JPH07129588A - Pattern-guided document content analyzer - Google Patents

Abstract

PURPOSE:To analyze document contents without preparing a dictionary for processing natural language. CONSTITUTION:A document content pattern extracting means 1 extracts a document content pattern 5 from the contents of a document described for a purpose similar to that of a subject document. A document content compressing pattern 2 extracts a similar pattern in the document content pattern 5 and compresses the document content pattern 5. A document content analyzing pattern 3 analyzes a newly inputted sentence 7 through the use of a document content pattern 6 compressed by the document content compressing pattern 2 and outputs an analytic result 8.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document content analysis device for analyzing the content of a document in an office.

[0002]

2. Description of the Related Art Conventionally, when analyzing document contents, a dictionary required for the analysis is prepared in advance, and the document contents are analyzed according to the analysis knowledge held by the dictionary. Therefore, there are cases where contents that do not appear in a specific document are registered as knowledge in the dictionary, or necessary knowledge is not described in the dictionary.

[0003]

When analyzing document contents, it is laborious to prepare a dictionary required for the analysis in advance, and it is very difficult to construct necessary and sufficient dictionary knowledge.

It is an object of the present invention to provide a document content analysis apparatus capable of analyzing document content without preparing a dictionary for natural language processing.

[0005]

A document content analyzing apparatus of the present invention extracts a document content pattern indicating a characteristic of a target document content from the content of a document described for the same purpose as the document content. Using a pattern extracting means, a document content pattern compressing means for extracting a similar pattern from the extracted document content patterns and compressing the document content pattern, and a document content pattern compressed by the document content pattern compressing means , And has document content analysis means for analyzing the newly input document content.

[0006]

When the document contents are analyzed, the information necessary for the document analysis is appropriately extracted from the contents of the document created for the same purpose as the target document, and the extracted text expression peculiar to the target document is extracted. Hold and use the information to analyze the document contents. Therefore, the document contents can be analyzed without preparing a dictionary for natural language processing.

[0007]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 1 is a block diagram of a pattern-guided document content analyzing apparatus according to an embodiment of the present invention, FIG. 2 is a flow chart showing the operation of the document content pattern extracting means 1, and FIG. 3 is an operation of the document content pattern compressing means 2. A flow chart showing
FIG. 4 is a flowchart showing the operation of the document content analysis means 3.

The document content pattern extraction means 1 extracts the document content pattern 5 from the content 4 of the document described for the same purpose as the target document. The document content pattern compression unit 2 extracts a similar pattern from the extracted document content patterns 5 and compresses the document content pattern 5. The document content analysis unit 3 analyzes the newly input document content 7 using the document content pattern 6 compressed by the document content pattern compression unit 2 and outputs an analysis result 8.

Next, the operation of this embodiment will be described with reference to a concrete example.

Here, a newspaper article is considered as a target document. Newspaper articles are edited and published for the general public on what happened in the company. In this case, the purpose of the newspaper article may be considered to be constantly constant in that it is written to convey an event that the general public should know.

In the document content pattern extraction means 1, first,
Sentence 4 is input (step 11), the input sentence 4 is divided into clauses (step 12), and the clauses appearing in the document are processed according to whether they are proper nouns or other proper nouns. (Steps 13 to 17). If the phrase appearing in the document is a proper noun, depending on the function of the proper noun,
A distinction is made between an agent (main body) and an object (target) (step 14). Usually, in newspaper articles, etc., ag
When acting as an ent, information related to the subject is added in brackets after the proper noun. For example, if it is a company name, who is the president, where is the address of the head office,
Information about what the representative telephone number is is written in brackets. From this information, it is automatically determined that the preceding clause in parentheses is a proper noun, which acts as an agent. If it is an object, the proper noun is rhetorical using brackets. Proper nouns are automatically judged by these rhetoric. If it is not a proper noun, the word that most appropriately expresses the meaning of each clause is extracted (step 15).

Input sentence 4: Company A (President X, Y city, Tokyo, telephone number Z) transfers from "B" Toyohashi factory to Maebashi factory.

[0014] In the following, it is divided into the following clauses (separation of clauses is represented by "/").

Company A (President X, Y city, Tokyo, telephone number Z)
/ Transfer "B" from Toyohashi Plant to Maebashi Plant.

