JPH0934905A - Key sentence extraction method, abstract method and sentence search method - Google Patents

Abstract

(57) [Abstract] [PROBLEMS] In evaluating the importance of a sentence for extracting a key sentence in a document, a more appropriate importance based on the context is obtained by being based on the degree of association with other sentences in the document. To improve the accuracy of key sentence extraction. SOLUTION: A sentence cutout unit 2 evaluates the degree of association between sentences in a digitized document 1, a sentence importance degree evaluation unit 4 evaluates the degree of importance of a sentence, and a key sentence extraction unit 5 extracts a key from within the document 1. Sentences are extracted, and an abstract sentence is created from the extracted key sentences. The degree of relevance between sentences is based on the degree of overlap between keyword candidate words, which are mainly nouns in sentences, and the degree of importance of a sentence is based on the degree of relevance to other sentence groups and whether or not they are related. Ask.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a key sentence extraction method for extracting an important sentence (key sentence) from a document, an abstract method using the key sentence extraction method, and a sentence in the key sentence extraction method. The present invention relates to a sentence search method using an inter-relationship evaluation means.

[0002]

2. Description of the Related Art Document summarization is important for grasping the outline of a document, and an automatic summarizing device is expected. However, it is even more difficult to automate the creation of summaries that are difficult for humans. So, as an alternative to the summary,
It is more feasible to extract key sentences (important sentences) from documents and to automate the abstract created by connecting those important sentences. To extract key sentences, high-frequency words [1: Japanese Patent Laid-Open No. 61-117658 (text abstraction device)] and keyword importance [2: Japanese Patent Laid-Open No. 3782/1982]
70 (Abstract sentence creating device)], collation with key syntax [3: JP-A-61-100861 (document editing device)], logical structure of important words and original sentence [4: JP-A-2-
No. 181261 (automatic abstract generation device)], knowledge [5: Japanese Patent Laid-Open No. 4-74259 (document summarization device)] for recognizing important parts, and the like.

In the invention described in JP-A-61-117658 [1], a sentence is sequentially divided into paragraphs and the like, and a term is added to each sentence existing in each divided paragraph. Each sentence is divided and the importance of the sentence is discriminated based on the frequency of use of the term, and the abstract of the sentence is automatically edited by editing the abstract in such a way that the sentences of the highest importance are sequentially connected to each paragraph. It was edited.

Further, in the invention described in JP-A-3-278270 [2], the keyword extraction / evaluation means extracts a keyword candidate from the input document and the importance of the keyword candidate is set in advance. Calculated based on the conditions, and from the calculated importance of the keyword and the input document, the sentence evaluation means calculates the appropriateness for the content expression of the document for each sentence as a sentence evaluation value and the sentence evaluation value. A predetermined sentence is selected by comparing with a preset threshold value, and the selected sentence is arranged to create an abstract sentence.

Further, the invention described in the above-mentioned JP-A-61-100861 [3] analyzes syntactically or semantically each sentence in an inputted sentence, and analyzes the syntactically or semantically. The whole sentence structure or a part of the sentence structure is compared with the partial sentence structure (sentence key structure) registered in advance in the dictionary,
A sentence including a partial sentence structure registered in the dictionary is extracted from the input sentence, and the sentence is reconstructed using these extracted sentences. For example, a partial sentence registered in the dictionary The sentence element corresponding to the deletion rule in the structure is deleted from the sentence extracted from the input sentence to reconstruct the sentence.

Further, the invention described in Japanese Unexamined Patent Publication No. 2-181261 [4] uses a Japanese dictionary to completely remove functional words, and targets general nouns and proper nouns with their frequency information and Contents that the author thinks are important based on the information of the structure of the text, by extracting the important words that are the core of the text and the core of the description from the position information with high accuracy and analyzing the logical structure of the text of the original text. It is intended to include in the abstract a description of what you want to emphasize.

Further, according to the invention described in Japanese Patent Application Laid-Open No. 4-74259 [5], the sentences constituting the document to be summarized are each analyzed, and the analysis result and the knowledge storage means are stored. Knowledge for recognizing an important part of a document, for example, character modification information is used to recognize and determine an important sentence in the document, and according to the result of the determination, the important sentence is extracted from the document and summarized. It is designed to create sentences.

