JPH07219953A - Document summarizing device - Google Patents

Abstract

PURPOSE:To output contents to be a point of an original document as a summarized document by checking a conception system by using important words extracted from the original document and extracting an important concept node managing many important words from the original document. CONSTITUTION:A conception system 1 for an event or operation is retrieved by using important words (2) extracted from an original document (1) by prescribed technique and how many extracted important words (2) is managed by a concept node (3) for managing previously set individual words to the important words in the system 1 is checked. A concept node (3) capable of managing more important words (2) is selected and extracted as an important concept node (3a) to obtain a point indicating what the original document (1) means. Then work such as the addition of a proper sentence to the concept in the node (3a) is executed to summarize the original document (1).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention checks the concept system in which the concept of things, actions, etc. is expressed in a tree structure from the input original document by using the important words selected by a predetermined method. The present invention relates to a document summarizing device for selecting an important concept node a indicating the concept of an original document and summarizing the original document using the selected concept of the important concept node a.

[0002]

2. Description of the Related Art FIG. 6 is a diagram for explaining a conventional document summarizing device. First, the original document is converted into a morphological analysis unit (input unit).
2 is input, the original document is divided into morphemes by using the syntax of the original language of the original document, and the important word extraction unit 3 extracts from the divided morphemes, ~ "ha", ... " The words in front of the particle such as "to" are extracted as important words.

Next, the appearance frequency analysis unit 4 extracts the important word (word) having the highest frequency among the extracted important words as the most important word a, and the extracted most important word a and the important word a. The sentence including the word is treated as an important sentence in the document summarization unit (output unit) 7, and the document is summarized.

As a concrete summarizing method of the document, a summary sentence is formed by the extracted most important word a and a simple enumeration of important words, or a processed sentence suitable for the extracted most important word a. Is added, the "original document is a document regarding" ... ". There is a method such as doing a summary.

[0005]

In the conventional document summarizing apparatus described above, the word having the highest appearance frequency is extracted from the extracted important words by using only the syntax of the language of the original document.
Selected as the most important word a. The most important word a extracted in this way is simply an important word extracted only from the surface layer such as the syntax of the original document. To select the important word, the meaning of the entire original document is checked. Not not.

Therefore, the most important word a extracted as described above may not be an appropriate most important word a. There is a problem in that a document that is summarized centering on this unsuitable most important word a is not necessarily an appropriate summary for the original document. That is, there was a problem that the original document did not extract the point of what it meant.

A similar technique is disclosed in Japanese Patent Laid-Open No. 63-17.
There is a document processing device in Japanese Patent No. 5965, in which a keyword is extracted from the meaning content of a term in a document, a concept candidate representing the content of the document is selected by referring to the extracted keyword,
Based on this, the concept of the document is decided.

However, in the document summarization technique disclosed in Japanese Patent Laid-Open No. 63-175965, "Document Processing Device",
There is a problem in that it is necessary to recreate a dictionary and a concept keyword for each field depending on the field of documents, for example, the field of paperwork such as “employment”, “shipment”, and “remittance”.

In view of the above-mentioned conventional drawbacks, the present invention is a general or document technical field in which concepts of things, actions, etc. are expressed in a tree structure using important words selected by a predetermined method. A document summarization device that checks a concept system that does not depend on the original document, extracts the point to be said in the original document as a concept, uses the extracted concept as the above-mentioned most important word, and performs the most appropriate summary for the original document. It is intended to provide.

[0010]

FIG. 1 is a principle configuration diagram of the present invention, and FIG. 2 is a diagram showing an example of a concept system used in the present invention. The above problems are solved by the document summarizing device configured as follows.

(1) The document summarizing device according to the present invention includes an input unit 2 for inputting an original document, an important word extracting unit 3 for extracting an important word from the original document by a predetermined method, and an extracted important word. The concept system 1 that represents the concept of things, actions, etc. in terms of words in a tree structure is searched, and the important concept node selection unit 6 that selects the important concept node a and the concept of the selected important concept node a are It is composed of an output unit 7 that outputs a summary document of the original document.

(2) The important concept node selection unit 6 of the document summarizing device searches for a node on the concept system 1 that controls a certain number of words, and each node extracts the number of important words. It is configured to check whether to dominate and select a node that dominates more important words as the important concept node a.

(3) When the number of important words governed by the nodes on the concept system 1 is the same, the important concept node selection unit 6 of the document summarizing device causes the dominant words to appear in the original document. The frequency is checked, and the node that controls the important word with the higher appearance frequency is selected as the important concept node a.

