JPH08278982A - Method for retrieving similar word or similar sentense - Google Patents

Abstract

PURPOSE: To lighten the load of retrieval and decision processing by making it easy to select a necessary document out of retrieval results and eliminating the need to classify terms. CONSTITUTION: An input document α as a routine document is decomposed into plural constituent terms, which are set as reference terms γ. Words similar to standard words matching the reference terms γ are stored in a similar word storage area βtogether with similarities by referring to a term dictionary (f). The similar words in this area are rearranged to obtain words similar to the reference terms γ, and those similar words are set as document retrieval reference terms. As for a document data base (g) where similar documents are stored and a retrieval area is set, a document having terms similar to the terms for document retrieval is stored as a similar document in a similar document storage area ε. The similar documents in this area are rearranged to find the similarities between the input document and similar documents and store them in the similar document storage area ε, and a similar document corresponding to the input document α is outputted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a similar word search method for searching a word dictionary for a similar word of a constituent term of an input sentence, and a similar case of an input sentence in the past based on the similar word. The present invention relates to a method of searching for a similar sentence that is searched from a file in which the sentence etc.

[0002]

2. Description of the Related Art Conventionally, in order to retrieve a sentence similar to an inquired sentence from among sentences registered in a computer, a keyword is registered in advance and a sentence having this keyword is searched. A general method has been to judge that they are similar to each other, or to classify the sentences of various cases in advance according to some classification and perform a search.

[0003]

Among these, in the method based on keyword registration, all sentences having keywords are searched and presented, and when there are many cases, the sentences that are searched and presented are Too many and it is often difficult to find the text you actually need. Further, in the search by classification, it is difficult for a searcher to find a sentence to be searched unless the search criteria is known.

Further, when creating a system proposal or a product instruction manual, a sentence with a high degree of similarity created in the past should be searched at the time of creating the document, and if this sentence can be diverted, it is expected that the number of production steps will be reduced. You can However, a search method from the viewpoint of searching and diverting sentences has not yet been realized. In addition, there is a method that considers the number of synonyms and the like in the similarity determination of sentences, but there is no method that pays attention to the degree of similarity in terms and the entire sentence structure.

The present invention has been made to solve the above problems, and an object of the present invention is to facilitate selection of a required sentence from a large number of similar sentences even if a large number of similar sentences are searched for. To provide a method of searching for similar words or similar sentences that does not require classification of terms and performs similarity determination of sentences based on similarity determination of terms, and that does not impose an excessive load on search and determination processing. is there.

[0006]

First, the terms used in the present invention are defined as follows. A. Synonym and similarity, and degree of similarity Synonym means complete agreement of meaning content, and similarity means partial agreement. Therefore, if the degree of coincidence of meaning contents is represented by a numerical value and called as similarity, the similarity is 1.0, for example, and 1.0 for similarity.
It is a decimal value less than.

(Specific example)-"Process" and "line" or "plan" and "schedule" are synonymous words- "Planned work" and "planning" or "plan adjustment" are similar words- "Machining process""Productionplan" and "machining line production schedule" are synonymous sentences ・ "Machining process production planning work" and "machining factory production plan adjustment" are similar sentences In the present invention, such (including synonyms ) Similar words and similar sentences are searched by performing similarity calculation processing.

B. Term Dictionary A synonym is collected in a daily dictionary, but if the contents are examined for synonyms or similarities, similar words are also collected. From this point of view, assuming a term dictionary as in the specific example of FIG. 2, a similar word search is performed based on this dictionary. In addition, among the similar words shown in FIG. 2, it is registered as a reference term and the similar relationship is matched,
Some are registered only as similar words and are not standard terms. It is generally very difficult to create a standard term dictionary as described above and use it in common. Therefore, depending on the field and target function, it can be shared within the scope of joint users of sentences. It is enough to change.

C. Fixed sentence and non-fixed sentence The sentence to be searched in the present invention is considered to be a collection of terms. Ordinary sentences are called non-standard sentences because they use terms with arbitrary configurations, while sentences described according to a certain format are called fixed sentences.
In the case of a fixed sentence, the terms that compose the sentence are classified according to the format, so that the comparison and determination can be efficiently performed, and the result is clear. The present invention is particularly useful as a method for retrieving similarities in this fixed text.

(Specific example) Sentence 1 (fixed form sentence): "Frequently change the line configuration based on production instructions" This sentence is an IP that expresses the functions of software and machines
O (Widely known as computer software technology; I is an input section that the corresponding sentence means, P is a processing section,
O is the output part) format, I and O; line configuration P; (frequently) changing auxiliary part; disassembled as follows based on production instructions. Such a term is called a constituent term of sentence 1, and a process corresponding to this format is executed.