[0008]

However, according to the above-mentioned conventional method, information from the outside [the above-mentioned gazettes 2, 3, 4, 5] and
It requires syntactic analysis [the above-mentioned gazettes 3, 4, and 5] and independently evaluates each sentence, so that the relevance of sentences in a document is not emphasized. Incidentally, JP-A-6-25942
In Japanese Patent Laid-Open No. 4 (document display device, document summarizing device, and digital copying device), only a heading in a document is limited, but a sentence highly relevant to a heading in a document is a word in the heading. It is extracted by matching with and is a contextual selection.

The present invention has been made in view of the above-mentioned circumstances, and particularly, in the evaluation of the importance of a sentence for extracting a key sentence in a document, the degree of relevance to other sentences in the document is determined. Based on this, we give more appropriate importance based on context, improve the accuracy of key sentence extraction, and as an evaluation of the importance of the sentence, a simple method without a large burden such as external knowledge and parsing is used. It is intended to be applicable.

[0010]

According to the first aspect of the present invention, there is provided "sentence cutting means" for cutting out a sentence from an electronic document.
A "sentence relevance evaluation means" that evaluates the degree of relevance between one sentence in the document and another sentence, and a "sentence" that evaluates the degree of importance of the sentence based on the degree of relevance between other sentences in the document And a "key sentence extraction means" for extracting a key sentence based on the degree of importance of the sentence. According to the invention of claim 2, in the invention of claim 1, the "sentence degree between sentences" is included. The “evaluation means” extracts a keyword candidate word group from the sentence and evaluates the degree of association based on the similarity between the keyword candidate words included in each sentence. The invention of claim 3 provides the invention of claim 2 The degree of overlap of characters between the character strings of the keyword candidate words is used as the similarity between the keyword candidate words in the “inter-sentence degree evaluation unit”, and the invention of claim 4 is the invention of claims 1 to 3. In the above, "the sentence importance evaluation means" is Evaluating the severity of the sentence by the one or both of the associated coverage degree with relevance strength and other Bungun the group, the invention of claim 5, claim 4
In the invention of, as the strength of the degree of association with the other sentence group, the average value of the degree of association with other sentences, as the coverage degree of the association with the other sentence group, of the degree of association with other sentences The invention of claim 6 uses the average value of presence or absence, and in the invention of claim 4, as the importance of the sentence, the degree of association with other sentence groups and the degree of coverage of association with other sentence groups. The invention of claims 1 to 6 makes it possible to extract a key sentence based on the degree of importance of a sentence based on the degree of association between the sentences in the document. In particular, the method for evaluating the degree of association between sentences and the degree of importance of a sentence provided in claims 2 to 6 does not use external knowledge or syntactic analysis, but does not use external knowledge or syntactic analysis, This is an effective method that can be easily obtained by processing.

The invention of claim 7 is characterized in that the key sentence extraction method according to any one of claims 1 to 6 is used to extract key sentences having a high degree of importance in a sentence and arrange them in the order of sentences in a document. By this, the key sentence is extracted and the abstract is created in the document based on the degree of importance of the sentence based on the degree of association between the sentences.

The invention of claim 8 is based on the "sentence relevance evaluation means" in the key sentence extraction method according to any one of claims 1 to 3, and is based on the degree of relevance with the sentence or word group of the search request. The present invention is characterized in that a sentence in a document is searched for. By using the degree of relevance between sentences according to the inventions of claims 1 to 3, a specific sentence selected from within the document or from the outside can be used. This is also effective as a method for searching a sentence that is highly related to a given search sentence or keyword group.

