus according to the present embodiment. A program for executing processing is stored in the storage unit 2 and executed under the control of the central processing unit 1.

  First, in step S11, morphemes of all document data held in the database are analyzed to extract words. In step S12, the extracted word is stored in the storage unit 2 as a word list. In step S13, an arbitrary word is extracted from the word list. In step S14, the effectiveness of this word for document classification is calculated. Here, the ratio of documents containing this word among the documents belonging to a certain category is calculated. In this way, all categories are normalized. Here, if there is a difference in the ratio of documents containing the word depending on the category, the word is considered to characterize a category having a high ratio, and can be said to be an effective word for classification.

  On the other hand, if the ratio of documents containing the word is not different depending on the category, it is considered that it is not effective for classification. In order to evaluate such distribution bias, entropy is calculated. Here, the entropy has a smaller value as the distribution bias is larger (effective for classification), and takes a larger value as the bias is smaller (ineffective for classification). The effectiveness is defined as “1-entropy”.

  Accordingly, in step S15, it is determined whether or not the obtained effectiveness is greater than a threshold value. If it is determined that the effectiveness is smaller than the threshold (effective for classification), the process proceeds to step S16, the word is regarded as an effective word, registered in the effective word dictionary, and the process proceeds to step S17. On the other hand, if the validity level is equal to or greater than the threshold (not valid for classification), the registration process to the valid word dictionary (provided in the storage unit 2) in step S16 is skipped, and the process proceeds to step S17. Delete from the word list. In this embodiment, the threshold value is set appropriately, and as many valid words as possible are extracted and registered in the valid word dictionary within a range where the document classification can be effectively performed.

  In step S18, it is determined whether or not the word list is empty (whether determinations for all words have been completed). If not, the process returns to step S13, and the same processing is performed for the next word. When the list becomes empty, the effective word extraction process ends. This process can also be performed manually.

  Next, a base word extraction process will be described. As the base words, it is desirable that each of them has a high effectiveness for classification, and at the same time has a low correlation with each other. For example, it is assumed that “exchange”, “highest price”, “economy”, “good technology”, “inflation”, etc. are registered as valid words in the valid word dictionary. These are words that often characterize the category “economic”. However, since “exchange” and “best” appear frequently in the same document, it is redundant to adopt both as base words. Rather, it is better to select “exchange”, “good tech”, “inflation” or the like as a base word and add correlation information between “highest value” and “exchange” in the effective word dictionary.

  The base word extraction process based on this idea will be described with reference to the flowchart shown in FIG.

  FIG. 4 is a flowchart showing a base word extraction process in the information search apparatus according to the present embodiment. A program for executing this process is stored in the storage unit 2 and executed under the control of the central processing unit 1. .

  In this base word extraction process, first, in step S21, a co-occurrence probability between words (effective words) registered in the effective word dictionary is calculated from the documents held in the document database. The co-occurrence probability of the word W with respect to the word W ′ is obtained by (number of documents including W and W ′ simultaneously) / (number of documents including W). Next, in step S22, as a default setting, the base word list (storage unit 2) holding the selected base word and the base word candidate list (storage unit 2) holding base word candidates are respectively emptied and selected. The number of base words n (storage unit 2) is “0”. In step S23, all of the words registered in the valid word dictionary except the words held in the base word list B are held as base word candidates in the base word candidate list, and the base word evaluation is performed. The maximum value is initialized to “0”. In step S24, an arbitrary word W is extracted from the base word candidate list, and an evaluation value as a base word of the word W is calculated (step S25).

This evaluation value is obtained as follows. If the selected base word list is still empty, the evaluation value is the validity. If a base word has already been selected, the evaluation value of the word as a base word increases as the effectiveness of the word for document classification increases, and decreases as the correlation with the base word increases. In step S26, it is determined whether or not the evaluation value calculated in step S25 is larger than the maximum value. If it is larger, the process proceeds to step S27, a word is set as the next base word candidate, and the maximum value is set as the word evaluation value. Update and go to step S28. On the other hand, if the evaluation value is equal to or less than the maximum value, the process of step S27 is skipped and the process proceeds to step S28. In step S28, the determined word is deleted from the base word candidate list. In step S29, it is checked whether or not the base word candidate list has been emptied (whether the calculation of the validity of all base word candidates has been completed). If not, the process returns to step S24 to return the remaining valid words (basis The same evaluation is performed for word candidates. When the basic word candidate list becomes empty in step S29, the process proceeds to step S30, where the basic word candidate is added to the basic word list, and the number of basic words is incremented by one. In step S31, it is checked whether or not the number of basic words has reached the number set in advance by the user. If not, the process returns to step S23 to select the next basic word candidate. When the number of base words reaches the set number by rubbing, the base word extraction process is completed.

  Correlation information between the base word extracted in this way and each word registered in the effective word dictionary is registered in the effective word dictionary. This is done by describing the co-occurrence probabilities for the word base words for each word in the valid word dictionary. In addition, the same correlation information is assigned to synonymous effective words using a synonym dictionary. This can also be processed manually in advance.

  Vector expression processing will be described with reference to the flowchart shown in FIG. The classification category is determined by the vector created by the vector expression process.