Sentence 2 (non-standard text): "Line structure is frequently changed." When such a format is also standardized, it can be processed as a standardized text, but it conforms to the IPO format. Since it is difficult to break it down into constituent terms, we will consider sentence 2 as an atypical sentence.

The fixed text to which the present invention is applied is an IP.
Although it is not necessary to limit to the O format, the present invention is mainly intended to search texts for information processing, and the IPO format is considered to be the most universal in actual information processing. It is a classification of constituent terms in accordance with.

D. Weight of similarity calculation for fixed text Since the fixed text can take into account the importance of its composition, this importance is used as a weight when calculating the similarity. (Specific example) In the IPO format, the portion corresponding to P is important, and I and O are often cases of P. In that case, a large weight is given to the P portion and a small weight is given to the I portion and the O portion. Can be judged to be very similar as a whole. On the other hand, when the I part and the O part have a high dependency due to a special function, the corresponding weights are increased to reduce the similarity if the I part and the O part are different, so that they are not so similar. You can judge.

The present invention is realized by the following means. a. Text input means This is a well-known means for inputting a text (reference text) to be searched for a similar text. When searching for a sentence that is similar to a fixed phrase, enter the format as well. b. Term Extraction Means in Sentence This is a well-known means for extracting terms in an input sentence, and a group of terms extracted by this means is referred to as a term (reference term) at the time of retrieval.

C. Similar word search means A term similar to the input reference term is searched from the term dictionary, and the similar word and the degree of similarity are stored in the storage area and displayed. d. Similar sentence retrieving means Decomposes an input sentence into constituent terms, retrieves a sentence in a sentence database that uses similar terms of these constituent terms, stores this as a similar sentence in a storage area, and displays it.

E. Sentence similarity calculating means, which is prepared as needed in the similar word searching means, calculates the similarity between the input sentence and the similar sentence based on the similarity of terms constituting each sentence. At this time,
A weighting process is executed when a similar sentence of a fixed sentence is searched. f. Term dictionary This is a table in which the table shown in FIG. 2 is stored.

G. Sentence database This is a database that stores the sentences to be searched. The non-standard text stores the character string itself. In the fixed text, a term decomposed for each structure, a character string that is not a term, and a weight are stored.

H. Retrieval instruction / result display means For instructing a process for retrieving a similar word or a similar sentence and displaying a retrieval result, the same one as the sentence input means can be used. i. Result printing means Used not only for displaying but also for printing as needed.

J. Similar sentence retrieval device Consistently executes the entire processing of the present invention. k. Text component weight table This is a table that stores the weights set for the components of a fixed text. The weight may be input for each process, but may be commonly determined in relation to the fixed text.

The gist of the present invention realized by each of the above means is as follows. A similar word search method according to the first invention of claim 1 decomposes an input sentence which is a fixed sentence into a plurality of constituent terms, sets these constituent terms as reference terms, and refers to a term dictionary. If there is a canonical term that matches the reference term, the similar word is stored in the similar word storage area together with the similarity, and if there is no canonical term that matches the reference term, the canonical term that the similar word matches is stored as a similarity of the similar word. It is characterized in that the similar words in the similar word storage area are stored together with the degree and the similar words in the similar word storage area are arranged to obtain the similar words of the reference term.

According to a second aspect of the present invention, there is provided a method of retrieving similar words according to the first aspect, wherein the term dictionary has general contents including explanation of standard terms and parts of speech. , Similarities corresponding to the reference terms and the similarities expressed by numerical values are stored.

A similar sentence search method according to a third aspect of the present invention is to execute the search method according to the first or second aspect to search for similar words corresponding to a plurality of constituent terms of an input sentence. , Setting these similar words as text search reference terms, and targeting a text database in which similar texts are stored and a search area is set, a text having a term similar to the text search reference term as a similar text It is characterized in that it is stored in the storage area, the similar texts in the similar text storage area are organized, and then the degree of similarity between the input text and the similar text is obtained and stored in the similar text storage area.

According to a fourth aspect of the present invention, there is provided a similar sentence search method according to the third aspect, wherein a similar word corresponding to a constituent term of the input sentence is included in the constituent terms of the similar sentence. In some cases, the similarity of the constituent terms of the similar sentence is set to the similarity of the constituent terms of the input sentence, and the similarity between the input sentence and the similar sentence is obtained based on the similarity of the constituent terms. To do.