[0013]

1 is a block diagram for explaining a key sentence extraction method, an abstract method, and a sentence search method according to the present invention. In FIG. 1, 1 is a digitized document, and 2 is a sentence segmentation means. Reference numeral 3 is an inter-sentence degree evaluation means, 4 is a sentence importance degree evaluation means, 5 is a key sentence extraction means, and a "sentence cutout means" 2 cuts out a sentence from the digitized document 1 one by one. Usually, the sentences in a document are separated by one sentence by a punctuation mark ".", Or one line forms one sentence without a punctuation like a heading. In general, the technique of cutting out sentences one by one from an electronic document file is a technique already realized by a machine translation system or the like, and a detailed description thereof will not be given here. FIG.
Shows the contents of the document used in the following explanation.
Shows the sentences cut out by the "sentence cutting means" (the sentence number is added to the beginning of each sentence. In the original sentence shown in FIG. 2, the first three sentences are headings). .

The inter-sentence relevance evaluation means 3 evaluates the degree of relevance between each clipped sentence and each of the other sentences, and obtains the relevance. Although various methods are conceivable for obtaining the degree of relevance between sentences, in the invention of claim 2, the degree of relevance is extracted based on the similarity between the keyword candidate words included in each sentence by extracting a group of keyword candidate words from each sentence. I am trying to evaluate.

As a keyword candidate word, a noun is generally mentioned. FIG. 4 shows the results of extracting nouns from each sentence, excluding numerical nouns such as dates and nouns with one kanji character that are difficult to be keywords. Extraction of nouns can rely on conventional morphological analysis techniques for Japanese sentences. Or you can just cut out kanji, katakana, or alphabetic strings,
It is possible to obtain almost the same result.

There may be various methods of obtaining the degree of association between sentences based on the similarity of keyword candidate words, but in the invention of claim 3, the degree of similarity between keyword candidate words is the overlap of characters between character strings of keyword candidate words. Use degrees. Here, the degree of association R (i, j) of sentence i with respect to sentence j is
Among the keyword candidate word characters of the sentence i, the total number of the character strings of the keyword candidate word of the sentence j and the character of the keyword candidate word of the sentence j is defined as a ratio of the total number of the keyword candidate word characters.

For example, the keyword candidate words of the fifth and sixth sentences are as follows: [5] Anglo-American major advanced G7 agreement regulated Iraq Target communist zone export control committee COCOM list [6] ] G7 target item Talks End of Cold War Export Control By substring matching in combination between keyword candidate words in both sentences, easily duplicate characters are "G7", "regulation",
It can be seen that they are “target” and “export”. Since the number of overlapping characters is 8, while the total number of characters of the keyword candidate words in each sentence is 34 and 18, respectively, the ratio is R (5,6): 8/34 for each sentence. = 0.235 R (6,5): 8/18 = 0.444.

FIG. 5 shows the degree of association of the fifth sentence with other sentences,
The degree of relevance between sentences is shown in a matrix in FIG. 6 (note that in the figure,
The degree of association is shown as an integer by multiplying the above value by 100 [percentage]. The above method is based on partial matching of word character strings, but if the ratio of the number of words that match the keyword candidate words is taken, in the above case, the three words “G7”, “restriction”, and “target” , While the number of words in each is 10 and 6, so: R (5,6): 3/10 = 0.30 R (6,5): 3/6 = 0.50

Further, returning to the invention of claim 2, as a more general degree of relevance, for example, the sentence i with respect to the sentence j is
It is conceivable that the degree of association R (i, j) of is the average of the sums of the similarities of the words in sentence i with each word in sentence j. This is because the set of keyword candidate words in sentence i is W (i), the similarity between word x and word y is r (x, y) [0 to 1], sentence i
If the number of keyword candidate words in the above is N (i), it can be expressed by the following equation (1).

[0020]

[Equation 1]

When the degree of similarity between words is strictly considered, a method of using a thesaurus system in which upper and lower relationships between words are given can be considered.

The sentence importance level evaluation means 4 evaluates the importance level of each sentence based on the degree of association of each sentence with other sentences. In the invention of claim 4, the degree of importance of the sentence is evaluated by the degree of the degree of association with other sentence groups and the coverage of the association with other sentence groups. The former shows how strongly they are related to other sentences, and the latter shows how widely they are related to other sentences. As a specific calculation method, in the invention of claim 6, the strength of the degree of association with another sentence group, the average value of the degree of association with another sentence, and the coverage degree of the association with another sentence group are ,
The average value of the degree of association with other sentences is used. Table 5 shows the fifth sentence.

