JPH06103311A - Text base information retrieving device - Google Patents

Abstract

PURPOSE:To enable interactive information gathering and acquisition by handling the whole text data as a plane text and performing retrieval under functional conditions. CONSTITUTION:The information retrieving device equipped with an input means 1, a large-scale text data base 6, an information extracting means 5, and an output means 8 is equipped with a function relation extracting means 4 which extracts function units as units carrying information in respective texts in the large-scale text data base 6, a function unit index 3 which holds the function units extracted by the function relation extracting means 4 and the positions in the texts as indexes, a function unit retrieving means 2 which retrieves a function unit matching with an input in the function unit index 3, and a majority decision evaluating means 7 which extracts representative information by making a majority decision basing on the coincidence of the function unit when the amount of the text outputted by the information extracting means 5 is large.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information retrieving apparatus, and more particularly to an information retrieving apparatus having a function of retrieving information spanning a plurality of texts regardless of a unit such as a document or a chapter.

[0002]

2. Description of the Related Art A conventional text-based information retrieval method has a form in which a text explaining a certain specific item is organized using those items as indexes, as seen in examples of encyclopedias and dictionaries. Therefore, the search method is to search for this item name and obtain information from the text of its description. In the electronic encyclopedia that digitizes this, in addition to the conventional way of drawing, it is possible to draw even with partial matching with the character string of the index by taking advantage of the digitization. However, in this conventional text-based information retrieval method, it is essential that the information be organized by items, such as encyclopedias and dictionaries, and the disadvantage that this item can be organized only manually. is there. In addition, since it is necessary to manually organize information, there is a disadvantage that a large number of words already appearing in a text and often used cannot be searched from a text base until someone arranges them.

To solve this, Japanese Patent Laid-Open No. 25337/1990
As can be seen in No. 1, a natural sentence semantic analysis processing device that performs a semantic analysis of a natural language text, creates a regular pattern of a sentence, and searches for a similar sentence by checking a match with a user input. Proposed. However, this device is a device for observing the matching of input patterns when relatively similar inputs are repeated, and is not effective in a situation such as an encyclopedia in which a wide variety of information requested by a user.

[0004]

There is also a full-text search technique in which only word search is performed on all texts as an information search method that does not incur the cost of manual information arrangement. However, even if a search is performed using this method, there are the following problems. (1) Necessary information spans multiple texts and cannot be searched only with words (2) A large amount of information of the same type is searched and unnecessary information is repeated (3) A large amount of information is processed by the user I can not reach the target information because I can not do it

In order to solve the above-mentioned problems, an information retrieval apparatus of the present invention comprises an input means for accepting an information retrieval request from a user, a large-scale text database holding a large amount of text, and In an information retrieval device equipped with an information extraction means for extracting necessary information using an index that matches the input from a large-scale text database and an output means for presenting the extracted information to the user,
Functional relation extraction means for extracting a functional unit as a unit for carrying information of each text in the large-scale text database, and a functional unit extracted by the functional relation extraction means and its position in the text are held as an index. A functional unit index and a functional unit search means for searching the functional unit index for a functional unit matching the input,
When a large amount of text is output from the information extraction means, a majority decision evaluation means is provided for taking a majority decision according to the number of matching functional units and extracting representative information.

[0005]

In the text-based information search device according to the present invention, a plurality of texts are searched at the same time to provide information by using the functional relationship between words in the text, which is a unit that carries information as a unit of information search. . Therefore, even the smallest amount of information required by the user can be searched for even if it spans a plurality of texts. Furthermore, when a large amount of information of the same type is retrieved, those information are grouped with the same functional relationship, and the amount is considered to be the reliability and importance of the information, and it is provided in order from the largest. To do. By selecting information by this majority vote, a large amount of information of the same type is searched and unnecessary information is no longer repeated. Furthermore, by interactively using the search that presents this minimum amount of information, it is possible to search for the target information without giving the user all the information that cannot be processed at once.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the present invention.
With reference to FIG. 1, an embodiment of the present invention, an input means 1 for receiving a user's information search request, a large-scale text database 6 for holding a large amount of text, and information of each text in the large-scale text database 6 A functional relationship extracting unit 4 for extracting a functional unit as a unit for carrying out the function, a functional unit index 3 for holding the functional unit extracted by the functional relationship extracting unit and its position in the text as an index, and the input matched. A functional unit searching means 2 for searching a functional unit in the functional unit index, and an information extracting means 5 for extracting necessary information from the large-scale text database 6 by using an index matching the input.
When a large amount of text is output from the information extraction unit 5, a majority decision evaluation unit 7 that extracts a representative information by making a majority decision based on the number of matching functional units, and presents the extracted information to the user. Output means 8 for

Regarding the text registered in the large-scale text database 1, the functional relationship extracting means 4 does not look at the text as a document unit but as a whole plain text, and regards it as a unit for carrying information. The functional unit is extracted and registered in the functional unit index 3. The functional relationship extracting means 4 may be activated at the time when the functional unit index 3 is accessed, but practically it is preferable that it be activated when a text is registered in the large-scale text database 6. In the functional unit index 3, indexes are grouped for the same functional unit, so that indexes of different places of a plurality of texts are often held for the same functional unit.