According to a fifth aspect of the present invention, there is provided a method for retrieving similar sentences according to the third aspect, wherein the similar words corresponding to the constituent terms of the input sentence are included in the constituent terms of the similar sentence. In the case where there is no correspondence, the non-corresponding composition similarity is set to the similarity of the constituent terms of the input sentence, and the similarity between the input sentence and the similar sentence is obtained based on the similarity of the constituent terms.

A similar sentence search method according to a sixth aspect of the present invention is the search method according to the fourth or fifth aspect, wherein the similarity between the input sentence and the similar sentence is determined based on the similarity of the constituent terms. When obtaining, the similarity between the input sentence and the similar sentence is obtained based on the maximum value, the minimum value, or the product of the similarities of the plurality of constituent terms of the similar sentence.

According to a seventh aspect of the present invention, there is provided a method for retrieving a similar word, wherein in the retrieval method according to the first or second aspect, the similarity among the retrieved similar words is equal to or higher than the standard similarity. It is characterized by processing only as a similar word.

According to an eighth aspect of the present invention, there is provided a method for retrieving a similar sentence according to any one of the third to sixth aspects, wherein the similarity among the retrieved similar words is a criterion. It is characterized in that only terms having a degree of similarity or higher are processed as similar terms.

According to a ninth aspect of the present invention, there is provided a method for retrieving similar sentences, wherein in the retrieval method according to any one of the fourth to sixth aspects, the significance of the constituent terms of the similar sentence which is a fixed sentence. The degree of similarity of the constituent terms is weighted according to, and the degree of similarity between the input sentence and the similar sentence is obtained based on the weighted degree of similarity.

According to a tenth aspect of the present invention, a similar sentence search method decomposes an input sentence which is a fixed sentence into a plurality of constituent terms, sets these constituent terms as reference terms, and sets a term dictionary. A similar word of the reference term that matches the reference term is stored together with the degree of similarity in the similar word storage area, and the similar words in the similar word storage area are organized to obtain similar words of the reference term. A word is set as a sentence search reference term, and a sentence having a term similar to the sentence search reference term is stored as a similar sentence in the similar sentence storage area for a sentence database in which similar sentences are stored and a search area is set. Then, after arranging the similar sentences in this similar sentence storage area, the similarity between the input sentence and the similar sentence is calculated, stored in the similar sentence storage area, and the similar sentence corresponding to the input sentence is output. And it features.

In the similar sentence search method according to the eleventh aspect of the present invention, the reference term is used when it is determined that the similar sentence output by the search method according to the tenth aspect is insufficient for the search. A reference term matching a similar word is searched from the term dictionary and stored in the similar word storage area together with the similarity of the similar word, and the similar words in the similar word storage area are sorted to obtain the similar word of the reference term. , Setting these similar words as text search reference terms, and targeting a text database in which similar texts are stored and a search area is set, a text having a term similar to the text search reference term as a similar text The similar sentence corresponding to the input sentence is stored in the storage area, the similar sentences in the similar sentence storage area are sorted, the similarity between the input sentence and the similar sentence is calculated, and the similarity is stored in the similar sentence storage area. And outputs.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows the overall structure of a similar sentence search device j for implementing the present invention, together with a work area and peripheral input / output means. First, the terms to be processed and the situation of sentences will be described based on this figure.

The main cases are as follows: Sentence 1; “Frequently change the line configuration based on production instructions” is assumed, and this sentence is used as the input sentence of a key to search for a similar sentence. This text is input by the text input means a and stored in the work area α. The fixed text is based on the IPO format as described above. In this case, the constituent elements of the sentence are “input part; I”, “auxiliary part”,
The sentence 1 in the work area α is decomposed and stored in the work area β as a constituent term. The extraction of the constituent terms is executed by the term extracting means b.

Input section; line configuration auxiliary section; output section based on production instructions; line configuration processing section; change frequently

A sentence database g is used for sentences similar to this.
The following sentence is stored in the sentence database g. -Sentence 2; "Line configuration is frequently changed" -Sentence 3; "Line configuration is often changed" -Sentence 4; "Processing machine allocation is determined based on production instructions" -Sentence 5; “Optimize production planning to reduce waiting time as parts arrive”

Of the above-mentioned sentences, only sentence 2 is stored as a non-standard sentence, while sentences 3 to 5 are stored as standard sentences and are decomposed into constituent terms as shown in Table 1. It has been done.

[0035]

[Table 1]

The term dictionary f shown in FIG. 3 is used as an example of the term dictionary f necessary for specifically explaining the contents of the present embodiment. The similar relations of the three terms “change”, “create”, and “determine” in FIG. 3 have multiple stages, and when such a dictionary is created,
This means that the similarity is interpreted as shown in FIG.