[0023]

[Table 1]

FIG. 7 shows the strength of the degree of relevance of each sentence based on the degree of relevance of FIG. 6, the coverage, and the value of the product thereof (multiplied by 100 and integerized. The product is '=>'). (Shown on the right). Focusing on the strength of relevance, the second and third sentences are.
It is as high as 33, 27, and the fourth, sixth, seventh and tenth sentences are .15 ~.
It forms the second group of 17 ranges. Focusing on the related coverage, the fifth and seventh sentences are high at .91, and the fourth and sixth sentences continue at .73.

It is assumed that a sentence group having a high degree of relevance is a sentence that is strongly related to the theme of the document, as the headline is high. In addition, the sentence having a high related coverage includes the sentence of the first paragraph which is said to have a summary content in newspaper articles and the like, and it is assumed that there is a high possibility that it includes the entire content. Conversely, a sentence with this low value
It is considered to be a sentence that describes a very specific topic. As described above, all the values are significant as the importance of the sentence for extracting the key sentence.

Further, in the invention of claim 6, the product of both of them gives the importance of the sentence in consideration of both effects. Based on this importance, for .10 and above, the second sentence,
The 7th sentence, the 6th sentence, the 4th sentence, the 10th sentence and the 5th sentence are in this order, and the 3rd sentence of the heading 2nd sentence and the 1st paragraph (4th, 5th, 6th
Sentence), the first sentence (seventh sentence) and the final sentence (second sentence) of the second paragraph
Sentence) is extracted as a key sentence.

According to the invention of claim 7, the extracted key sentences are shown in order to form an abstract. Based on the above-mentioned importance, the abstract is shown in FIG. 8 (upper 2 sentences: second sentence, seventh sentence).
Sentence) and FIG. 9 (upper 6 sentences: second sentence, fourth sentence, fifth sentence, sixth sentence)
Sentence, the 7th sentence, the 10th sentence) are obtained.

According to the eighth aspect of the present invention, a sentence related to the sentence or keyword group given as the retrieval request is retrieved by the degree of relevance of the sentence of the present invention. For example, because the headline is considered to contain important keywords,
If the first sentence of the heading sentence, “Industrial products related to conventional weapons”, is used as the search sentence, the fourth sentence, “Export restrictions on industrial products that can be diverted to conventional weapon parts and processing machines, is ...” is searched. According to the second sentence of the headline sentence (FIG. 10), the three sentences of the first paragraph and the first sentence of the last paragraph are searched as the sentences having a high degree of association (FIG. 11).

[0029]

According to the first aspect of the present invention, the "sentence cut-out means" for cutting out a sentence from an electronic document and the "sentence relevance between sentences" for evaluating the degree of relevance between one sentence and another sentence in the document. Evaluation means "
And a "sentence importance evaluation means" that evaluates the importance of the sentence based on the degree of association with other sentence groups in the document, and a "key sentence extraction means" that extracts the key sentence based on the importance of the sentence. 3. The method according to claim 2, wherein
The invention of claim 1 is the invention of claim 1, wherein the “sentence relevance evaluation means” extracts a keyword candidate word group from a sentence,
The degree of relevance is evaluated based on the degree of similarity between the keyword candidate words included in each sentence. The invention of claim 3 is the invention of claim 2, wherein the keyword candidate words in the "inter-sentence relevance evaluation means" are The degree of overlap of characters between the character strings of the keyword candidate words is used as the degree of similarity. In the invention of claim 4, in the inventions of claims 1 to 3, the “sentence importance evaluation means” is another sentence. The importance of a sentence is evaluated based on one or both of the degree of association with a group and the degree of coverage of association with another sentence group. The invention of claim 5 is the invention of claim 4, wherein Use the average value of the degree of association with other sentences as the degree of association with the sentence group, and the average value of the degree of association with other sentences as the degree of coverage of the association with the other sentence group. The invention of claim 6 is the same as the invention of claim 4, Is used as the degree of importance of the relation between the degree of relevance to another sentence group and the degree of coverage of the relation to another sentence group. The invention according to any one of claims 1 to 6 is characterized in that Enables the extraction of key sentences based on the degree of importance of the sentence based on the degree of relation between the sentences in the document,
In particular, the method of evaluating the degree of association between sentences and the degree of importance of a sentence provided in claims 2 to 6 does not use external knowledge or syntactic analysis, but performs analysis processing of noun determination degree and character string matching processing. It is an effective method that can be easily obtained by.

