JP4009937B2 - Document search device, document search program, and medium storing document search program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a document retrieval technique for accumulating a large amount of document information and retrieving and presenting a document including an inputted character string (phrase).
[0002]
[Prior art]
In recent years, with the rapid development of information distribution infrastructure represented by the spread of the Internet, a large amount of document information (hereinafter simply referred to as documents) has been distributed. A system has been realized in which these documents are collected in an exhaustive manner, and a document that includes a word / phrase inputted as a search request in the text is searched and presented. In particular, as a service called a Web search engine for searching a document on the WWW (World Wide Web), goo (http://www.goo.ne.jp/) and Google (http://www.google.com/) Etc. are realized.
[0003]
In these Web search engines, a large number of documents of thousands to millions of documents are often obtained as search results for the input search request phrases. Since it is virtually impossible for a user to look through all of such a large number of search result documents, a document that is considered to be better suited to the search request phrase is selected and presented, or a search result document is displayed. The search result documents are grouped and displayed in units of sites and the subject units of the documents, and the user's desired documents are narrowed down.
[0004]
In natural language, a single phrase often has multiple meanings, that is, ambiguity, so when a search is performed using a certain phrase, documents that use that phrase with different meanings are mixed as search results. May be output. A technique for classifying search results according to the subject matter of the documents and presenting the search results in a group is useful as means for accurately narrowing down the documents requested by the user from these documents.
[0005]
As a method for classifying search results, a characteristic word (phrase) is extracted from each document using a TF × IDF method or the like, and the document is arranged in a vector space of a dimension corresponding to the number of these phrases. For example, a method is used in which the similarity between documents is defined by the cosine of the angle formed by and the documents having a high similarity are classified using cluster analysis.
[0006]
The details of the TF × IDF method are described in, for example, G. Salton, “Automatic Text Processing” 1989, Addison-Wesley Pubrishing, etc. This is a technique for calculating the degree of matching between a document and a phrase based on two principles that it is regarded as a simple phrase and a phrase with a higher appearance frequency in a document is regarded as a word that well represents the characteristics of the document.
[0007]
[Problems to be solved by the invention]
However, when the number of search result documents is large, if the above cluster analysis is performed on all documents, there is a problem that the search response time becomes long.
[0008]
Also, when calculating the feature value of each document using the TF × IDF method, the importance of each word is considered to be static and does not change. When used in the title of a book or movie, it will be regarded as a phrase that is important to the user, or a combination of a surname and a surname that is very common will be recognized as a celebrity name, and a phrase that is important to the user In spite of the fact that there are many cases that become such cases, it was not possible to accurately classify them reflecting this.
[0009]
An object of the present invention is to shorten the search response time and enable accurate classification corresponding to dynamic changes in the importance of phrases for users.
[0010]
[Means for Solving the Problems]
In order to solve the above-described problem, the present invention is a document search apparatus that searches and presents a document including a desired phrase from a set of previously stored search target documents, and uses a phrase input by a user as a search request phrase. A search request input means for receiving, a document search means for searching and outputting a document including a search request phrase input by the user from a set of search target documents, and a document output by the user by the user among the documents output by the document search means A document search apparatus is provided that includes search result display means for grouping and displaying documents including related phrases that are highly related to the input search request phrase.
[0012]
Here, as a search result display means, a phrase adjacent to the search request phrase input by the user is extracted from the history of search request phrases (a set of search request phrases input and used in the past). The appearance frequency is calculated, the high frequency word / phrase is regarded as the related word / phrase having a high degree of association with the search request word / phrase inputted by the user, and the document search means outputs the document inputted by the user. Search result display means for grouping and displaying documents including concatenated phrases of the search request phrases and these related phrases may be used. The search result display means is composed of search input phrases and adjacent phrases. Of the phrases, phrases that appear frequently in the history of search request phrases can be considered as phrases that many users tend to seek in common at that time. By considering the classification of the reference phrases, it is possible to classify the search results by the high concept of importance for the user, presented.
[0013]
As described above, according to the present invention, both the problem that the search response time becomes long and the problem that accurate classification cannot be performed in response to the dynamic change of the importance of the phrase for the user are performed. It can be solved.
[0014]
DETAILED DESCRIPTION OF THE INVENTION
Next, embodiments of the present invention will be described in detail with reference to the drawings.
[0015]
FIG. 1 shows an example of an embodiment of a document search apparatus according to the present invention. In the figure, 1 is a search request input processing unit, 2 is a document search processing unit, 3 is a search result display processing unit, This is a set of stored search target documents (a storage unit that stores them).
[0016]
The search request input processing unit 1 receives a phrase input by the user as a search request word and passes it to the document search processing unit 2.
[0017]
The document search processing unit 2 searches for a document including the search request word / phrase from the set 4 of search target documents, and outputs this as a search result document to the search result display processing unit 3.
[0018]
The search result display processing unit 3 displays, among search result documents, documents including related words / phrases having high relevance to the search request word / phrase, grouped for each related word / phrase.
[0019]
In the search result display processing unit 3, there are various possible ways to determine the related word (extract from what is determined and how to determine it). In the present invention, the word that appears adjacent to the search request word is related. as the phrase candidates, extracts words appearing adjacent to the search request terms from the history of the search request term, regarded as adjacent number appearing often words (the high frequency terms) related phrases, the search request terms Documents containing concatenated phrases with these related phrases are displayed grouped for each related phrase.
[0020]
In addition, there are various possible ways of determining the number of adjacent appearances (the appearance frequency is high), but (1) the appearance ratio in all search result documents is greater than or equal to a certain level, and (2) the higher order in the number of appearances. To what number, (3) a combination of these is conceivable.
[0021]
In addition, although the adjacent phrase to be extracted varies depending on the search request phrase (part of speech), it is usually a noun (including proper nouns such as personal names and place names), and may include adjectives, adverbs, etc. in some cases. However, conjunctions, prefixes, suffixes, etc. are excluded.
[0022]
FIG. 2 shows an example of document search and search result display processing (however, it is not included in the scope of claims) . Here, an example in which the adjacent phrase of the search request phrase in the search result document is used as the related phrase. It is a flowchart shown.
[0023]
That is, based on the search request phrase input by the user in the search request input processing unit 1, the document search processing unit 2 searches for a document including the search request phrase from the set 4 of search target documents and outputs it as a search result document. After that (S1), the search result display processing unit 3 extracts words / phrases adjacent to the search request word / phrase from the search result document, calculates the appearance frequency (S2), and uses the word / phrase with high frequency as the related word / phrase. Assuming that the documents including the search request phrases and the concatenated phrases of these related phrases are grouped for each related phrase (S3) and displayed in descending order of appearance frequency (S4).
[0024]
FIG. 3 shows a specific example of search result display processing by this method.
[0025]
FIG. 4 is a flowchart showing another example of the document search and search result display processing, here an example in which the adjacent phrase of the search request phrase in the history of the search request phrase is a related phrase.
[0026]
That is, based on the search request phrase input by the user in the search request input processing unit 1, the document search processing unit 2 searches for a document including the search request phrase from the set 4 of search target documents and outputs it as a search result document. After that (S1), the search result display processing unit 3 extracts a phrase adjacent to the search request phrase from the history of the search request phrase (a set of search request phrases input and used in the past), and its appearance The frequency is calculated (S11), the phrase having the high frequency is regarded as the related phrase, and the documents including the search request phrase and the concatenated phrase of these related phrases are grouped for each related phrase (S3). Display in order from the highest (S4).
[0027]
FIG. 5 shows a specific example of search result display processing by this method.
[0028]
At this time, the history range of the target search request phrase is limited to a certain period (for example, a range of 72 hours, etc.) retroactively from the time when the search request phrase is input, and the appearance frequency of the corresponding phrase from this is limited. By obtaining the above, it is possible to accurately follow the change in the importance of the search request phrase due to factors such as fashion.
[0029]
【The invention's effect】
The effect of the present invention is that, firstly, by extracting a related phrase for the search request phrase and performing classification related thereto, classification processing can be performed in a shorter time compared to a technique such as clustering. Second, In other words, it is possible to follow an event in which the importance of the search request word changes reflecting the fashion or the like, and it becomes possible to classify the search request word according to the words in line with the interests of many users.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an example of an embodiment of a document search apparatus of the present invention. FIG. 2 is a flowchart showing an example of a document search and search result display process. FIG. 3 is a search result display according to the process flow of FIG. FIG. 4 is a flowchart showing another example of document search and search result display processing. FIG. 5 is an explanatory diagram showing a specific example of search result display processing according to the processing flow of FIG. Explanation of]
1: search request input processing unit, 2: document search processing unit, 3: search result display processing unit, 4: set of search target documents.