That is, the similarity between "create" and "determine" is 0.9, and the similarity between "change" and "create" is 1 or less (0.8). Therefore,
Similarity between "change" and "determine" (0.7
2) uses the term as being smaller than the similarity (0.8) between “change” and “create”. From the opposite viewpoint, “create” is more similar to “change” than “determine” rather than the similarity (0.72) between “decide” and “change”. , The similarity (0.8) between "change" is larger.

Here, the similarity (0.72) between “change” and “determine” is the similarity (0.8) between “change” and “create”. It is calculated by the product of the similarity (0.9) between "create" and "determine", and the term dictionary in FIG. 3 is created by the product type. Other than this, a maximum value type term dictionary that takes the maximum value of the similarity or a minimum value type term dictionary that takes the minimum value may be used.

Next, FIG. 5 is a flowchart showing a process of searching for a similar word of a reference term by using the term dictionary f and storing it, and corresponds to the embodiment of the invention described in claim 1.
The term dictionary f shown in FIG. 3 is used in the embodiment of the invention described in claim 2.

The series of processing shown in FIG. 5 is executed by the similar word searching means c shown in FIG. As a preparation for searching a similar sentence, it is only necessary to search for a similar word.
“Change” is set in the work area γ as a reference term in (101). Further, since it is difficult to make the term dictionary f common, a term that suits the purpose is designated (102).

When the reference term and the term dictionary f are determined in this way, the reference term of the term dictionary f is searched for. In the term dictionary f shown in FIG. 3, since “change” is registered as the reference term, a similar word area (similar word −
As long as there are 1, similar words-2, ...) (105), the storage in the similar word storage area δ is continued (104). As a result,
Similar words of “change”, “create”, “determine”, and “perform” are stored in the storage area δ together with the respective degrees of similarity (104). Since only the same reference term is registered in the term dictionary f, this process is executed only once.

Even if the term does not match the reference term in 103, the similar word "change" is searched for from the word registered in the term dictionary f (106). The standard terms “create” and “determine” (these two are registered in duplicate), in which “change” is registered as a similar word, are stored in the storage area δ as similar words together with their respective similarities. (107). Although not shown in FIG. 3, for example, also for the reference term “optimize”, if “change” is registered in the similar word area, the similar word is stored in the storage area δ together with the similarity. Is stored.

When the reference terms change and the dictionary area ends (108), the duplicated "create" and "determine" stored are deleted and only one is stored. (109). This completes the search for similar words related to the reference term “change”. The search instruction and the search result are displayed by the search instruction / result display means h in FIG. 1, and are printed out by the result printing means i as necessary.

FIG. 6 is a flow chart showing a process of searching for similar sentences from the sentence database g of FIG. 1 and storing them, which corresponds to the embodiment of the invention described in claim 3. This processing is executed by the similar sentence search means d in FIG. Hereinafter, this processing procedure will be described assuming that the input sentence is the sentence 1; “The line configuration is frequently changed based on the production instruction”.

Since the constituent terms of sentence 1 are "based on production instruction", "line configuration", and "change frequently", these three are set as reference terms (201). Then, the above-mentioned similar word search means c executes similar word search processing until all of these three reference terms are completed (203), and stores the searched similar words in the similar word storage area δ (202). At the same time when the similar words are stored, the input sentence, that is, the constituent terms of the sentence 1 are also stored in the work area β.

In this process, when "change frequently" is set as the reference term, it is necessary that the term dictionary f contains a reference term including a modifier such as "frequently". In that case, the processing is quick and the reliability is improved. On the other hand, however, the effort for creating such a term dictionary becomes large.

A technique for decomposing "changing frequently" into "frequently" and "changing" and deciphering that "changing" is the main part has already been established. Therefore,
In the present invention, since it is considered that “production instruction”, “line configuration”, and “change” are the same as the configuration terms, as shown in FIG. 3 for “change”. The similar words are stored in the similar word storage area δ. Furthermore,
Similar words for “production instruction” and “line configuration” are also searched for in the term dictionary f, and all similar words stored in the storage area δ are organized in the same manner as 109 in FIG. 5 (204).

Of course, "frequently" may have an important meaning, so if such processing is to be realized,
It is sufficient to include the modifiers in the term dictionary f, to give the degree of similarity, and process the fixed forms in which the modifiers are taken into consideration for each of the constituent terms. Since the text database g is diverse, a region to be searched is set (205) to limit the processing.
In this example, it is assumed that the area of the sentence database g in which the sentences 3 and 4 are stored is set.

The similar words stored in the similar word storage area δ are
Set as reference terms for sentence search in order (20
6), a similar sentence including the reference term is searched for in the entire region of the sentence database set in 205 (208) and stored in the similar sentence storage area ε (207). For example, sentence 3
"Line structure" matches sentence 1 and sentence 4 "decides"
Are similar to "change" in sentence 1, so these are stored in the similar sentence storage area ε (207).