The invention of claim 7 uses the key sentence extraction method according to any one of claims 1 to 6 to extract key sentences with high importance in sentences and arrange them in the order of sentences in a document. In this way, a key sentence is extracted and an abstract is created in a document based on the degree of importance of the sentence based on the degree of association between the sentences.

The invention according to claim 8 is based on the "sentence relevance evaluation means" in the key sentence extraction method according to any one of claims 1 to 3, and is based on the relevance to the sentence or word group of the search request. The present invention is characterized in that a sentence in a document is searched for. By using the degree of relevance between sentences according to the inventions of claims 1 to 3, a specific sentence selected from within the document or from the outside can be used. This is also effective as a method for searching a sentence that is highly related to a given search sentence or keyword group.

[Brief description of drawings]

FIG. 1 is a configuration diagram for explaining a key sentence extraction method, an abstract method, and a sentence search method according to the present invention.

FIG. 2 is a diagram showing an example (original text) of a digitized document for explaining an embodiment of the present invention.

FIG. 3 is a diagram showing a result of sentence segmentation of the original sentence shown in FIG.

FIG. 4 is a diagram showing keyword candidate words for each sentence.

FIG. 5 is a diagram showing a degree of association of a fifth sentence with another sentence.

FIG. 6 is a diagram showing a relevance matrix between sentences.

FIG. 7 is a diagram showing strength of association, coverage of association, and a product thereof.

FIG. 8 is a diagram showing an example of an abstract (abstract 1).

FIG. 9 is a diagram showing another example of abstract (abstract 2).

FIG. 10 is a diagram showing a search result of a related sentence based on a heading first sentence.

FIG. 11 is a diagram showing a search result of a related sentence based on a heading second sentence.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Digitized document, 2 ... Sentence extraction means, 3 ... Inter-sentence degree evaluation means, 4 ... Sentence importance evaluation means, 5 ... Key sentence extraction means.

Claims (8)

[Claims] 1. A sentence segmentation unit for segmenting a sentence from an electronic document, an inter-sentence degree evaluation unit for assessing the degree of association between one sentence in the document and another sentence, and another sentence in the document. A key sentence characterized by having a sentence importance degree evaluation means for evaluating the importance degree of a sentence based on the degree of association with a group and a key sentence extraction means for extracting a key sentence based on the importance degree of the sentence. Extraction method. 2. The inter-sentence relevance evaluation means extracts a keyword candidate word group from a sentence and evaluates the relevance based on the similarity between the keyword candidate words included in each sentence. The key sentence extraction method according to claim 1. 3. The key sentence extraction according to claim 2, wherein the degree of overlapping of characters between the character strings of the keyword candidate words is used as the similarity between the keyword candidate words in the inter-sentence degree evaluating unit. method. 4. The sentence importance level evaluation means evaluates the importance level of a sentence based on one or both of the strength of the degree of association with another sentence group and the coverage degree of the association with another sentence group. The key sentence extraction method according to any one of claims 1 to 3. 5. As the strength of the degree of association with the other sentence group,
5. The key according to claim 4, wherein an average value of the degree of association with other sentences is used as the average value of the degree of association with other sentences and the coverage degree of the association with the other sentence group. Sentence extraction method. 6. The key according to claim 4, wherein a product of a degree of association with another sentence group and a coverage degree of association with another sentence group is used as the importance of the sentence. Sentence extraction method. 7. An abstract method, wherein the key sentence extraction method according to any one of claims 1 to 6 is used to extract a key sentence having a high degree of importance in a sentence and arrange the sentences in a document in order. 8. A sentence in a document is searched based on the degree of relevance to a sentence or a word group in a search request, based on the inter-statement relevance evaluation means in the key sentence extraction method according to any one of claims 1 to 3. A sentence search method characterized by: