JP2546515B2 - Information extraction device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention analyzes a document described in natural language, extracts information in a predetermined field, and outputs information including a relationship between words included in the document in a fixed format. Information extraction system.

[0002]

2. Description of the Related Art Information extraction methods for extracting information in a specific field from a text up to a relationship between words and outputting the information in a fixed format have hitherto been used, such as a method of utilizing the appearance state of a keyword,
There was a method of using parsing. The method of using a keyword is a method of previously providing a keyword related to the field of information to be extracted and a format to be output, and performing information extraction based on the appearance and co-occurrence of the keyword in the input document. However, since this method ignores the sentence structure, it often extracts inappropriate information that includes keywords but does not establish a correct relationship between words. Further, the method using the syntactic analysis is to re-parse the syntactic tree obtained by the syntactic analysis to generate a certain interpretation tree that does not depend on the meaning. However, since the target of analysis is a syntax tree, the extraction result will be different due to a slight difference in the syntax tree.

[0003]

The method using the appearance of keywords can output a large amount of information to be extracted by increasing the number of keywords provided. However, since the output is determined by simply ignoring the sentence structure and simply appearance of the keyword, the extracted output contains a lot of inappropriate information that does not establish the relationship between words. In addition, information could not be extracted from words that were not registered as keywords. The method using the syntactic analysis recognizes the syntactic structure of the sentence so that the correct extraction result can be obtained. However, it was difficult to get the correct parse tree because the ambiguity in the syntax could not be resolved only by the syntax analysis technology. Therefore, when it is applied to an actual document, there are many cases where a document including information cannot be extracted.

It is an object of the present invention to extract a large number of correct information by performing a syntactic analysis while recognizing a relationship between keywords.

[0005]

According to a first aspect of the present invention, a document input section for receiving a document described in natural language as an input, a morpheme and a morpheme dictionary section in which syntax information is written for each morpheme are predetermined. The keyword related to the field of information to be extracted and the keyword dictionary describing the role of each keyword in the format to be finally output for each keyword, and the sentence input from the document input unit is cut into words, A morpheme dictionary unit, a morpheme analysis unit that allocates the dictionary contents of the keyword dictionary unit to each word, and a syntactic analysis rule that stores rules for performing syntactic analysis of an input sentence using the syntactic information stored in the morpheme dictionary unit. The storage unit and the key that stores the rules that generate the semantic structure showing the inter-keyword relationships while controlling the parsing rules by the keyword information given to the keywords. Using the interword relation calculation rule storage unit, the syntactic analysis rule stored in the syntactic analysis rule storage unit and the inter-keyword relation calculation rule stored in the inter-keyword relation calculation rule storage unit, the morpheme analysis unit uses morphemes. A document information extraction unit that analyzes the parsed sentence and outputs a semantic structure indicating a relationship between keywords, and an extraction result output unit that converts the semantic structure of the entire document output by the document information extraction unit into an output format and outputs the result. It is characterized by having.

In a second aspect based on the first aspect, a keyword estimation rule storage unit for estimating a morpheme sequence that does not exist in the keyword dictionary unit as a keyword from a syntactic structure, and a syntax stored in the syntactic analysis rule storage unit. Analyzing the sentence morphologically analyzed by the morpheme analysis unit using the inter-keyword relation calculation rule stored in the keyword relation calculation rule storage unit and the keyword estimation rule stored in the keyword estimation rule storage unit However, it is characterized in that it is provided with a document information extraction unit that outputs a semantic structure indicating a relationship between keywords.