Further, similar to 204, the stored similar sentences are sorted in consideration of the processing efficiency (209). In this way, instead of organizing by 204 or 209, 202
Although it is more advantageous to save the storage area when storing items 207 and 207, this difference is not important, and thus the same treatment as 109 in FIG. 5 is performed. In this way, since the sentences 3 and 4 similar to the sentence 1 of the input sentence are stored in the similar sentence storage area ε, the similarity between the sentences 1 and 3 and the similarity between the sentences 1 and 4 are similar. Is calculated according to the procedure shown in FIG.
These are stored in the similar sentence storage area ε (21
0).

FIG. 7 is a flowchart of a processing procedure for calculating the similarity between the input sentence and the retrieved similar sentence,
This corresponds to the embodiment of the invention described in claim 4. This process
It is executed by the sentence similarity calculation unit e in the similar sentence search unit d.

The constituent terms of the input sentence for which similar sentences are to be searched are set as the input sentence constituent terms, the constituent terms of the searched similar sentence are set as the similar sentence constituent terms, and the similarity is determined based on the input sentence constituent terms. The similarity between the input sentence and the similar sentence is obtained based on the set similarity of the terms. Table 2 is a summary of the specific examples described above in association with the procedure here.

[0053]

[Table 2]

The processing procedure of FIG. 7 will be described below by taking the case where the input sentence, the term dictionary, and the sentence database are as shown in Table 2 as a specific example. First, when the input sentence is indefinite (301), it is meaningless to decompose it into the constituent terms as shown in Table 2. Therefore, the conventional method of calculating the similarity between sentences using the number of synonyms is used ( 311). In this case, the sentence 2 described above can be determined to be similar to the input sentence (sentence 1), but it cannot be associated with the degree of similarity of terms.

When the input sentence is a fixed form (301), the process target element in the input sentence constituent term is set as a constituent element (306) until there is no target (302), and the similar sentence constituent term is processed. It is determined whether there is a similar word corresponding to the target element (303). When there is a similar word, the similarity between the input unit and the output unit for sentence 3 is 1.0, and the similarity between the output unit and the processing unit for sentence 4 is 0.85 and 0.72, respectively. And set (30
5).

When this processing is executed for a component having no similar word, it is substantially the case that the input part "based on production instruction" of sentence 4 is not similar to the component of sentence 1; the input part. Has no meaning. That is, in Sentence 1, the "line configuration" is used as an input unit and an output unit in consideration of the "change" processing target, but in Sentence 4 it is assumed that the same thing is not a constituent element, and the auxiliary unit of Sentence 1 I just assigned the constituent terms in the input section. If you do this, sentence 5
It is considered that the example of sentence 4 is normal because the positioning of another auxiliary part becomes clearer, such as.

From the above consideration, in the decomposition of the constituent terms such as the sentence 1, it is judged that the input part does not actually have to be described as a constituent element. Then, it is possible that the input part of sentence 1 is sufficiently similar even if there is no term corresponding to the input part. Therefore, in such a sentence, the input part does not actually need to be strictly defined. . In fact, although the original sentence 1 is missing, it was only analyzed strictly and added, so it is decided to give a similar word even if there is no similar word corresponding to this term. “Structure similarity”.

This non-corresponding structural similarity is set to the similarity of the constituent elements of the input sentence and is used for calculating the similarity between the input sentence and the similar sentence, which corresponds to the embodiment of the present invention. . What value to set as the non-corresponding configuration similarity may be determined depending on the text. Considering the constituent elements of sentence 4 simply, it is natural to set the non-corresponding structural similarity to the auxiliary unit. Furthermore, if the constituent terms of the auxiliary section are similar to those of the input section, the output section, and the processing section, it can be determined that the entire sentence is similar.

How to determine the similarity of each sentence from the similarity of the constituent terms depends on the usage of the similarity.
However, since it is considered that the processing is performed by using a computer, only the maximum value type, the minimum value type, or the product type is discriminated as the similarity calculation method (307). These are described in claim 6. It corresponds to an embodiment of the invention.

In 308, when the similarity is calculated with the maximum value (the maximum value among the similarities of the respective constituent elements is taken as the similarity of the entire sentence), the similarity of sentence 3 becomes 1.0 and the sentence 4 becomes
The similarity (assuming that the non-corresponding configuration similarity is not set) is 0.85. Making the maximum value of the similarity of the constituent elements the similarity of the entire sentence means placing importance on the most similar constituent elements, and when even one of the constituent elements is the same or the similarity is high. Is equivalent to determining that the entire sentence has a high degree of agreement or similarity.