Claims (3)

A document search apparatus for searching and presenting a document including a desired phrase from a set of search target documents stored in advance,
A search request input means for receiving a phrase input by a user as a search request phrase;
A document search means for searching and outputting a document including a search request phrase input by the user from a set of search target documents;
A phrase adjacent to the search request phrase input by the user is extracted from the history of search request phrases that are a set of search request phrases that have been input and used in the past, and the appearance frequency is calculated. Of the documents output by the document search means, the search request words and their related words and phrases are output from the document search means, assuming that the frequently used words and phrases are related to the search request words inputted by the user. And a search result display means for displaying a document including a concatenated word phrase and a group for each related word phrase. A document search program for searching and presenting a document including a desired phrase from a set of search target documents stored in advance,
The program is stored on the computer
A search request input step for receiving a phrase input by a user as a search request phrase;
A document search step of searching and outputting a document including a search request word input by the user from a set of search target documents;
A phrase adjacent to the search request phrase input by the user is extracted from the history of search request phrases that are a set of search request phrases that have been input and used in the past, and the appearance frequency is calculated. A search request phrase entered by the user and these related phrases in the searched and output document are regarded as a related phrase having a high degree of association with the search request phrase input by the user. And a search result display step of displaying a document including a concatenated word and a grouped display for each related word.   A computer-readable medium having the document search program according to claim 2 recorded thereon.

Images