JPH0877196A - Extracting device for document information - Google Patents

Abstract

PURPOSE: To highly precisely extract a desired sentence containing important information by collating the sentence with a sentence expression pattern on the basis of the mark and the part-of-speech information of the morpheme of every sentence having been morpheme-analyzed, and extracting the coincident sentence. CONSTITUTION: A document information extracting device 2 morpheme-analyzes an input document 1, and extracts the sentence coincident with the sentence expression pattern 5 on the basis of the mark, the part-of-speech information and the meaning information of the morpheme, and it is constituted of a morpheme analyzing part 3, a sentence pattern collating part 4 and the sentence expression pattern 5. Then, the morpheme analyzing part 3 morpheme-analyzes the input document 1, and the sentence pattern collating part 4 executes the collation with the sentence expression pattern 5 on the basis of the mark and the part-of-speech information of the morpheme of every morpheme-analyzed sentence, and extracts the coincident sentence. Accordingly, since the result of the morpheme analysis of the inputted document 1 is collated with the sentence expression pattern 5, and at the time of coincidence, the sentence is extracted as the desired sentence, the sentence in which the improtant information is contained can be highly precisely extracted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document information extracting apparatus for morphologically analyzing a document and matching a sentence expression pattern based on morpheme notation, part-of-speech information and semantic information to extract a desired sentence. .

[0002]

2. Description of the Related Art A conventional document extracting apparatus uses a character string when extracting a sentence containing a certain expression. For example, when it is desired to extract a sentence containing the expression "○ means XX" (○ is an arbitrary character), the sentence containing the character string "TOHA" was searched and extracted.

[0003]

[Problems that the Invention is to try to solve this case, is a problem that the statement containing the string "and is" in addition to the sentence that contains the expression you want to extract the original, such as "can not and editing child 'is being extracted there were.

Further, when it is desired to extract a sentence such as "when the e command is entered, the process can be terminated", "when you can do XX, you can do XX", the word immediately before "to" is "do XX". I want to specify the condition that it is the final form of the verb like, but I can not specify it in the conventional search by character string, so I will extract all the sentences containing the character string "○○ and ○○ can" , "to come to an update of the document can be specified as command," there is a problem that statements such as from being extracted.

The present invention solves these problems.
Prepare a sentence expression pattern that describes part-of-speech information and semantic information in association with the notation of characteristic morphemes of sentences that include important information, and match the result of morphological analysis of the input document with the sentence expression pattern. The purpose is to extract a desired sentence that includes important information with high accuracy, when it matches.

[0006]

FIG. 1 is a block diagram showing the principle of the present invention. In FIG. 1, an input document 1 is a document from which sentences are extracted.

The document information extraction device 2 is a morpheme analysis unit that performs morphological analysis on the input document 1 and extracts sentences that match the sentence expression pattern 5 based on the notation, part-of-speech information, and semantic information. 3, a sentence pattern matching unit 4, and a sentence expression pattern 5.

The morphological analysis unit 3 is for performing morphological analysis on the input document 1. The sentence pattern matching unit 4 matches the sentence expression pattern 5 on the basis of the notation, part-of-speech information, and semantic information of the sentence subjected to the morpheme analysis by the morpheme analysis unit 3, and extracts the matching sentence.

The sentence expression pattern 5 is a pattern in which a sentence pattern to be extracted (notation of a characteristic morpheme, part-of-speech information, and semantic information) is registered. Extracted sentence 6
Is a sentence extracted from the input document.

[0010]

In the present invention, as shown in FIG.
Performs morphological analysis of the input document 1, and the sentence pattern matching unit 4 matches with the sentence expression pattern 5 based on the morpheme notation and part-of-speech information of each sentence analyzed by the morpheme to extract matching sentences.

The morpheme analysis unit 3 performs morpheme analysis on the input document 1, and the sentence pattern matching unit 4 performs morpheme analysis on the basis of the morpheme notation and the semantic information of each sentence.
And the matching sentence is extracted.

The morphological analysis unit 3 morphologically analyzes the input document 1, and the sentence pattern matching unit 4 matches the sentence expression pattern 5 with the sentence expression pattern 5 on the basis of the morpheme notation, part-of-speech information, and semantic information of each sentence. It tries to extract the matching sentences.