[0014]

With the document summarizing device according to the present invention, the concept system 1 in which the concept of things, actions, or the like is expressed in a tree structure is searched by using the important word (word) extracted from the original document by a predetermined method. However, for the words that correspond to the important words in Concept System 1, check how many of the extracted important words do the concept nodes that dominate a preset number of words in a tree structure, and A concept node that governs many important words is selected as an important concept node a, this important concept node a is extracted to obtain a point of what the original document is trying to say, and the important concept node a Using the concept, the original document is summarized, for example, by processing such as adding an appropriate sentence to the concept.

Regarding the above concept system 1, for example,
An example can be seen in "Classified Vocabulary Table", National Institute of Japanese Language Books 6, National Institute for Japanese Language Studies, published by Hideei Publishing Co., Ltd., March 31, 1964, first edition issued. As shown in
It is a tree structure that expresses the concept of things or actions.
There are the above things or actions in the highest node,
In the lower node, the lower concept that constitutes the higher concept is represented in a hierarchical structure, for example, in a tree structure format. The words (important words) that each concept governs belong to the lowest layer.

There are several known methods for extracting important words (words) from the original document. For example, "" automatic abstraction method ", Yasushi Iwabuchi, Mikio Arai, Minoru Aizawa,
IEICE Communications Research Group, CMN-89-23 ”.

In the above-mentioned documents, 1) select words constituting the title and subtitle. 2) "ha", "to" in the text,
Select the word consisting of kanji and kana characters before "toha". This is because the above "ha" is used when describing the subject. 3) If there is "to" or "ya" in front of the word extracted in 2) above, if the word before that also consists of Kanji or Kana characters, it is extracted as an important word. 4) The word extracted as above is
If it is a single letter, it is deleted from the rule of thumb as it cannot be an important word. Such a method is shown.

In the document summarizing apparatus according to the present invention, the method of extracting the important word from the original document is not limited, but the important word can be extracted by using the method shown in the above-mentioned document, for example. Extract.

Therefore, with respect to the input original document, the important word is extracted by using some method, and the extracted important word is used for the above-mentioned prebuilt construction, or A concept system that expresses concepts such as actions in a tree structure 1
Is searched to find out how many of the extracted important words are controlled by a node that controls a specified number of words, and a node that controls more important words is selected as an important concept node a. By doing so, the highest point of the original document can be obtained, and the original document can be summarized by an appropriate document.

[0020]

Embodiments of the present invention will be described in detail below with reference to the drawings. The above-mentioned FIG. 1 is a block diagram of the principle of the present invention.
FIG. 3 is a diagram showing an example of a concept system used in the present invention, FIGS. 3 to 5 are diagrams showing an embodiment of the present invention, and FIG. 3 is an embodiment of a document summarization process. 4 is a flow chart, FIG. 4 shows an example of a concept system used in this embodiment, and FIG. 5 shows an example of selection of important concept nodes.

In the present invention, by using the syntax of the original document, important words (words) are extracted, and in order to select the most important word a from the extracted important words, things or actions are performed. The concept system 1 in which the concepts such as are expressed in a tree structure are searched, nodes that control a certain number of words are compared, and the concept node with the most number of dominant words is selected as the important concept node a. A means for summarizing an original document using the selected concept of the important concept node a is a means necessary for implementing the present invention. The same reference numerals indicate the same objects throughout the drawings.

Hereinafter, referring to FIGS. 1 and 2, FIG.
The configuration and operation of the document summarizing device of the present invention will be described with reference to FIG. As shown in FIG. 1, the document summarizing device of the present invention includes an input unit 2 for inputting an original document and an important word extracting unit 3 for extracting an important word in the original document by some method.
Then, using the extracted important words, the concept system 1 is searched, and the concept nodes that dominate a preset number of words dominate the important words extracted by the important word extracting unit 3. Alternatively, by using the important concept node selecting unit 6 that selects the important concept node a corresponding to the original document by checking the appearance frequency of the important word, and the concept indicated by the selected important concept node a. It consists of an output unit 7 that summarizes and outputs the original document, and the concept system 1 above.

In the present embodiment, the original document used is, for example, "the profit of the textile export to the country B of the country A is not profitable when a heavy anti-dumping duty is applied. Increasingly, dumping tariffs are increasing. ”

FIG. 4 is an example of the concept system 1 used in this embodiment, which is composed of tree-structured concept nodes as shown in FIG. Important words (words) controlled by the concept node of are arranged.