[0007]

The present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the first invention. Referring to FIG. 1, the present invention includes a document input unit 1 that receives a document described in natural language as an input,
A morpheme and a morpheme dictionary section 2 in which syntax information is written for each morpheme, a keyword related to a predetermined field of information to be extracted, and a role of each keyword in a format to be finally output for each keyword The keyword dictionary 3, the sentence input from the document input unit 1, is word-cut, and the morpheme dictionary unit 2 and the morpheme analysis unit 4 assigns the dictionary contents of the keyword dictionary unit 3 to each word; and the morpheme dictionary unit. A parsing rule storage unit 5 that stores rules for parsing an input sentence using the syntax information stored in 2.
And the inter-keyword relationship calculation rule storage unit 6 in which the rules for generating a semantic structure indicating the inter-keyword relationship are stored while controlling the syntactic analysis rule according to the keyword information given to the keyword, and the syntax analysis rule storage unit 5. Meaning indicating the inter-keyword relationship by analyzing the sentence morphologically analyzed by the morphological analysis unit 4 using the stored syntactic analysis rule and the inter-keyword relation calculation rule stored in the inter-keyword relation calculation rule storage unit 6. It is composed of a document information extraction unit 7 that outputs a structure and an extraction result output unit 8 that converts the semantic structure of the entire document output by the document information extraction unit 7 into an output format and outputs the output format.

FIG. 2 is a block diagram showing an embodiment of the second invention. Referring to FIG. 2, the present invention is provided with a keyword estimation rule storage unit 9 for estimating a morpheme sequence that does not exist in the keyword dictionary unit 2 as a keyword from a syntactic structure, in addition to the first invention. At this time, the document information extraction unit 7 adds the syntax analysis rules stored in the syntax analysis rule storage unit 5 and the inter-keyword relationship calculation rules stored in the inter-keyword relationship calculation rule storage unit 6 to the keyword estimation rule storage unit. Using the keyword estimation rules stored in 9,
The sentence subjected to morpheme analysis by the morpheme analysis unit 4 is analyzed, and a semantic structure indicating a relationship between keywords is output.

Next, the operation of the embodiment of the present invention will be described with reference to FIGS.

As an embodiment of the present invention, consider a case of extracting information as to where a layering device, which is a semiconductor manufacturing process, is developed, manufactured, sold, or used. Further, assume that an output format as shown in FIG. 3 is given. Here, the relationship column shown in FIG. 3 contains some of the items that apply from development, manufacturing, sales, and use, and companies and devices are filled with companies and devices in the relationship column.

For example, assume that the sentence "NEC has developed a sputtering device" is input from the document input unit 1. This sentence is divided into words by the morpheme analysis unit 4, and the dictionary information of the morpheme dictionary unit 2 and the keyword dictionary unit 3 is given to each word. Here, "NEC", "CVD
When “device” and “development” are keywords of a company name, a device name, and a relationship, respectively, it can be recognized from the syntactic structure that these three keywords are related, and the output of FIG. 3 can be obtained.

Further, in the method using the keyword, the information in FIG. 2 cannot be extracted if, for example, "NEC" is not registered as the company name keyword. On the other hand, in the present invention, the character string “NEC” stands for the subject of the sentence “developed the device” according to the keyword estimation rule stored in the keyword estimation rule storage unit 9, and thus the device developer. Therefore, it is possible to approve “NEC” as the company name. Therefore, even if "NEC" is not registered as a keyword, the output of FIG. 3 can be obtained.

Further, even a sentence in which information is erroneously extracted only by the appearance of a keyword can be recognized as having no information to be extracted by using the syntactic structure.

For example, the sentence "NEC has developed a sputtering material" includes three types of keywords "NEC", "sputtering", and "development", and it is judged only by the appearance of the keywords. The information of FIG. 3 is extracted. However, the device keyword "sputtering" is related to "material", and NEC developed the material. In the present invention, it is possible to detect that "NEC has developed a material" from the syntactic structure.
It can be certified that it was not a layering device that was developed as shown in.

Further, the present invention can reduce the ambiguity generated during parsing by utilizing the lexical knowledge contained in the keyword.