As the functional unit, a ternary relation of word pairs and the relation between them is used. It is considered that this ternary relationship is a unit indicating the functional meaning of a pair of words. As the simplest functional unit, there is one in which a pair of words are two independent words having a dependency relationship and a relationship is a relationship between those words (case particle, connecting particle, connecting particle, etc.).
When these independent words are represented by A and B, and the relation between them is represented by R, the functional units are ABC: eg, Gulf War-participated in Gulf War-allied warfare-and-bombing It is expressed as done. Further, more complicated relations between words can be expressed by connecting the functional units. A-R1-B-R2-C → A-R3-C (R3 =
B) Example: Gulf War-Participated-USA This functional unit connection is not enough that the words of B are simply the same, and the context in which the relationship between A and B is stated is the relationship between B and C. Concatenates only in the sentence in which is stated.

Next, the operation when the user inputs a search request will be described. In the user's input obtained by the input means 1, first, the functional unit search means 2 extracts the functional relationship required by the user by using the functional unit. A text is extracted from the large-scale text database 6 by the information extracting means 5 with an index of a functional unit that matches this functional relationship. The character string length of the extracted text is not the entire text, but the length of the character string of a unit that carries information is appropriate, and it is conceivable to use a unit of one sentence. Explaining with an example, in response to a search request "Where participated in the Gulf War", the functional unit "Gulf War-participated in the United States" was searched, and by tracing the index, it became the source of the functional unit. The text "The United States participated in the Gulf War" is displayed.

When the amount of text extracted by the information extraction means 5 is too large and it is considered to be unsuitable for presentation to the user, the majority evaluation means 7 matches the texts in functional units. To take a majority vote and present it to the user from the representative opinion. If the amount of text extracted by the information extraction means 5 is not so large, the majority decision evaluation means 7 does nothing. The information selection by the majority evaluation avoids giving the amount of information that the user cannot process at one time, and presents the most universal information. Moreover, if a large amount of text is added as an unlimited information source, there will inevitably be erroneous descriptions and unnecessary noise information, but the selection of information based on this thickness plays a role in removing these. Is also playing.

Next, referring to FIG. 2, a search process in the text-based search apparatus will be described. FIG. 2 is a conceptual diagram showing a search process by this device. First, suppose the user wants to retrieve information about the Gulf War. As users knew the United States was involved in the Gulf War,
Enter "What happened to the United States in the Gulf War?" With this input, the functional unit search means searches the functional unit index for the functional unit "Gulf War-US-War". Further, as the original text from which the functional unit is extracted, "the United States has warned Iraq in the Gulf War" is retrieved by the information extracting means and presented to the user. The user sees this text, recalls the memory that the United States was the center of the Allied Powers, and inputs it as a question. Then, it is searched that "the United States was the main allies in the Gulf War." By repeating this process, it becomes possible for the user to know that "electronic weapons were used for the first time in the Gulf War."

[0012]

As described above, in the text-based information retrieval apparatus according to the present invention, the text is not viewed as a document unit, but the entire text is viewed as plain text, and an index for identifying the position of the text is prepared. The text corresponding to the user's information search request can be
It is possible to search from the information held in the entire database. In addition, since the index of each functional unit is automatically created by the functional relationship extracting means when the text is registered in the large-scale text database, there is a feature that the cost of creating the index is low. Another feature is that the text that describes the latest information can be searched and used by simply registering it in a large-scale text database.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a conceptual diagram showing a search process according to the present invention.

[Explanation of Codes] 1 Input Means 2 Functional Unit Searching Means 3 Functional Unit Indexes 4 Functional Relationship Extracting Means 5 Information Extracting Means 6 Large Scale Text Databases 7 Majority Evaluating Means 8 Output Means

Claims (1)

[Claims] 1. Input means for receiving an information search request from a user, a large-scale text database holding a large amount of text, and information extraction for extracting necessary information from the large-scale text database by using an index matching the input. Means, and an information retrieval device equipped with an output means for presenting the extracted information to the user, a functional relationship extracting means for extracting a functional unit as a unit for carrying the information of each text in the large-scale text database, A functional unit index that holds the functional unit extracted by the functional relationship extracting unit and its position in the text as an index; a functional unit searching unit that searches for a functional unit that matches the input in the functional unit index; If there is a large amount of text output by the information extraction means, An information retrieval apparatus comprising a majority decision evaluating means for taking a majority decision and extracting representative information.