Therefore, a sentence expression pattern in which the notation of characteristic morphemes of a sentence including important information, part-of-speech information, and semantic information is described is prepared, and the result of morphological analysis of the input document and the sentence expression pattern are prepared. By matching and extracting as a desired sentence when they match, it is possible to accurately extract a sentence containing important information.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the construction and operation of an embodiment of the present invention will be described in detail with reference to FIGS.

FIG. 2 shows a flowchart for explaining the operation of the present invention. This is a flow chart for explaining the overall operation of the configuration of FIG. In FIG. 2, S1 reads the input document 1 and performs morphological analysis. In this, the morpheme analysis unit 3 included in the document information extraction device 2 of FIG. 1 reads the input document 1 and performs morpheme analysis to generate a morpheme notation of a sentence, part-of-speech information, and semantic information. For example, FIG.
The manual sentence which is the input document 1 is read, and as shown in FIGS. 4 and 5, morpheme notation, part-of-speech information, and semantic information are generated. For example, the sentence class of sentence number “1” in FIG. 3 is an abstract object that defines data attributes and operations. Is morphologically analyzed to generate morpheme notation, part-of-speech information, and semantic information, as shown in FIG.

In S2, the sentence pattern matching unit 4 performs matching with the sentence expression pattern 5. This is the S1
The morpheme notation, part-of-speech information, and semantic information of the sentence subjected to morpheme analysis are compared with the sentence expression pattern of FIG. 6 to find a matching sentence.

In S3, a sentence that matches the sentence expression pattern 5 is extracted. This extracts the matched sentence by the collation of S2. Input document 1
Morphological analysis is performed to generate morpheme notation, part-of-speech information, and semantic information of the sentence, and these are collated with a sentence expression pattern 5 (for example, the sentence expression pattern example of FIG. 6) registered in advance to find a matching sentence. Extract. At this time, the sentence is extracted not only when the notation of the morpheme of the sentence is matched, but also when the part-of-speech information and the semantic information are also matched. It becomes possible to register and specify in the sentence expression pattern 5. By specifying part-of-speech information and semantic information in addition to the notation of the sentence to be extracted, even if the notation of the morpheme of the sentence matches, a sentence in which the part-of-speech information and the semantic information do not match (match) is not extracted. It became possible to prevent the extraction of sentences and selectively extract only important sentences. The details will be sequentially described below.

FIG. 3 shows an example of the manual sentence of the present invention. This is an example of the input document 1 in FIG. Sentence numbers 1 to 7 are described on the right side, and are related to the later explanation. Morphological analysis of sentence numbers 1, 5, 6, and 7 of the manual sentence to generate the morpheme notation, part-of-speech information, and semantic information in FIGS. 4 and 5.

4 and 5 show examples of morphological analysis according to the present invention. These are the sentence numbers 1 of the manual sentence in FIG.
Morphological analysis is performed on the sentences 5, 6, and 7 to generate the morpheme notation, the part-of-speech information, and the semantic information of the sentence.

FIG. 6 shows an example of a sentence expression pattern of the present invention. Here, as patterns for expressing important sentences, three sentence expression patterns 1, 2, and 3 are described in advance as follows.

-Sentence expression pattern 1: [noun] / and /////-Sentence expression pattern 2:-[verb end form] / and / can / suffer-Sentence expression pattern 3: [noun] <Means> / to / yo / tsu /
Here, in [], part-of-speech information is represented, in <>, semantic information is represented, and in other cases, morpheme notation is represented. ~ Represents an arbitrary character string. / Represents a morpheme delimiter of a sentence.

As described above, the sentence expression patterns 1, 2 specifying the morpheme notation of the sentence, the part-of-speech information, and the semantic information.
By describing 3, the sentence expression patterns 1, 2,
Only sentences matching 3 will be extracted from the input document 1.

FIG. 7 shows a flow chart of collation with the sentence expression pattern of the present invention. This describes morphological analysis of the input document 1 to generate morpheme notation, part-of-speech information, and semantic information of a sentence, and to compare these with the sentence expression pattern 5 of FIG.