In 309, when the similarity is calculated with the minimum value (the minimum value among the similarities of the respective constituents is taken as the similarity of the entire sentence), the sentence 3 and the sentence 4 (the non-corresponding configuration similarity is set). However, since the similarity becomes 0.0, the above-described non-corresponding configuration similarity is effective for preventing this. Making the minimum value of the similarity of the constituent elements the similarity of the entire sentence means placing importance on the constituent elements that are most different, and when even one of the constituent elements is different or the similarity is low. Is equivalent to determining that the entire sentence is also different or has low similarity.

As described above, the maximum value type evaluates high similarity, resulting in a large number of similar sentences, and the minimum value type evaluates low similarity, resulting in a small number of similar sentences. Since both of these have the drawback that the similar number and the dissimilar number are difficult to be reflected in the constituent terms, in order to reflect a finer difference in the similarity, the product type (similarity of the similarity of each component is The product is taken as the similarity of the whole sentence).

When the product of the similarities is calculated in 310, the similarity of both sentence 3 and sentence 4 (assuming that the non-corresponding composition similarity is not set) becomes 0.0, but the non-corresponding composition similarity. If the degree is set and is not 1.0 or 0.0, the degree of similarity changes depending on the number. Then, if the sentence 4 has a similar sentence of “changing the allocation of the processing machine based on the production instruction”, a higher degree of similarity can be given.

In the present invention in which a sentence is decomposed into constituent terms and the similarity between terms and the similarity between sentences correspond to each other, it is significant to finely determine the similarity between sentences. Therefore,
This product-type similarity calculation plays an important role, and is very effective in connection with the weighting described later.

The product-type similarity calculation formula is as follows. Sentence similarity = similarity of input section × similarity of auxiliary section × similarity of output section × similarity of processing section Now, assuming that a sentence is divided into n constituent elements, the above similarity calculation formula is generally used. When it is described in, the equation 1 is obtained.

[0066]

[Equation 1]

The calculation formula such as the formula 1 is used when the sentences match when all of the constituent terms match, and if the matching of the sentences is considered to be completely similar, the constituent elements It is based on the fact that it corresponds to the logical product of the match discrimination. Therefore, the logical product (AN
I tried to think about it in D), which results in the calculation of the product.

FIG. 8 shows a procedure of a process of discarding similar candidates, and corresponds to the embodiment of the invention described in claims 7 and 8. Specifically, similar word candidates (similarity candidates) having a degree of similarity equal to or higher than the criterion similarity degree given in advance during similarity search.
Since the purpose is to reduce the processing load by narrowing down to 104, 107 in FIG.
2), used when storing similar words by 207 and 210 in FIG. This processing is executed by the similar word search means c and the similar sentence search means d. That is, if the degree of similarity searched or calculated in each of the steps 104, 107, 202, 207, and 210 is equal to or greater than the preset reference degree of similarity (401), the similar candidate is processed as a similar word (402). , If it is smaller than the standard similarity (4
01), it is processed as not a similar word (403).

FIG. 9 shows the procedure of the weighting process for considering the importance of the sentence constituents. This processing corresponds to the embodiment of the invention described in claim 9, and is executed by the sentence similarity calculation means e. Specifically, the importance is considered by multiplying the similarity of the retrieved similar word by a preset weight stored for each component in the sentence component weight table k shown in FIG. To do.

That is, when the similarity is calculated by 308, 309, and 310 in FIG. 7, a determination like 501 in FIG. 9 is made, and if the weight is not considered, the similarity of the search result is used as it is ( 503), if the weight is taken into consideration, the degree of similarity is recalculated according to Equation 2 (50
2). Note that x in Expression 2 is obtained by Expression 3 from "weight-i" stored in the weight table k.

[0071]

[Equation 2]

[0072]

(Equation 3)

The weighted similarity calculated in this way has a numerical value different from the similarity set for the term, but since it has the following characteristics, it is effective as a weight for the similarity. The similarity of matching (1.0) is unchanged. A component with a small weight approaches a match (close to 1.0). The greater the weight, the closer to the term similarity. Originally, since the numerical value of the degree of similarity itself is relative, it is possible to use such weights if it is applied uniformly as a whole.

In general, the processing section (P) is important in the text of the IPO format that describes the functions of software and automated machines, so the purpose is to allow similar words to be searched for. Therefore, if the weight of the processing unit is set to be larger than the others according to the above-mentioned formula 2, this object can be achieved.

