JPH0765030A - Text search method and apparatus - Google Patents

Abstract

(57) [Summary] [Purpose] Considering how words are used in one sentence, the degree of similarity with the input sentence that is the search target is determined for the entire sentence, and the search is performed, enabling a wide range and highly accurate search. To do so. [Structure] A character string input from the input unit 1 is processed and displayed on the display unit 3. One sentence from the input unit 1 is divided into word units by the word division unit 4, and a phrase synthesis unit 5 creates a phrase from the divided words. The similarity between the phrase and the phrase is calculated by the phrase similarity calculation unit 6, and the similarity of the entire sentence is determined by the similar sentence determination unit 7. By this determination, it is determined that the entire sentence is similar. Only the target sentence is output to the display unit 3.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sentence retrieval method and a device for retrieving a similar sentence from a large amount of stored documents, which is used in a Japanese word processor, an optical character reader (OCR) and the like.

[0002]

2. Description of the Related Art Recently, Japanese word processors, optical character readers, etc. are widely used for document storage. In these devices, documents are stored using a storage medium such as a magnetic disk or a magneto-optical disk instead of a conventional recording paper or microfilm. In this case, a text search technique for searching a desired document from a large amount of stored documents is adopted. In this search, a match between the input search sentence and the character string is determined, or a match between the used words is determined, and the matching sentence is searched.

[0003]

However, in the sentence retrieval in the conventional example as described above, the sentence including the character string in the entire sentence is retrieved. Alternatively, after cutting the word into words, a sentence matching this word is searched. In this case, in the search of the character string, even if one character of this character string is different, it cannot be searched. Furthermore, in the word search, the sentence is searched without considering the usage of the word. Therefore, it may not be possible to search for a necessary sentence or may search for a useless sentence, and there is a drawback that a wide range and high precision cannot be searched.

The present invention solves the above-mentioned drawbacks of the prior art by taking into account the usage of words in one sentence, and by considering the degree of similarity between the entire sentence and the input sentence to be searched. It is an object of the present invention to provide a text search method and a device thereof that enable a judged search and can perform a wide range and highly accurate search.

[0005]

In order to achieve the above object, the sentence search method of the present invention divides a search input sentence and a search target sentence into word units, and then combines them into phrase units to obtain a search input sentence. The search target sentence is compared for each phrase, and the similarity between the phrases is calculated, and only the search target sentence determined to be similar as a whole sentence by this calculation is output.

In this case, the similarity between the bunsetsus is the independent word similarity calculated by calculating the similarity between the independent words in the bunsetsu,
The degree of similarity between adjunct words in a clause is combined with the calculated adjunct word similarity or either.

Further, the sentence retrieval device of the present invention comprises an input means for inputting a character string, a display means for processing the character string and displaying it, and a word dividing means for dividing one sentence from the input means into word units. , A bunsetsu synthesizing unit that synthesizes bunsetsu from words divided by the word slicing unit, a bunsetsu similarity calculating unit that calculates the similarity between bunsetsu obtained by the bunsetsu synthesizing unit, and a similarity of the entire sentence. And a similar sentence judging means for judging.

[0008]

As described above, in the sentence search method and apparatus according to the present invention, the search input sentence divided into word units and collected in phrase units is compared with the sentence to be searched for each phrase, and the similarity between the phrases is determined. Only the search target sentence is output when the whole sentence is similar after the calculation. Therefore, as compared with the conventional case of searching only by matching words, a highly accurate search can be performed without searching for sentences that are used differently in one sentence. In addition, since the independent word similarity and the adjunct word similarity are individually evaluated, it is possible to perform a search while judging the degree of similarity of the entire sentence, and a wide range of advanced searches are performed.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, an embodiment of a text search method and apparatus according to the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of a text search method and apparatus according to the present invention. In FIG. 1, this text search method and apparatus are provided with an input unit 1 for inputting character strings and control instructions.
And a control unit 2 for controlling each unit of the apparatus and an input unit 1.
It has a display unit 3 for displaying a character string and the like input from, and a word dividing unit 4 for analyzing the input character string and dividing it into word units. Furthermore, the words divided by the word dividing unit 4 are put together, and a bunsetsu synthesizing unit 5 for synthesizing them into bunsetsu, a bunsetsu similarity calculating unit 6 for comparing the bunsetsu with other bunsetsu, and calculating their similarity, A similar sentence determination unit 7 that determines the degree of similarity of the entire sentence from the calculation result of the phrase similarity calculation unit 6;
The search target sentence storage unit 8 stores and stores the search target sentence.