  FIG. 5 is a flowchart for explaining vector expression processing in the information search apparatus according to the present embodiment. A program for executing this processing is stored in the storage unit 2 and executed under the control of the central processing unit 1.

  First, in step S41, the vector is initialized. Next, in step S42, the target document is subjected to morphological analysis, and in step S43, a word list including a pair of words included in the document and the frequency thereof is created in the storage unit 2. In step S44, a set of one arbitrary word and its frequency is extracted from the word list. Next, in step S45, the word is drawn with the effective word dictionary created according to the flowchart of FIG. 3, and in step S46, it is checked whether or not the word is registered in the effective word dictionary. If it is not registered here, the process returns to step S44. On the other hand, if the word is registered in the valid word dictionary, the process proceeds to step S47, a vector representation of the word is generated based on the valid word dictionary, and the vector is updated (step S48). In step S49, the combination is deleted from the word list, and in step S50, it is checked whether or not the word list is empty (processing of all the words extracted in step S43 is completed). If it is not empty, the process returns to step S44 and the same processing is performed for the next word. If the word list is empty, the process ends.

  In this way, the content of the document can be expressed by a vector using the effective word dictionary. That is, first, as many valid words as possible for classification are registered in the valid word dictionary as much as possible. From this, as few base words as possible are extracted as axes of the vector space representing the document. Then, the effective word dictionary has correlation information between these base words and each effective word registered in the effective word dictionary. This increases the probability that a word included in a document input as a classification target matches one of the valid words registered in the dictionary, and makes the document a vector with fewer dimensions based on the correlation information between the valid word and the base word. By expressing it, the processing cost on the vector space is reduced.

  Note that the present invention is not limited to the above-described embodiment. For example, when extracting a word from a document as described above, processing in word extraction using a method such as character type cutting without using morphological analysis. You may make it raise speed. In the above embodiment, the entropy calculation is used to evaluate the effectiveness for classifying words. However, other evaluation functions may be used as long as the distribution bias can be evaluated.

  Furthermore, the evaluation function of the base word is not limited to the one shown in the above embodiment, and other evaluation functions can be used as long as the effectiveness for the word itself and the correlation between the base words are taken into consideration. It may be used.

  In the above embodiment, correlation information with all base words is given to each word in the dictionary in the effective word dictionary, but only correlation information with several higher-order base words with high correlation is provided. To reduce the scale of the effective word dictionary. A simple field dictionary, synonym dictionary, co-occurrence dictionary, or the like may be used. Registration of search data is performed by registering necessary data according to the search method.

  FIG. 6 is a flowchart showing a registration process for concept search in the information search apparatus according to the present embodiment. A program for executing this process is stored in the storage unit 2 and executed under the control of the central processing unit 1. Is done.

  The concept search registration process includes a language analysis result acquisition process in step S61, a document vector generation process in step S62, and an index registration process in step S63.

  In FIG. 6, at the stage where a document is registered, first, in step S61, a result of language analysis is acquired, and a word included in the document is acquired. In step S62, a document vector is generated, and the document vector is calculated from the words appearing in the document using a basic vector dictionary.

  FIG. 7 is a diagram showing the configuration of this basic vector dictionary, which is stored in the storage unit 2.

  As shown in the figure, this basic vector dictionary stores feature amounts corresponding to respective dimensions (Dim.) At the time of vector expression for each word. For the dimension, a standard classified according to the original meaning of the word, a standard classified according to the field of use of the word, or the like is adopted. It can be seen that the feature value of dimension 01 (“Dim.01”) of “word 1” is “0”, and the feature value of dimension 02 (“Dim.02”) is “23”.

  As described above, it is possible to obtain the feature amount of each dimension (Dim.) In one word from the basic vector dictionary. This feature amount can be interpreted as a value indicating how much the document may characterize the classification standard (= dimension) by using the word. From the feature values for each classification criterion (by dimension) obtained from all the words constituting the document, the feature amount of the entire document is expressed by a vector having the classification criterion as a dimension. A value obtained by normalizing the vector thus obtained with norm = 1 is stored as a document vector.

  In the subsequent index registration process (step S63), the document vector is stored in an index as shown in FIG. It can be seen that the feature quantity of “Dim.01” of the document vector “document ID = 6947” is “183”, and the feature quantity of “Dim.02” is “0.214”.

  Next, FIG. 9 is a diagram illustrating an example in which a plurality of documents to be searched are classified into four categories (here, “sports”, “politics”, “economy”, and “environment”).

  FIG. 10 is a flowchart for explaining a search process in the information search apparatus according to the present embodiment. A program for executing this process is stored in the storage unit 2 and executed under the control of the central processing unit 1.

  First, in step S71, a query (search data) input by the input unit 3 is acquired. Next, in step S72, language analysis of the acquired query is performed. In step S73, the query is classified based on the result of language analysis of the acquired query. Based on the classification result in step S73, the category of the document to be searched is determined in step S74. In step S75, the search target (category) document determined in step S74 is searched. In step S76, a list of search results obtained by this search process is displayed on the display unit 4 in descending order of coincidence.