In FIG. 7, S11 determines whether the morpheme list is complete. If YES, end (EN
D). On the other hand, in the case of NO, the list of morphemes has not ended, and therefore the process proceeds to S12. Here, the list of morphemes is the notation of morphemes obtained by morphological analysis of sentences, part-of-speech information, and semantic information. For example, the list of morphemes with sentence number 1 in the manual sentence in FIG. 3 is described as sentence number 1 in FIG. It is a list of notation, part-of-speech information, and semantic information.

In step S12, it is determined whether the sentence expression pattern ends. This determines whether the sentence expression pattern, for example, the sentence expression patterns 1, 2, and 3 in FIG. If yes,
Returning to S11, the morpheme list of the next sentence is repeated. On the other hand, if NO, the process proceeds to S13.

In step S13, it is determined whether the morpheme list matches the sentence expression pattern until the end. If yes,
Since the morpheme list and the sentence expression pattern match until the end, the matching sentence is extracted in S14, and the process returns to S11.
On the other hand, in the case of NO, there is no match, so S12
Return to and repeat.

As described above, when the list of sentence morphemes of the input document 1 and the sentence expression pattern are sequentially collated and matched, the matched sentence is extracted. The matching of the manual sentence of FIG. 3 with the sentence expression pattern of FIG. 6 will be described below.

(1) Sentence number 1 in FIG. 3 A class is an abstract object that defines data attributes and operations. The Upon morphological analysis, the morphological notation statement class / A /, / data / Roh / attribute / a / operation / a / provisions / to / abstract / Do / object / in / is. (Part-of-speech information and semantic information are document number 1 in FIG. 4)
See part-of-speech and semantic information). This notation of the morpheme of sentence number 1 and the pattern of part-of-speech information is the sentence expression pattern 1 of FIG. 6. [Noun] / and ///// The underlined part of the notation matches, that is, "class" and "noun (common noun)"-"what" and "to"-"de" and "de"-"ru" Since "ru" matches, the sentence of sentence number 1 is extracted as shown in FIG.

(2) The sentence of sentence number 5 is extracted as shown in FIG. 8 as the sentence expression pattern 2 in the same manner as (1). (3) Sentence No. 6 in FIG. 3 • The generated instance differs depending on the class type. When the morphological analysis, morphological notation of the statement, class / Bruno / kind / to / I / Tsu / Te /, / generation / /
Re / ru / instance / is / different / ru. (See the part-of-speech information and the meaning information of the document number 6 in FIG. 5 for the part-of-speech information and the meaning information). The morpheme notation of sentence number 6 and the pattern of part-of-speech information and sentence expression pattern 3 of FIG. 6 are obtained. When [noun] <means> / ni / yo / t / te / ...・ The "class", "common noun (part of speech information)", "concrete (semantic information)" and [noun] <means> do not match. That is, although the part of speech information of the class is “noun”, the semantic information of “concrete” and “means” do not match, and the sentence number 6 does not match the sentence expression pattern 3.

Therefore, even if the notations match, the meaning information does not match, and the meanings do not match as a whole, so they do not match and the sentence of sentence number 6 is not extracted. (4) On the other hand, the sentence of sentence number 7 ・ The slot of the instance can be specified by the editor. When the morphological analysis, morphological notation of the statement, editor / on / by / Tsu / Te /, / instance / Bruno /
Slot / specify / specify / specify. (See the part-of-speech information and the meaning information of the document number 7 in FIG. 5 for the part-of-speech information and the meaning information). The pattern of the morpheme of the sentence number 7, the part-of-speech information, and the semantic information is the sentence expression pattern 3 of FIG. 6. [Noun] <means> / ni / yo / t / te / ~ is collated. And the morpheme notation of the above sentence, part-of-speech information,
And the semantic information match, that is, "class", "common noun", "means" and "noun",
"Means"-"ni" and "ni"-"yo" and "yo"-"tsu" and "tsu"-"te" and "te" match, so the sentence of sentence number 7 is expressed as a sentence expression pattern. It is extracted as shown in FIG.

From the above, from the manual sentence of FIG.
As shown in FIG. 8, the sentences of sentence numbers 1, 5, and 7 can be extracted as the sentences that match the sentence expression patterns 1, 2, and 3 of FIG.