Morphological analysis can be used to divide the bunsetsu, but the place where the character type is changed from hiragana to kanji or symbol can be used as the bunsetsu delimiter. After the phrase segmentation is completed, it is judged whether each segment is a proper noun (step 13). In this example, "A company (President X, Y city, Tokyo, telephone number Z)" is a proper noun, and "B" is a proper noun, and agent and object are respectively added (steps 16 and 17). , Except for proper nouns, "Toyohashi Factory", "Maebashi Factory", "Transfer". The three most appropriate words are extracted from these, and the word that most appropriately expresses the meaning of each word is extracted and assigned to the word (step 15). Usually, the last word in the word and the adjunct word mean the word. This assumption can be used to automatically process clauses other than proper nouns. With the input sentence 4 above, the following document content pattern 5 can be obtained. Is extracted.

Extracted document content pattern 5: agent /
object / from factory / to factory / transfer In the same manner, the document content pattern 5 is extracted from a similar newspaper article and stored (steps 18 and 19).

The document content pattern compression means 2 compresses the document content pattern 5 extracted by the document content pattern extraction means 1 (step 21). In compression processing, agent, ob
If there are multiple jects, etc., they are treated as one.
Furthermore, in order to deal with the degree of freedom of word order peculiar to Japanese,
Regardless of the order of the appearing clauses, if the appearing clauses are the same, it is judged as the same pattern and compressed. Further, when there is an omitted clause, if the other parts of the pattern match, they are compressed as the same pattern. The compressed document content pattern 6 is registered in the device (step 2).
2). Here, the following three pattern examples are determined to be the same.

Agent / agent / object / object / from factory / to factory / transfer from factory / agent / object / to factory / transfer agent / object / to factory / transfer In the document content analysis means 3, the document content pattern compression means 2
The document content is analyzed using the document content pattern 6 compressed in. First, the sentence 7 is input (step 31), and the input sentence 7 is divided into sentence units in the same manner as the processing of the document content pattern extraction means 1 (step 32), and each sentence is matched with the document content pattern 6. Take (Step 3
3). In the matching, as in the case of compression, regardless of the order of the appearing phrases, if the appearing phrases are the same, it is determined that they match. In addition, if there are omitted clauses,
If the other parts of the pattern match, it is determined that they match.

Input sentence 7: Transferred to the Mihashi factory for large bridges.

The input sentence 7 is divided into clause units: for a large bridge / to / transferred to the Mitsuhashi factory.

Matched document content pattern 6: agen
t / object / from factory / to / factory / matching part: to Mitsuhashi factory = transferred to factory = transferred The document content analysis means 3 finally sets the part matching the document content pattern 6 as the analysis result 8 of the document content. Output (step 34).

Output (analysis result 8): Transferred to the Mitsuhashi factory.

[0024]

As described above, according to the present invention, the document content pattern indicating the characteristics of the target document content is extracted from the content of the document described for the same purpose as the target document and is similar. By compressing the document content pattern and registering it in the device, and analyzing the newly input document content using the registered document content pattern, the document content can be stored without manually creating a large-scale dictionary. There is an effect that it is possible to analyze automatically.

[Brief description of drawings]

FIG. 1 is a block diagram of a pattern-guided document content analysis apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the document content pattern extraction means 1.

FIG. 3 is a flowchart showing the operation of the document content pattern compression unit 2.

FIG. 4 is a flowchart showing the operation of the document content analysis means 3.

[Explanation of symbols]

 1 Document Content Pattern Extracting Means 2 Document Content Pattern Compressing Means 3 Document Content Analyzing Means 4 Input Statements 5 Document Content Patterns 6 Compressed Document Content Patterns 7 Input Statements 8 Analysis Results 11-19, 21, 22, 31-34 Steps

Claims (2)

[Claims] 1. A document content pattern extracting means for extracting a document content pattern indicating a characteristic of a target document content from the content of a document described for the same purpose as the document, and among the extracted document content patterns. A document for analyzing a newly input document content by using a document content pattern compression unit for extracting a similar pattern by using the document content pattern compression unit and a document content pattern compressed by the document content pattern compression unit A pattern-guided document content analysis device having content analysis means. 2. The document content pattern extraction means divides an input sentence into phrases, and for each of the divided phrases, extracts a word representing the meaning of the phrase from the phrase according to the function of the phrase. Then, the document content analysis unit divides the input sentence into units of phrases, and matches each divided phrase with the document content pattern compressed by the document content pattern compression unit, The document content analysis apparatus according to claim 1, wherein the matched portion is output as an analysis result.