  FIG. 11 is a detailed flowchart when a phrase search is applied to the search process in step S75 of FIG.

  This phrase search includes word acquisition processing included in the query in step S81, processing for acquiring a relationship between words included in the query in step S82, and similarity calculation processing in step S83.

  In the word acquisition process in step S81, the word included in the query is acquired from the morpheme analysis result in step S42 (FIG. 5). Next, the relationship acquisition processing between words included in the query in step S82 is processing for acquiring syntax information of the words included in the query from the morpheme analysis result in step S42 (FIG. 5). In addition, the similarity calculation processing in step S83 is performed by comparing the word included in the query acquired in step S81 and step S82 with the relationship between the word and the word included in the document with the relationship between the words. This is a process for calculating the similarity between a query and a sentence in a document.

  FIG. 12 is a detailed flowchart when the concept search is applied to the search processing in step S75 of FIG.

  This concept search includes a process for obtaining the language analysis result of the query in step S91, a query vector generation process in step S92, and a similarity calculation process in step S93.

  Here, the query language analysis result acquisition processing in step S91 is processing for acquiring the query language analysis result generated in step S72 (FIG. 10). The query vector generation process in step S92 is a process for generating a query vector from the analysis result acquired in step S91, and the similarity calculation process in step S93 stores the query vector obtained in step S92 and the registration time. This is a process of comparing the document vector for each document stored in the section 2, calculating the similarity between the query and a sentence in the document, and listing the result as a search result.

  Here, a case will be described where a plurality of documents classified into four categories of “politics”, “economy”, “environment”, and “sports” are stored in the storage unit 2 as shown in FIG.

  FIG. 13 is a diagram illustrating a display example of an input screen for a query (character string) input from the input unit 3.

  In FIG. 13, “Regarding the effect of interest rates in the Japanese economy” is input as a query.

  Next, a search process based on this query will be specifically described with reference to the flowchart shown in FIG.

  First, in step S71, a query “about the influence of interest rates in the Japanese economy” input using the input unit 3 is acquired. Next, in step S72, language analysis of the query acquired in step S71 is performed. Here, based on “Japan”, “economy”, “interest rate”, “no”, “influence”, and “about” broken down into morphemes, a query vector is generated by a method similar to the above-described classification of registered documents. Then, the query vector is compared with the category of FIG. Here, it is classified into the category of “economy”.

  As a result of the classification in step S73, the search target is determined to be the “economy” category (step S74). In step S75, documents belonging to the determined “economy” category are searched as search targets. The search result list of this search step is displayed as a list in descending order of similarity on the display unit 4 as shown in FIG. 14 (step S76).

  FIG. 14 is a diagram illustrating a search result in the information search apparatus according to the present embodiment.

  In FIG. 14, five items are displayed in order from the highest similarity among the search results for the query “influence of interest rates in the Japanese economy”.

<Specific description of search operation>
Next, the search processing according to the present embodiment realized by the processing described in each of the flowcharts described above will be specifically described.

1. Phrase Search Details when the phrase search is applied to the search processing in step S75 in FIG. 10 will be described with reference to the data examples in FIGS. 15A to 15C and the flowchart in FIG.

  As shown in FIG. 15A, when “print image” is input as a query, in the process of acquiring words included in the query (step S81 in FIG. 11), “image” “ ”Is acquired.

  Next, in the relationship acquisition process between words in step S82 (FIG. 11), it is acquired that “image” is related to “print”. In the subsequent similarity degree calculation process in step S83 (FIG. 11), the degree of similarity between the words included in the query and the words is calculated.

  For example, in the document 1 in FIG. 15B, the word “image” and the word “print” are included, and ““ image ”is related to“ print ”” and ““ high speed ”is“ It is shown that “it is related to“ printing ”. Here, “the“ image ”is related to“ print ”” ”is similar between the query in FIG. 15A and the document 1 in FIG. Therefore, the document 1 has a similarity degree including similar phrases (= 1−a deduction by a word (0) −a deduction by a relationship (because the distance between words is 0.1)) = 0.9.

  Next, in the case of the document 2 in FIG. 15C, the word “image” and the word “print” are common, but the “image” is related to “reading”. Since “print” is related to “document”, the similarity is (= 1−deduction by word (0) −reduction by relationship (0.5 with different relationship)) = 0.5.

  Accordingly, the document 1 shown in FIG. 15B having a high similarity is displayed on the display unit 2 as a search result.

  FIG. 19 is a diagram showing a display example of a search result when a search process based on the above-described query “About the effect of interest rates in the Japanese economy” is performed by a phrase search. In FIG. 19, the search results of (1) to (3) have substantially the same similarity because “Japan economy” is included, but in order of the time closer to the present time (2004). Is attached.

2. Concept Search Next, the details when the concept search is applied to the search processing (step S75) in FIG. 10 will be described with reference to the flowchart in FIG.

  First, the result of the query language processing in step S91 is acquired, and words included in the query are acquired. Next, in the query vector generation processing in step S92, a query vector is generated by the basic vector dictionary, as in the case of document vector generation. In the similarity calculation process in step S93, the query feature vector and the similarity of each document vector are calculated.