That is, if the weight of the processing unit is increased and the weights of other portions are decreased, the similarity of the processing unit becomes closer to the similarity of the term, and the weights of other elements are reduced by the larger the processing unit. Therefore, since the weighted similarity is close to 1.0, it is not reflected in the similarity of sentences. Therefore, if a weight of 0.0 is set for an insignificant component, processing can be performed regardless of the degree of similarity between terms.

The above weights can be defined without using Equation 3. For example, the present invention is characterized by Formula 1 and Formula 2 because there are various possibilities such as setting x itself as a weight for each component. Note that an example of the relationship between the original similarity of the constituent elements and the weighted similarity according to Expression 2 is shown in FIG. 11.

FIG. 10 shows the consistent processing of the similar sentence search by the similar sentence search device j of FIG.
0 corresponds to an embodiment of the invention. Since the individual processing has already been described, an example of the processing as a whole will be outlined here using the specific example.

First, in 601 sentence 1 is input. At the same time, whether or not to process according to the present invention is distinguished depending on whether the format of the input sentence is fixed or irregular. At 602, sentence 1 is decomposed into constituent terms as previously described. In 603, if the standard similarity for truncation processing is input in order to speed up the processing, in the processing of 605 and 609, the processing with a similarity smaller than that value is stopped as a similarity candidate. Generally, since a special term is used or a special use of the term is made for each field, the term dictionary used in 604 is designated.

After 605, the process is similar to that of FIG.
As actual processing, similar words are displayed using the search instruction / result display means h in 606, weights are input as necessary in 607, a text database is specified in 608, and then similar text search and storage processing is performed. Is executed (609), the search result is output, and printed / displayed as necessary (61).
0). If the search is satisfactory, the process ends (611).

Although the sentence 5 cannot be retrieved by the standard processing as described above, "optimizing" the "production plan" and "changing" the "line configuration" are essentially extremely difficult. It is unknown whether the information used as the input unit or the auxiliary unit at that time is the same or similar or different between the sentences 1 and 5. Nevertheless, sentence 5 was not a similar candidate because "optimize" was not searched directly from the term dictionary.

In order to improve this, when there are few sentences in the retrieval result or when the expected sentence is not retrieved, it is determined in 611 that the retrieval is unsatisfactory, and the processing in 612 is performed to continue the retrieval. Run. That is, the contents of the storage area for the constituent terms are replaced with the constituent terms of the input sentence with similar words of the constituent terms, and the processing from 605 is repeated.

This reprocessing corresponds to the embodiment of the invention described in claim 11. By this reprocessing, the constituent term of the processing part of sentence 1 is replaced from "change" to "determine",
Since “determine” is registered as a similar word of the reference term “optimize” in the specific example of the term dictionary shown in FIG. 3,
“Optimize” is searched, and sentence 5 is searched and stored.

[0083]

As described above, the present invention finally searches for similar sentences by associating the degree of similarity between terms and the degree of similarity between sentences. For example, in the case of processing a sentence explaining the function of software. Extremely useful for In other words, in order to enable matching of software functions and determination of similarities by the matching and similarity of sentences, it is necessary to clarify the partial matching and difference of the constituent terms of the sentences. It is possible to realize partial coincidence and difference discrimination. In this respect, according to the present invention, it is possible to search for similar software and a matching portion and a different portion of the function thereof.

Further, according to the present invention, by truncating or weighting similar candidates as necessary,
It is possible to narrow down the similar search target and improve the processing speed. Furthermore, since classification of terms is not required, a searcher can easily perform an effective and accurate search without knowing the classification criteria.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a similar sentence search device to which an embodiment of the present invention is applied.

FIG. 2 is a diagram showing a specific example of a term dictionary.

FIG. 3 is a diagram showing a specific example of a term dictionary.

FIG. 4 is a diagram illustrating the degree of similarity between similar words.

FIG. 5 is a flowchart showing a processing procedure for searching for and storing a similar word using a term dictionary.

FIG. 6 is a flowchart showing a processing procedure for searching for similar sentences in a sentence database and storing the same.

FIG. 7 is a flowchart showing a processing procedure for calculating the similarity between an input sentence and a retrieved similar sentence.

FIG. 8 is a flowchart showing a processing procedure for discarding similar candidates.

FIG. 9 is a flowchart showing a processing procedure for weighting a text component.

FIG. 10 is a flowchart showing a consistent processing procedure for similar sentence search.

FIG. 11 is a diagram showing an example of a relationship between original similarity of components and weighted similarity.