For example, the two sentences 1) "NEC has developed a sputtering apparatus and a CVD apparatus" and 2) "NEC has developed a CVD apparatus with Sumitomo Metal Industries" indicate that the syntactic structure is ambiguous. Shows. That is, although both have the same case particle alignment as "A developed B and C," in 1), B is C
In 2), B is involved in the development together with A as an adjoint case. These ambiguities cannot be resolved by grammatical information alone. On the other hand, in the present invention, “NEC” and “Sumitomo Metal Industries” are the company keywords, “sputtering equipment”,
“CVD equipment” can be recognized as a device keyword. Therefore, by using the inter-keyword relation rule that parallel structures are formed from the same kind of keywords, as shown in FIG. 5, “sputtering device” and “CVD device” are used in 1) and “NEC” is used in 2). And "Sumitomo Metal Industries" can be certified as parallel.

[0018]

According to the present invention, many correct outputs can be extracted by giving sufficient keywords. Moreover, since the information to be extracted is generated as a result of recognizing the relationship between the keywords, it is possible to reduce the output of erroneous extraction results. Further, even if the given keywords are insufficient, it is possible to identify the word that should be the keyword from the syntactic structure, so that more correct output can be extracted. Furthermore, since the semantic structure consisting of the relationships between the keywords is generated while performing the syntactic analysis, the syntactic ambiguity can be greatly reduced. Since syntactic analysis uses vocabulary knowledge contained in keywords,
The ambiguity can be reduced by controlling the syntax rules. Also, words that are not registered as keywords can be inferred as keywords from the syntactic structure, and information can be extracted for these as well.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a block diagram according to an embodiment of the first invention.

FIG. 2 is a diagram illustrating a block diagram which is an embodiment of the second invention.

FIG. 3 is a diagram illustrating input / output according to an embodiment of the present invention.

FIG. 4 is a diagram illustrating input / output according to an embodiment of the present invention.

FIG. 5 is a diagram illustrating input / output according to an embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 document input unit 2 morpheme dictionary unit 3 keyword dictionary unit 4 morpheme analysis unit 5 syntax analysis rule storage unit 6 keyword relation calculation rule storage unit 7 document information extraction unit 8 extraction result output unit 9 keyword estimation rule storage unit

Claims (2)

(57) [Claims] 1. A document input section for receiving a document described in natural language as an input, a morpheme dictionary section in which syntactic information is written for each morpheme, and a keyword relating to a predetermined field of information to be extracted. For each keyword, a keyword dictionary that describes the role that the keyword plays in the format to be finally output, and a sentence that is input from the document input unit is cut into words, and the dictionary of the morpheme dictionary unit and the keyword dictionary unit is cut. A morphological analysis unit that assigns contents to each word, and a syntactic analysis rule storage unit that stores rules for parsing an input sentence by using the syntactic information stored in the morpheme dictionary unit, and a keyword given to a keyword A keyword relation calculation rule storage unit that stores a rule for generating a semantic structure indicating a keyword relation while controlling a parsing rule by information. , Using the syntactic analysis rule stored in the syntactic analysis rule storage unit and the inter-keyword relation calculation rule stored in the inter-keyword relation calculation rule storage unit,
Analyzing the sentence morphologically analyzed by the morpheme analysis unit, and a document information extraction unit that outputs a semantic structure indicating a relationship between keywords, and convert the semantic structure of the entire document output by the document information extraction unit to an output format, An information extraction device comprising an extraction result output unit for outputting. 2. A keyword estimation rule storage unit that estimates a morpheme string that does not exist in the keyword dictionary unit as a keyword from a syntactic structure, a syntactic analysis rule stored in the syntactic analysis rule storage unit, and the inter-keyword relation calculation rule storage. Using the keyword relation calculation rule stored in the section and the keyword estimation rule stored in the keyword estimation rule storage unit, the sentence morphologically analyzed by the morpheme analysis unit is analyzed, and a semantic structure indicating a keyword relation is determined. The information extracting apparatus according to claim 1, further comprising a document information extracting unit for outputting.