  FIG. 20 is a diagram showing a display example of a search result when the search process based on the above-described query “About the influence of interest rates in the Japanese economy” is performed by concept search.

  As described above, according to the first embodiment, when searching for information such as a target document or an annotated image, the input query is subjected to linguistic analysis, and which search is performed from the analysis result (syntax information, etc.). It is determined whether to apply the mode, and the determined search mode is selected and executed. As a result, a highly accurate search result with less noise is automatically obtained without the user selecting a search mode. Therefore, the target information can be searched effectively.

[Embodiment 2]
FIG. 16 is a flowchart showing a search procedure according to the second embodiment of the present invention.

  First, in step S101, a query input by the input unit 3 is acquired. Next, in step S102, language analysis of the acquired query is performed. In step S103, the query is classified into categories based on the result of language analysis of the query. As a result of this classification, a search target is determined, and the determined search target is searched (step S4). If the search target is “A”, the search target (category A) is searched in step S105. If the search target is “B”, the search target (category B) is searched in step S106. In this way, the search result list of the search step of step S105 and the search result list of step S106 is displayed as a list in descending order of similarity on the display unit 4 (step S107).

  Next, the flow of the flowchart showing the search procedure of FIG. 16 will be described along the data.

  FIG. 17 shows a state where a plurality of documents classified into seven categories are accumulated in the storage unit 2. Here, seven categories of “politics”, “world economy”, “Japanese economy”, “environment”, “international”, “sports”, and “entertainment” are included.

  First, in step S101, the query “about the influence of the Japanese economy in the world economy” is acquired by the input unit 3 using the screen, and then in step S102, the acquired query is subjected to language analysis. “World” “Economy” “In” “Japan” “Economy” “No” “Impact” “About”, and based on the results of linguistic analysis of the query, the same method as the classification of registered documents to be searched A query vector is generated, compared with each category, and classified into two categories, “world economy” and “Japan economy” (step S103). As a result of this classification, the search target is determined to be the category of “world economy” and “Japanese economy” (step S104). In step S105, only the “world economy” category thus determined is searched as a search target. In step S106, only the determined “Japan economy” category is searched as a search target. In this way, the search result list of the search step of step S105 and step S106 is displayed as a list in descending order of similarity on the display unit 4 as shown in FIG. 14 (step S107). Although the example classified into two classification categories was described here, it is not limited to two.

[Embodiment 3]
FIG. 18 is a flowchart showing an information search procedure according to Embodiment 3 of the present invention.

  First, in step S111, a query input by the input unit 3 is acquired. Next, in step S112, language analysis of the acquired query is performed. In step S113, when the search target is designated as “A” (here, server A managing documents of a predetermined category), in step S114, the search target is classified into one designated search target category. Therefore, the query is classified based on the result of the language analysis of the query acquired in step S112. Next, in step S115. Search is performed based on the classification result of step S114.

  On the other hand, when the search target is designated as “B” (here, the server B that manages the document of the predetermined category) in step S113, the process proceeds to step S116, and the result of the language analysis of the query in step S112 is performed. Classify into: Next, a search is performed in step S117. In this way, in step S118, the search process list in step S115 and the search result list obtained by the search process in step S117 are displayed on the display unit 4 in a descending order of similarity.

  Next, the flow of the information search procedure of FIG. 18 will be described along the data of FIG.

  As shown in FIG. 9, a server A in which a plurality of documents classified into four categories of “politics”, “economy”, “environment”, and “sports” are stored in the storage unit 2, and “politics” “ A plurality of servers such as server B in which a plurality of documents classified into the seven categories of “world economy”, “Japanese economy”, “environment”, “international”, “sports”, and “entertainment” are stored in the storage unit 2 respectively Search for multiple documents that are classified into different categories.

  First, a query “Regarding the impact of the Japanese economy on the global economy” is input from the input unit 3 (step S111), and then in step S112, the acquired query is subjected to language analysis. Next, in step S113, the designated search target server is determined. Here, server A and server B are search targets. In step S114, when server A is designated in step S113, the result of language analysis of the query acquired in step S112 in order to classify server A into the classification categories (four types) “Japan” “Economy” “ Based on "", "interest rate", "no", "impact", and "about", a query vector is generated by server A in the same manner as the classification of registered documents, and is compared with each category of server A. ”Category. In step S115, the determined “economy” category is searched for as a search target (step S115).

  When the server B is classified into the classification categories (seven types), in step S116, as a result of the linguistic analysis of the query acquired in step S112, "Japan" "Economy" "In" "Interest rate" "No" "Influence" Based on “about”, a query vector is generated in the same manner as the classification of the registered document of the server B. The query vector is compared with each category of the registered document of the server B, and is classified into the category “Japanese economy”. In step S117, the determined “Japanese economy” category is searched as a search target. In this way, the search process of step S115 and the search process result of step S117 are merged, and the search result list is displayed as a list in descending order of similarity on the display unit 4 as shown in FIG. 14 (step S118).

  In this third embodiment. Although the case of two servers A and B has been described, the present invention is not limited to two, but can be applied to the case of one or a plurality of servers. Different classification methods may be applied to each server.