[Explanation of symbols]

 α, β, γ, δ, ε Work area or storage area a Text input means b Term extraction means c Similar word search means d Similar text search means e Text similarity calculation means f Term dictionary g Text database h Search instruction / result display Means i result printing means j similar sentence search device k sentence constituent element weight table

 ─────────────────────────────────────────────────── ─── Continued Front Page (72) Inventor Takashi Ikematsu 1-1, Tanabe Nitta, Kawasaki-ku, Kawasaki-shi, Kanagawa Fuji Electric Co., Ltd.

Claims (11)

[Claims] 1. An input sentence, which is a fixed sentence, is decomposed into a plurality of constituent terms, these constituent terms are set as reference terms, and if there is a reference term that matches the reference term by referring to a term dictionary, it is similar. A word is stored in the similar word storage area together with the similarity, and if there is no reference term that matches the reference term, the reference term in which the similar word matches is stored in the similar word storage area together with the similarity of the similar word, and the similar word storage A similar word search method characterized by organizing similar words in a region to obtain similar words to a reference term. 2. The search method according to claim 1, wherein in the term dictionary, in addition to general contents including explanations of standard terms and parts of speech, similar terms and numerical values corresponding to the standard terms are used. A method of searching for a similar word, characterized in that the similarity is stored. 3. The search method according to claim 1 or 2 is executed to search for similar words corresponding to a plurality of constituent terms of an input sentence, and these similar words are set as sentence search reference terms. Storing sentences having a term similar to the sentence search reference term in a sentence database in which sentences are stored and a search area is set is stored as a similar sentence in the similar sentence storage area, and the similar sentences in the similar sentence storage area are organized. A similar sentence search method, characterized in that the similarity between the input sentence and the similar sentence is then obtained and stored in the similar sentence storage area. 4. The search method according to claim 3, wherein when a similar word corresponding to a constituent term of the input sentence is included in the constituent terms of the similar sentence, the similarity of the constituent terms of the similar sentence is input. A method for retrieving similar sentences, characterized in that the similarity between the input sentence and the similar sentence is obtained based on the similarity of the constituent terms of the above. 5. The retrieval method according to claim 3, wherein when the similar word corresponding to the constituent term of the input sentence is not included in the constituent terms of the similar sentence, the uncorresponding constituent similarity is set to the constituent term of the input sentence. A similar sentence search method characterized in that the similarity is set and the similarity between the input sentence and the similar sentence is obtained based on the similarity of the constituent terms. 6. The search method according to claim 4 or 5, wherein, when obtaining the similarity between the input sentence and the similar sentence based on the similarity of the constituent terms, the maximum among the similarities of the plurality of constituent terms of the similar sentence. A similar sentence search method, wherein the similarity between an input sentence and a similar sentence is obtained based on a value, a minimum value, or a product of similarities. 7. The search method according to claim 1 or 2, wherein, of the searched similar words, only the words whose similarity is equal to or higher than the standard similarity are processed as the similar words. Method. 8. The search method according to claim 3, wherein, of the searched similar words, only terms whose similarity is equal to or higher than the standard similarity are processed as similar words. How to search for similar sentences. 9. The search method according to claim 4, wherein the degree of similarity of the constituent terms is weighted according to the importance of the constituent terms of the similar sentence that is a fixed sentence, and the weighting is performed. A similar sentence search method, characterized in that a similarity between an input sentence and a similar sentence is obtained based on the calculated similarity. 10. An input sentence which is a fixed sentence is decomposed into a plurality of constituent terms and these constituent terms are set as reference terms,
By referring to the term dictionary, the similar terms of the reference terms that match the reference term are stored in the similar term storage area together with the degree of similarity, and the similar terms in the similar term storage area are organized to obtain the similar terms of the reference term, These similar words are set as text search reference terms, and a text database in which similar texts are stored and a search area is set, and a text having a term similar to the text search reference term is stored as a similar text as a similar text. Storing in a region, rearranging the similar sentences in this similar sentence storage region, obtaining the similarity between the input sentence and the similar sentence, storing it in the similar sentence storage region, and outputting the similar sentence corresponding to the input sentence. A search method for similar sentences characterized by. 11. When it is determined that the similar sentence output by the search method according to claim 10 is insufficient for the search, a reference term having a similar word to the reference term is searched from the term dictionary, and the reference term is searched. The similar words in the similar word storage area are stored together with the similarities of the similar words, and the similar words in the similar word storage area are organized to obtain similar words of the reference term. A sentence having a term similar to the sentence search reference term is stored as a similar sentence in a similar sentence storage area for a sentence database in which a sentence is stored and a search area is set, and the similar sentence in this similar sentence storage area is stored. A method for searching for similar sentences, characterized in that the similarities between the input sentence and similar sentences are calculated, stored in a similar sentence storage area, and the similar sentence corresponding to the input sentence is output.