Next, the operation of the configuration of this embodiment will be described. A character string to be a search input sentence is input from the input unit 1. This search input sentence is
It is sent to the word division unit 4 and is divided here in word units.

FIG. 2 is a diagram showing a state in which the word dividing unit 4 analyzes a character string and divides it into word units. In FIG. 2, in the word division unit 4, the character string "Taro's father is a doctor" is changed to "Taro (noun)", "no (particle)", "father (noun)", "ha (particle)". , "Doctor (noun)", "is de (auxiliary verb end form)", grammatical terms such as part-of-speech and inflected forms are added to the divided words. The words thus divided are combined by the phrase combining unit 5 in units of phrases.
As a general rule, this bunsetsu consists of an independent word followed by an annex.

FIG. 3 is a diagram showing a case where a phrase is synthesized from the words analyzed in FIG. In FIG. 3, the words shown in FIG. 2 are synthesized as the clauses of “Taro's”, “father's”, and “is a doctor”.

Next, one sentence of the retrieval target sentence is fetched from the retrieval target sentence storage unit 8, processed in the same manner as the retrieval input sentence, divided into words, and the synthesizing of clauses is sequentially performed. The phrase similarity calculator 6 compares one phrase in the search input sentence with one phrase in the search target sentence to calculate the similarity. This calculation is “(number of phrases in the search input sentence) x (number of phrases in the search target sentence)”
Will be brute force. Based on this calculation result, the similar sentence judgment unit 7 evaluates the similarity of the whole sentence. If they are similar in this evaluation, this search target sentence is output to the display unit 3. When there is an unsearched search target sentence in the search target sentence storage unit 8, the next one stored sentence is taken out and the same processing is repeated.

Next, the processing procedure from the phrase similarity calculation unit 6 to the similar sentence determination unit 7 will be described. FIG. 4 is a flowchart showing the processing procedure of the operation from the phrase similarity calculation unit 6 to the similar sentence determination unit 7. Here, as a search input sentence,
And “Taro's father is a doctor” shown in FIG. 3 is entered,
It is assumed that word division and phrase synthesis are performed. Further, a case where "Jiro's father is a doctor" is retrieved as the search target sentence will be described.

FIG. 5 is a diagram showing a state in which the sentence "Jiro's father is a doctor" is analyzed by the word dividing unit 4 and divided into word units. In FIG. 5, in the word division unit 4,
The character string "Jiro's father is a doctor" is "Jiro (noun)", "no (particle)", "father (noun)", "ha (particle)", "doctor (noun)", " As shown in “(Auxiliary verb end form)”, grammatical terms such as parts of speech and inflectional forms are added to the divided words. The words thus divided are combined by the phrase combining unit 5 in units of phrases. In principle, a bunsetsu consists of an independent word followed by an adjunct.

FIG. 6 is a diagram showing a state in which a phrase is synthesized from the words analyzed in FIG. In FIG. 6, the words in FIG. 5 are synthesized into the clauses of “Jiro no”, “father is”, and “is a doctor”.

The phrase similarity calculation unit 6 sequentially extracts the phrases of the search input sentence from the beginning. At this time, if there is an unprocessed phrase (step (S) 10 in FIG. 4), the phrases of the search target sentence are sequentially taken out from the beginning and the unprocessed phrase is examined (step 11). If there is no unprocessed phrase in this check, the process proceeds to the process of the next phrase in the search input sentence (step 12), and the phrase process at the beginning of the search target sentence is started (step 13).

If there is any unprocessed phrase, the similarity of the independent word is calculated (step 14). For example, an evaluation score is set to 10 points when the independent words are completely the same and 7 points when the synonyms are similar. Since the independent words “Taro” and “Jiro” do not match between the first phrase “Taroha” of the search input sentence and the first phrase “Jiroha” of the search target sentence, the score is 0. This calculation is performed by a brute force of (all phrases of search input sentence) × (all phrases of search target sentence).

FIG. 7 is a diagram showing the calculation result of the similarity of the independent word. In FIG. 7, since the “father” between the second bunsetsu and the “doctor” between the fourth bunsetsu of the search input sentence and the search target sentence match each other as independent words, 10
Points are assigned. Then, the similarity of the attached words is calculated (step 15). Also in the case of the calculation of the degree of similarity of the adjunct words, if they are completely in agreement, the evaluation point is set to 10 points,
If they are similar to each other, an evaluation score is set to 7 points and a brute force calculation is performed.