[Embodiment 4]
FIG. 21 is a block diagram showing a schematic configuration of the information processing apparatus according to Embodiment 4 of the present invention. Portions that are common to the configuration of FIG.

  In the figure, the output unit 103 includes the display unit 4 and a printing unit such as a printer. A registration document input from the input unit 3 such as a keyboard or a file is processed by the CPU 1 and stored in the storage unit 2 by the search processing program 111 and the classification processing program 112 developed in the storage unit 2 such as a memory or a hard disk. be registered. The classification processing program 112 extracts keywords from the registration document, executes classification processing based on the extracted keywords, and generates classification information of the document. The search processing program 111 extracts keywords from the query document and registers them in the search index data 113 that is also stored in the storage unit 2.

  When searching for a document, the search processing program 111 extracts words that serve as search conditions from search conditions input from the input unit 3 such as a keyboard. The expanded word dictionary data 114 is acquired according to the search condition, and the search is executed. The search processing program 111 extracts a search keyword from the search condition, uses the word information as a headline, and matches the input search condition from the search index data 113 storing the document information and the characteristics of the word included in the document. Get the document. Then, it is determined whether similarity is given to the search condition or whether it is a final result, and the result is output to the output unit 103 such as a display.

  Further, the operating environment of this embodiment can correspond to a local network environment or an Internet environment in addition to a single computer.

  FIG. 22 is a flowchart showing an operation at the time of document registration in the information processing apparatus according to the fourth embodiment of the present invention.

First, in step S121, document analysis processing is performed on the input document. In this document analysis process, a document is divided into units of words by processing such as morphological analysis. For example, a morphological analysis is performed on document 1 as “Prospects for domestic violence is expected to be enacted in the National Diet” and divided into words. As a result, it is divided as follows. "Book (suffix) / Diet (noun) / De (particle) / Domestic violence (noun) / Ni (particle) / Related (verb) / Bill (noun) / Ga (particle) / Established (verb) / Prospect (noun) /. "
In step S122, words necessary for search processing and classification processing are extracted from the analyzed document. In the fourth embodiment, the words whose part of speech is “noun”, “verb”, “adjective”, and “adjective verb” are extracted as keywords. Accordingly, “National Diet”, “Home Violence”, “Regarding”, “Bill”, “Establishment”, and “Prospect” are extracted as keywords.

  In step S123, classification information is created by categorizing the document.

  FIG. 23 is a diagram illustrating an example of high-frequency keyword information for each category.

  The high-frequency keyword information for each classification is referred to, and the classification information is added to the document in consideration of the usage status of keywords in the document. Here, it is understood that “National Diet”, “Bill”, “Established”, etc. are words that frequently appear in the category of “politics”. Therefore, the document 1 is classified into the category [politics].

  In step S124, search index data is created. The minimum information necessary as index data for this search is notation character string information that becomes a headline when searching for data and information on a document in which the character string appears.

  FIG. 24 is a diagram showing an example of a search index for document 1 (category “politics”) according to the fourth embodiment.

  Here, the keyword extracted in step S122 is used as a heading character string, and “document 1” which is a document registered as an appearance document ID is stored. In step S125, the document and classification information are linked and stored (see FIG. 25). Here, the document ID and the classification information generated in step S123 are linked and stored as document classification data.

  FIG. 25 is a diagram showing an example of this document classification data.

  Here, the document ID for identifying the document and its classification information are stored. Here, as described above, “Document 1” is classified as “Politics”, so “Document 1” is stored in the document ID, and “Politics” is stored in the classification information.

  With the above processing, the registration processing of “Document 1” is completed.

  When registering a plurality of documents, the processes in steps S121 to S125 described above are repeated a number of times corresponding to the number of documents. In the fourth embodiment, “Document 1” to “Document 5” shown in FIG. 26 are registered.

  As a result, the search index data shown in FIG. 27 and the document classification data shown in FIG. 28 can be created.

  FIG. 27 is a diagram illustrating an example of extraction of a heading character string included in each document illustrated in FIG.

  FIG. 28 is a diagram showing an example of classification information corresponding to each document classified based on these heading character strings.

  Next, search processing by the information processing apparatus according to Embodiment 4 of the present invention will be described.

  FIG. 29 is a flowchart for explaining search processing by the information processing apparatus according to the fourth embodiment. A program for executing this processing is stored in the storage unit 2 and executed.

  First, in step S131, a search condition (query) is input from the input unit 3. Here, the search condition is input as a natural sentence or a logical expression. In the fourth embodiment, a description will be given in the case where “natural violence” is input in a natural sentence. In step S132, keywords included in the query are extracted. Here, character cutting processing such as morphological analysis is performed on the search condition (query) input in step S131 to extract a character string (word) that becomes a keyword in the search. Here, the natural sentence “problem of domestic violence” as a search condition is cut into words as follows.

"Home violence (noun) / no (particle) / problem (noun)"
From the character string thus cut out, a keyword condition at the time of registration, that is, a word whose part of speech is a noun, a verb, an adjective, or an adjective verb is used as a keyword. Accordingly, “domestic violence” and “problem” are extracted as keywords.

  In step S133, it is determined whether keyword expansion is designated. If keyword expansion is not designated, the process proceeds to step S136 using the acquired keyword as it is as a search condition. On the other hand, if keyword expansion is designated, the process proceeds to step S134, and the keyword is expanded with reference to the word expansion dictionary 114.

  FIG. 30 is a diagram illustrating an example of a part of the contents of the expansion dictionary 114 used in the fourth embodiment.

  In the figure, it can be seen that, for example, the expansion target word “problem” is expanded into “trouble” and “problem”. In addition, when the target word is “DV”, there are expanded words of “digital video”, “domestic violence”, and “domestic violence”, but these expanded words are restricted. For example, “digital video” is “digital video (industry)”, and “industry” in parentheses is information for designating a development category. As a result, this “digital video” is valid only when the classification information (category) of the document to be searched is “industry”, and is invalid for other classification information (category). ing. Similarly, “domestic violence” and “domestic violence” are given “politics and society”. Therefore, “domestic violence” and “domestic violence” are effective only when the classification information of the document to be searched is “politics” or “society”.

For example, when it is desired to make only “domestic violence” effective for the classification information “culture”, it may be “domestic violence (politics / culture / society)”. Thereby, “domestic violence” is also effective when the classification information of the document to be searched is “politics”, “culture”, and “society”. In addition, general expansion words that do not depend on the classification information of the document to be searched are not given such classification information (such as “trouble”, “problem”, “issue”).
Since the keywords according to the fourth embodiment are “domestic violence” and “problem”, the expansion dictionary 114 is searched, and “domestic violence” is “DV (politics / society)” “digital video (industry)” ) ”“ Domestic violence ((politics / society)). Similarly, “problem” is developed into “trouble” and “issue”.

  Next, in step S135, the expanded word thus expanded is acquired and a search condition is created. In other words, the search conditions “family violence” and “problem” that are search conditions before deployment are the search conditions “domestic violence or DV (politics and society) or domestic violence (politics and society)” and “problem or trouble or problem”. become. Here, “or” indicates an alternative condition “or”.

  When the keyword used for the search is determined in this way, a search process using the search index is executed in step S136.

  In the fourth embodiment, a case will be described where search index data (FIG. 27) registered in documents 1 to 5 is searched. The search index data can be searched using the search conditions created in step S135, and the following information can be acquired.

Domestic violence Document 1
Domestic Violence (Politics / Society) Document 3
DV (Politics / Society) Document 2, Document 5
Problem Document 2, Document 5
Trouble, problem No search results.

  Thus, in “domestic violence”, document 1, document 2, document 3, and document 5 are search result candidates, and in “problem”, document 2 and document 5 are search result candidates.

  In step S137, referring to the document classification data in FIG. 28, a search result is created from the result candidate obtained in step S136. Here, “domestic violence” and “problem” are keywords and are effective for all categories. Therefore, the search result candidate in step S136 can be used as the final search result as it is.

  On the other hand, for “domestic violence (politics / society)”, document 3 is a candidate for a search result. This search word is an expanded word that is effective only when the document classification information is “politics” and “society” from the expanded dictionary of FIG. Therefore, referring to the classification information of the document in FIG. 28, the classification information of the document 3 is “society”, so that the search result of the document 3 can be determined to be valid.

  Next, for "DV (politics / society)", there are search result candidates for document 2 and document 5. The classification information for which the expanded word “DV” is effective is “society” and “politics”. Here, referring to the document classification information of FIG. 28, the classification of the document 2 is “society”, and therefore, it is effective as a search result. On the other hand, since the classification of the document 5 is “industry”, it can be seen that the document 5 cannot be a candidate for the expanded word “DV”.

  In this way, Document 1, Document 2, and Document 3 are extracted as search results for “domestic violence”, and Document 2 and Document 5 are extracted as search results for “Problem”. Thus, when the search condition is “problem of domestic violence”, the document 2 can be output as the search result.

  Further, by providing score information, a search result with a low score can be output without discarding other search candidates. Also, for example, the search results may be output in the order of score, such as document 2> document 1, document 3, document 5.

[Embodiment 5]
The fifth embodiment will be described using the search index data of the fourth embodiment described above. In the above-described fourth embodiment, “DV” can be applied to “domestic violence”. However, when “DV” is input, it is difficult to control the application of the expanded words “domestic violence (politics and society)” and “digital camera (industry)”.

  Therefore, in the fifth embodiment, classification processing is performed on the search condition, and the search processing is performed by determining where the search intention is. Here, a case where a character string “DV social problem” is input as a search condition will be described.

  First, keywords are acquired from the input search conditions. That is, “DV”, “society”, and “problem” are keywords. Next, classification processing is performed based on the keywords thus extracted.

  The classification information for the search condition is generated with reference to the high frequency keyword information for each sentence shown in FIG. In the fifth embodiment, the classification “society” can be attached to the search condition.

  Next, search processing is performed using the extracted keywords as search keywords. If there is a keyword expansion instruction, the search keyword is expanded in the same manner as in step S134 in the fourth embodiment. Here, as shown in FIG. 30, the expanded words of “DV” include the expanded words of “digital video (industry)”, “domestic violence (politics / society)” and “home violence (politics / society)”. I understand that.