In FIG. 3, first, when the original document is input to the input unit 2 of FIG. 1, the important word extracting unit 3 of FIG. 1 extracts the original document by the method described in the above-mentioned document. , Extract important words in original document.

In the above original document, for example, "profitability"
(1), "Recent" (1), and "Export" (2) are extracted as important words. Here, the number in () is the frequency of appearance. {Refer to processing step 100 in FIG. 3} Next, the important concept node selection unit 6 in FIG. 1 searches the concept system 1 shown in FIG. Extract the concept node containing the key word from system 1. FIG. 5A is an example of a concept node that controls the important word obtained by searching the concept system 1 of FIG. 4 based on the important word.

That is, the word "export" (2) is controlled by the concept node "export / import" (2) (1), and the concept node "export / import" is a superordinate concept node "transaction" (4). ) (1). Similarly, the word "recently" (1) is dominated by the concept node "time" (2) (1), and the word "profitable" (1)
Is governed by the concept node "Accounting" (2) (1). Here, the number after the word "export" or the like is the frequency of appearance of important words, and the next number such as the concept node "export / import" is the number of words that the concept node controls, to the right of it. The number is the number of important words that dominate. {Refer to processing step 101 in FIG. 3} Next, in the same important concept node selecting unit 6, a concept node that controls a certain number (for example, 2) of words is compared, and the concept that has the most important words is the most important concept. A node is selected as an important concept node a. In the above example,
As is clear from the example shown in FIG. 5 (a), the concept nodes that control the words “export”, “recent”, and “profit” are “import / export”, “hour”, and “accounting”. The number of dominant important words (words) is “1” and the same.

In this case, the important word (word) having the highest appearance frequency in the original document is checked, and the higher concept node is selected as the important concept node a. In this embodiment, as shown in FIG. 5B, the important concept node “export / import” is selected. {Refer to processing steps 103 and 104 in FIG. 3} In the output unit 7 in FIG. 1, for example, a sentence such as "document is related" is added to the concept "export / import" of the selected important concept node a (processing of document). ) Is performed, and the summary sentence “It is a document related to“ import / export ”” is output.

As described above, the document summarizing apparatus according to the present invention uses a plurality of important words selected by a predetermined method from the input original document to create a tree structure of a concept such as a thing or a motion. By searching the expressed concept system, the concept node having the largest number of important words in the original document is extracted as the important concept node a, and the original document is summarized by the extracted concept of the important concept node a. It is characterized by the fact that it is set.

[0030]

As described above in detail, according to the document summarizing apparatus of the present invention, the important words extracted from the original document are used to check the concept system, and more important items in the original document are checked. By extracting the important concept node a that governs a word and obtaining the meaning pointed to by the original document, there is an effect that the content that is the point of the original document can be output as a summary sentence.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention.

FIG. 2 is a diagram showing an example of a concept system used in the present invention.

FIG. 3 is a diagram showing an embodiment of the present invention (No. 1).

FIG. 4 is a diagram showing an embodiment of the present invention (part 2).

FIG. 5 is a diagram showing an embodiment of the present invention (part 3).

FIG. 6 illustrates a conventional document summarizing device.

[Explanation of symbols]

1 concept system 2 input part, morpheme analysis part 3 important word extraction part 4 appearance frequency analysis part 5 document summarization part, output part 6 important concept node selection part 100-105 processing steps original document important word (word) a most important word concept Node a Important concept node

Claims (3)

[Claims] 1. An input section (2) for inputting an original document (), and an important word extraction section (3) for extracting an important word () from the original document () by a predetermined method. Important word ()
Then, a key concept node selection unit (6) that selects a key concept node (a) by searching the concept system (1) that represents the concept of things or actions in a tree structure, and the selected key concept A document summarizing device comprising an output unit (7) for outputting a summarized document of an original document () using the concept of the node (a). 2. An important concept node selecting unit of the document summarizing device.
(6) is the above concept system that controls a certain number of words
(1) Search the above nodes (), check how many each of the nodes () dominates the important word (), and find more nodes () that dominate the important word (). The document summarizing device according to claim 1, wherein the document summarizing device is selected as the concept node (a). 3. An important concept node selecting unit of the document summarizing device
(6) is the dominant word that is dominant if the number of important words () governed by the node () on the conceptual system (1) is the same
The appearance frequency in the original document () of () is checked, and the node () that controls the important word () with the higher appearance frequency is selected as the important concept node (a). The document summarizing device according to claim 1,