FIG. 8 is a diagram showing the result of calculating the degree of similarity of attached words. In FIG. 8, "no (particle)", "wa (particle)", and "is (auxiliary verb end form)" in the first input phrase, the second output phrase, and the third output phrase in the search input sentence and the search target sentence. Are coincident with each other, and 10 points are assigned to each. Then, the phrase processing of the search target sentence is advanced to the next phrase (step 16). Further, when this processing is repeated and there is no unprocessed phrase in the search input sentence (step 10), the similar sentence judgment unit 7 calculates the similarity of the entire sentence (step 17). For the evaluation of the entire sentence, the scores of the independent word similarity and the adjunct word similarity shown in FIGS. 7 and 8 are added and evaluated. From the result of this evaluation, the combination of the phrase of the search input sentence and the phrase of the search target sentence is in one-to-one correspondence, and the combination with the maximum score is obtained.

FIG. 9 is a diagram in which the points of the independent word similarity and the adjunct word similarity are added. In FIG. 9, in the search input sentence and the search target sentence, 10 points of “Taro no” and “Jiro no” of the first phrase, and 20 points of “Father wa” and “Father wa” of the second phrase A total of 50 points, which is the sum of the points and 20 points of “being a doctor” and “being a doctor” between the third clauses, are obtained. Based on these 50 evaluation points, it is determined that the sentences are similar to each other (step 18), and the search target sentence "Jiro's father is a doctor" is output to the display unit 3 (step 19). If the evaluation score obtained here is 0, the search target sentence "Jiro's father is a doctor" is displayed on the display unit 3
It ends without displaying it without outputting to.

In this embodiment, the bunsetsu is composed of independent words and the following adjuncts, but the independent words and adjuncts can be freely constructed without departing from the scope of the present invention. Also,
When the similar sentence judgment unit 7 judges the similarity of the whole sentence, it does not simply add the independent word similarity and the adjunct word similarity as in this embodiment, but judges only by the independent word similarity, Also,
The determination may be made based only on the adjunct word similarity, or the similarity of the bunsetsu evaluated by both the independent word similarity and the adjunct word similarity may be changed.

[0023]

As is clear from the above description, the sentence search method and apparatus according to the present invention compare the search input sentence divided into word units and the sentence to be searched for each phrase. However, since the similarity between the clauses is calculated and only the search target sentence is output when the entire sentence is similar, compared with the conventional case of searching only by matching the words,
This has the effect that a highly accurate search can be performed without searching for sentences that differ in how they are used within one sentence. Furthermore, since the independent word similarity and the adjunct word similarity are individually evaluated, it is possible to perform a search while judging the degree of similarity of the entire sentence, and it is possible to perform a wide range of advanced searches.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a text search method and apparatus according to the present invention.

FIG. 2 is a diagram showing a state in which a character string is divided into word units by a word dividing unit in FIG.

FIG. 3 is a diagram showing clauses synthesized from the analyzed words in the embodiment.

FIG. 4 is a flowchart showing a processing procedure of an operation from a phrase similarity calculation unit to a similar sentence determination unit in the embodiment.

FIG. 5 is a diagram showing a state in which a character string is analyzed by a word dividing unit and divided into word units in the embodiment.

FIG. 6 is a diagram showing clauses synthesized from the analyzed words in the embodiment.

FIG. 7 is a diagram showing a calculation result of the similarity of independent words in the embodiment.

FIG. 8 is a diagram showing calculation results of similarity of attached words in the embodiment.

FIG. 9 is a diagram in which points of independent word similarity and adjunct word similarity are added in the embodiment.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Input part 2 ... Control part 3 ... Display part 4 ... Word division part 5 ... Phrase synthesis part 6 ... Phrase similarity degree calculation part 7 ... Similar sentence judgment part 8 ... Search target sentence storage part

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location 9194-5L 15/403 350 C

Claims (3)

[Claims] 1. A search input sentence and a search target sentence are divided into word units, then grouped into phrase units, the search input sentence and the search target sentence are compared for each phrase, and the similarities between the clauses A sentence search method characterized by calculating a degree and outputting only a search target sentence determined to be similar as a whole sentence by this calculation. 2. The degree of similarity between bunsetsu is a combination of an independent word similarity calculated by calculating the degree of similarity between independent words within a bunsetsu and an adjunct word similarity calculated by calculating the degree of similarity between auxiliary words within a bunsetsu, or The text search method according to claim 1, wherein the text search method is one of the two. 3. Input means for inputting a character string, display means for processing and displaying the character string, word dividing means for dividing one sentence from the input means into word units, and division by the word dividing means. A bunsetsu synthesizing means for synthesizing and creating bunsetsu from the selected words, a bunsetsu similarity calculating means for calculating the similarity between the bunsetsu obtained by the bunsetsu synthesizing means, and a similar sentence judgment for judging the similarity of the whole sentence. A text search device comprising means.