  Next, by referring to the search index data shown in FIG. 24 according to the search keyword and the expanded word of the search keyword, the matching document information is acquired.

  Since the search condition of the fifth embodiment is a search condition classified as “society”, priority is given to a search keyword having classification information of “society” as an important expansion word. Therefore, the priority “DV” = “domestic violence” = “domestic violence”> “digital video” is established. Therefore, the importance of the retrieved document is document 1 = document 2 = document 3 = document 5> document 4.

  Furthermore, “DV” used here can be expected to be a word belonging to the classification “society”. Therefore, with reference to the document classification data of FIG. 28, priority is given to the retrieval result of the document having classification information “DV” is “Society”. As a result, document 1 = document 2 = document 3> document 5> document 4 is obtained. Finally, by giving priority to a document whose search condition classification information and document classification information match, a search result of document 2 = document 3 = document 1> document 5> document 4 can be obtained.

  As described above, according to the fifth embodiment, by including the classification information (category) of the document including the search word in the search condition, there are a plurality of documents including the search word. Even in such a case, it is possible to display the retrieved documents with the degree of matching ranking.

  As described above, according to the fourth and fifth embodiments, by using the classification information, an erroneous document is searched for due to the ambiguity of the search word, or a document to be searched can be retrieved. Can solve the problem. As a result, the search accuracy can be improved.

  The present invention is not limited to the apparatus of the above-described embodiment, and may be applied to a system constituted by a plurality of devices or an apparatus constituted by one device. A storage medium storing software program codes for realizing the functions of the above-described embodiments is supplied to a system or apparatus, and a computer (or CPU or MPU) of the system or apparatus reads and executes the program codes stored in the storage medium. Needless to say, it will be completed by doing.

  In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiments, and the storage medium storing the program code constitutes the present invention. As a storage medium for supplying the program code, for example, a floppy (registered trademark) disk, hard disk, optical disk, magneto-optical disk, CD-ROM, CD-R, magnetic tape, nonvolatile memory card, ROM is used. Can do. In addition, by executing the program code read by the computer, not only the functions of the above-described embodiments are realized, but also the OS running on the computer based on the instruction of the program code performs the actual processing. It goes without saying that a case where the functions of the above-described embodiment are realized by performing part or all of the processing, is also included.

Furthermore, after the program code read from the storage medium is written to the memory provided in the function expansion board inserted in the computer or the function expansion unit connected to the computer, the program code is expanded based on the instruction of the next program code. It goes without saying that the functions of the embodiment described above are realized by performing some or all of the actual processing by the CPU or the like provided on the expansion board or expansion unit.

It is a block diagram which shows schematic structure of the information search device which concerns on embodiment of this invention. 4 is a flowchart illustrating automatic classification registration processing for documents and the like by the information search apparatus according to Embodiment 1; It is a flowchart explaining the process which extracts the word from the document memorize | stored in the database, determines whether it is a valid word, and registers in the information search device which concerns on this Embodiment. It is a flowchart which shows the extraction process of the base word in the information search device which concerns on this Embodiment. It is a flowchart explaining the vector expression process in the information search device which concerns on this Embodiment. It is a flowchart of the concept search registration process in the information search device concerning this embodiment. It is a figure which shows the structure of the basic vector dictionary which concerns on this Embodiment. It is a figure which shows an example of the document vector of the registration document which concerns on this Embodiment. It is a figure which shows the example by which the some document used as search object is classified into four categories. It is a flowchart explaining the search process in the information search device which concerns on this Embodiment. It is the detailed flowchart at the time of applying a phrase search to the search process in step S76 of FIG. It is the detailed flowchart when a concept search is applied to the search process of step S75 of FIG. It is a figure which shows the example of a display of the input screen of a query. It is a figure which shows the search result in the information search device which concerns on this Embodiment. It is a figure which shows the example of a query. It is a flowchart which shows the information search procedure which concerns on Embodiment 2 of this invention. It is a figure which shows the state by which the some document classified into seven categories is stored and accumulate | stored in the memory | storage part. It is a flowchart which shows the information search procedure which concerns on Embodiment 3 of this invention. It is a figure which shows the example of a display of a search result at the time of performing the search process based on query "About the influence of the interest rate in the Japanese economy" by a phrase search. It is a figure which shows the example of a display of a search result at the time of performing the search process based on query "About the influence of the interest rate in the Japanese economy" by concept search. It is a block diagram showing schematic structure of the information processing apparatus which concerns on Embodiment 4 of this invention. It is a flowchart which shows the operation | movement at the time of the document registration in the information processing apparatus which concerns on Embodiment 4 of this invention. It is a figure which shows an example of the high frequency keyword information according to category. It is a figure which shows an example of the index for a search of the document 1 (category "politics") concerning Embodiment 4. FIG. It is a figure which shows an example of the document classification data based on Embodiment 4. FIG. 10 is a diagram illustrating an example of content of a document according to Embodiment 4. It is a figure which shows the example of extraction of the heading character string contained in each document shown in FIG. FIG. 20 is a diagram illustrating an example of classification information corresponding to each document classified based on a heading character string in the fourth embodiment. 15 is a flowchart for describing search processing by the information processing apparatus according to the fourth embodiment. It is a figure which shows a part example of the expansion | deployment dictionary used in this Embodiment 4. FIG.