FIG. 8 shows an extracted example sentence of the present invention.
As described above, this is a sentence extracted from the manual sentence of FIG. 3 as matching with the sentence expression patterns 1, 2, and 3 of FIG. At the time of this extraction, the type of sentence is registered in association with the “definition sentence”, “method sentence”, and “method sentence” registered in the sentence expression patterns 1, 2, and 3 that match as shown in the figure. deep.

FIG. 9 shows an application example of the present invention. this is,
With respect to the extracted sentences in FIG. 8, an application example in which an index is automatically generated for each sentence type will be shown. Here, the sentence with the sentence number 1 of the definition sentence in FIG. 8 is taken out, and what is the class in the illustrated column of (index) (definition)?・ ・ ・ P ××
(Page number) and edit automatically. Similarly, how to take out the sentences of sentence numbers 5 and 7 of the method sentence of FIG. 8 and generate an arbitrary instance in the index (method) column? ... pxx
(Number of pages) How to specify the instance slot? ... pxx
(Page number) and edit automatically.

As described above, from the manual sentence which is the input document 1 in FIG. 3, the sentence characterized by the morpheme notation, the part-of-speech information and the semantic information by the sentence expression patterns 1, 2 and 3 in FIG. 6 (sentence number 1, It is possible to extract (5, 7) only and automatically edit (generate) the index from the extracted highly accurate sentences by further dividing them into sentence types.

[0035]

As described above, according to the present invention,
Prepare sentence expression patterns that describe notation of morphemes that have characteristics of sentences containing important information, part of speech information, and semantic information,
Since this sentence expression pattern and the result of morpheme analysis of the input document are collated and the matched sentence is extracted as a desired sentence, the desired sentence including important information is extracted with high accuracy. I was able to do it. Particularly, as a sentence expression pattern, a desired sentence including important information can be extracted with high accuracy by arbitrarily designating a sentence morpheme notation, part-of-speech information, and semantic information, and the notation (character string)
Since it is possible to prevent the extraction of unnecessary sentences that only match, it is possible to reduce the trouble of deleting unnecessary sentences.

[Brief description of drawings]

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

FIG. 2 is a flowchart explaining the operation of the present invention.

FIG. 3 is an example of a manual sentence of the present invention.

FIG. 4 is a morphological analysis example (continued) of the present invention.

FIG. 5 is a morphological analysis example (continuation) of the present invention.

FIG. 6 is an example of a sentence expression pattern of the present invention.

FIG. 7 is a flowchart for matching with a sentence expression pattern of the present invention.

FIG. 8 is an extracted sentence example of the present invention.

FIG. 9 is an application example of the present invention.

[Explanation of symbols]

 1: Input document 2: Document information extraction device 3: Morphological analysis unit 4: Sentence pattern matching unit 5: Sentence expression pattern 6: Extracted sentence

Claims (3)

[Claims] 1. A sentence expression pattern (5) in which part-of-speech information is registered as necessary in association with the morpheme notation of a sentence to be extracted, and a morphological analysis unit (3) for morphologically analyzing an input document.
And a sentence pattern matching unit (4) that matches the above sentence expression pattern (5) based on the morpheme notation and part-of-speech information of each morphologically analyzed sentence and extracts a matching sentence. A document information extracting device, which extracts a sentence of. 2. A sentence expression pattern (5) for registering semantic information as necessary in association with the notation of morphemes of a sentence to be extracted, and a morphological analyzer (3) for morphologically analyzing an input document.
And a sentence pattern matching unit (4) that matches the above sentence expression pattern (5) and extracts a matching sentence based on the morpheme notation and semantic information of each morphologically analyzed sentence. A document information extracting device, which extracts a sentence of. 3. A sentence expression pattern (5) for registering part-of-speech information and semantic information as necessary in association with the notation of morphemes of a sentence to be extracted, and a morphological analyzer (3) for morphologically analyzing an input document. )
And a sentence pattern matching unit (4) for matching the sentence expression pattern (5) and extracting a matching sentence based on the morpheme notation, part-of-speech information, and semantic information of each morphologically analyzed sentence, A document information extraction device characterized by extracting a desired sentence from inside.