Claims (20)

An information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input means for entering a query string;
Query classification means for classifying the query string input by the input means into one or more categories;
Extraction means for extracting information having category information corresponding to the category classified by the query classification means as a search target;
A search unit that executes a search process based on the query character string input by the input unit, using the information extracted by the extraction unit as a search target;
Display means for displaying a search result by the search means;
An information processing apparatus comprising:   The information according to claim 1, wherein the search unit executes the search process based on words and syntax information included in the query, and words and syntax information included in a document to be searched. Processing equipment.   The information processing apparatus according to claim 1, wherein the search unit executes the search process based on a query vector that characterizes the content of the query and a document vector that characterizes the content of a document to be searched.   The information processing apparatus according to claim 1, wherein the query classification unit classifies the query classification unit into one of categories corresponding to category information of a document stored in the storage device.   The information processing apparatus according to claim 1, wherein the display unit displays the search results in descending order of similarity. An information processing apparatus for registering a document so as to be searchable,
Extracting means for extracting words contained in the document to be registered;
Determining means for determining classification information of the document according to the frequency of appearance of the word extracted by the extracting means;
A registration unit that registers the word extracted by the extraction unit as a search index in association with the classification information together with the document;
An information processing apparatus comprising: An information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input means for entering a query string;
Extraction means for extracting a character string as a keyword based on a word included in the query character string input by the input means;
An acquisition means for acquiring an expanded character string related to the character string serving as the keyword and a category of the expanded character string;
Candidate search means for searching for documents in the storage device using the character string as the keyword and the expanded character string as keywords and extracting search candidates;
Search means for making a search result a document having category information that matches the category of the expanded character string among the documents searched by the candidate search means;
An information processing apparatus comprising: An information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input means for entering a query string;
Identification means for identifying a character string as a keyword and a category of the query based on a word included in the query character string input by the input means;
An obtaining means for obtaining an expanded character string of a category that is related to the character string serving as the keyword and matches the category of the query;
Search means for searching for a document in the storage device using the character string as the keyword and the expanded character string as keywords;
An information processing apparatus comprising:   9. The information processing apparatus according to claim 7, further comprising display means for displaying a search result obtained by the search means. A method for controlling an information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input process for entering a query string;
A query classification step of classifying the query string input in the input step into one or more categories;
An extraction step of extracting information having category information corresponding to the category classified in the query classification step as a search target;
A search step for performing a search process based on the query character string input in the input step, using the information extracted in the extraction step as a search target;
A display step for displaying a search result in the search step;
A method for controlling an information processing apparatus, comprising:   11. The information according to claim 10, wherein in the search step, the search process is executed based on words and syntax information included in the query and words and syntax information included in a document to be searched. A method for controlling a processing apparatus.   The information processing apparatus according to claim 10, wherein in the search step, the search process is executed based on a query vector that characterizes the content of the query and a document vector that characterizes the content of a document to be searched. Control method.   13. The information processing apparatus according to claim 10, wherein in the query classification step, the information is classified into one of categories corresponding to category information of a document stored in the storage device. Control method.   The information processing apparatus control method according to any one of claims 10 to 13, wherein, in the display step, the search results are displayed in descending order of similarity. A method of controlling an information processing apparatus for registering a document so as to be searchable,
An extraction process for extracting words contained in the document to be registered;
A determination step of determining classification information of the document according to the frequency of occurrence of the word extracted in the extraction step;
A registration step of registering the word extracted in the extraction step with the document in association with the classification information as a search index;
A method for controlling an information processing apparatus, comprising: A method for controlling an information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input process for entering a query string;
An extraction step of extracting a character string as a keyword based on the word included in the query character string input in the input step;
An acquisition step for acquiring a development character string related to the character string to be the keyword and a category of the development character string;
A candidate search step of searching a document in the storage device using the character string as the keyword and the expanded character string as a keyword and extracting a search candidate;
A search step in which a document having category information that matches the category of the expanded character string among the documents searched in the candidate search step is a search result;
A method for controlling an information processing apparatus, comprising: A method for controlling an information processing apparatus for retrieving predetermined information from a storage device that holds a plurality of pieces of information each including a character string together with category information indicating a category of the information,
An input process for entering a query string;
An identification step of identifying a character string as a keyword and a category of the query based on a word included in the query character string input in the input step;
An acquisition step of acquiring a developed character string of a category that is related to the character string that is the keyword and matches the category of the query;
A search step of searching the document in the storage device using the character string as the keyword and the expanded character string as a keyword;
A method for controlling an information processing apparatus, comprising:   The information processing apparatus control method according to claim 16 or 17, further comprising a display step of displaying a search result obtained by the search step.   The program which performs the control method of any one of Claims 10 thru | or 18.   A computer-readable storage medium storing the program according to claim 19.

Images