JPH08129554A - Relation expression extracting device and retrieval device for relation expression - Google Patents

Abstract

PURPOSE: To provide an extracting device which can extracts a more appropriate key word without depending upon surface expression by utilizing relation between words and a retrieval device which can perform a retrieve at a more appropriage retrieval request. CONSTITUTION: An expression extraction part 2 analyzes respective texts in document data inputted from a data input part 1 by a morpheme analysis part 14 to divide them into words, and then extracts plural mutually related words as expressions by a word group extraction part 15 on the basis of a word group extraction rule together with information on parts of speech. The expressions extracted by the expression extraction part 2 are related by a relation extraction part 3 on the basis of a relation extraction rule and further normalized by a normalization part 4. The related expressions normalized by the normalization part 4 are registered in a storage part 5 together with the inputted original data. The registered original data can be retrieved by using the related expression as a retrieval key.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a search device for performing a search by a keyword, and more particularly, a relational expression extraction device for extracting a relational expression that can be a keyword from a document, and a relational expression to be input. The present invention relates to a relational expression search device that performs a search as a key.

[0002]

2. Description of the Related Art Conventionally, as a method for searching desired information from a large amount of information stored in a database or the like, a keyword is assigned to each data in advance, and a keyword matching a search key input by a user is used. Devices have been developed to search for and output the information that they have.

In the keyword search, it is general that an expert called an indexer gives an appropriate keyword to the text information in advance. However, since the addition of this keyword requires a huge amount of work, many studies have been conducted on a technique for automatically extracting the keyword. For example, in the keyword importance automatic evaluation device described in JP-A-1-112331, a noun is extracted as a keyword from a document, and evaluation of statistical, syntactic, and semantic importance is added. .

However, in such a conventional keyword extraction method, since the extraction is generally performed on a word-by-word basis, even if the degree of importance is added, there are many things that are irrelevant to the original search result. . That is, there is a problem in that the precision rate decreases. This is because the concept corresponding to the search request desired by the user does not always match the word-level expression. Therefore, when searching with a word-level keyword, a text used in a meaning different from the meaning assumed by the user is also searched.

On the other hand, in the keyword extracting device described in Japanese Patent Publication No. 58-33993, there is proposed a method of limiting the meaning of a keyword by using a compound word. As a result, it becomes possible to extract words having a dependency relationship and the like, and there is no restriction that the concept is expressed in word units.

However, in this method, since a compound word or an expression similar to this is used as a fixed keyword, it is impossible to handle the difference in meaning between "search a document" and "search a document", for example. ,as a result,
Only expressions such as "document search" are extracted. In other words, by using the compound word, the information such as the dependency relationship that is analyzed in detail is lost instead of being independent of the surface expression. For this reason, the search results often include unsuitable ones.

As described above, in the conventional technique, it was still difficult to efficiently search for a keyword.

[0008]

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned circumstances, and has been conventionally ignored by extracting a plurality of expressions such as words from a document and the relationship between the expressions. Retrieval that utilizes a relationship between words and provides an extraction device that can extract a more appropriate keyword without depending on a surface expression and that can perform a search based on a more appropriate search request The purpose is to provide a device.

[0009]

According to a first aspect of the present invention, in a relational expression extracting device for extracting a relational expression that can be a keyword from original data, a plurality of relational expressions that can have a relationship with each other from the original data. The present invention is characterized by having an expression extracting means for extracting the expression (1) and a relationship extracting means for giving a relationship to the expression group extracted by the expression extracting means.

According to a second aspect of the present invention, in the relational expression extracting device according to the first aspect, the normalization is performed to normalize into a representation that does not depend on a surface expression based on the relation given by the relation extracting means. It is characterized in that it further comprises a converting means.

According to a third aspect of the present invention, in the relational expression extracting device according to the first or second aspect, the expression extracting means is based on an analyzing means for analyzing the original data and an analysis result by the analyzing means. It is characterized by having a word group extraction means for extracting a plurality of words that may have a relationship with each other.

According to a fourth aspect of the present invention, in the relational expression extracting device according to any one of the first to third aspects, the relational expression given from the relation extracting means or the normalizing means depends on a language. It is characterized by further comprising a language normalization means for normalizing into a non-expression.

According to a fifth aspect of the present invention, in the relational expression extracting device according to any one of the first to third aspects, the expression extracting means is adapted to correspond to a language in which the original data is described. The relationship extracting means comprises language-specific analyzing means for analyzing the original data, and the relationship extracting means comprises language-specific relationship extracting means for extracting a relational expression consisting of a plurality of words which may have a relationship with each other depending on the language and a relationship between the words. A bilingual word dictionary in which words having an equivalent meaning in a language are registered, a bilingual relationship dictionary in which information indicating an equivalent linguistic relationship in a plurality of languages is registered, and a language using the bilingual word dictionary and the bilingual relationship dictionary Accordingly, it further comprises a language normalization means for normalizing the relational expression extracted by the relational extraction means into a language-independent expression.

In a sixth aspect of the present invention, in a search device that performs a search using a relational expression as a search key, the relational expressions extracted as a plurality of expressions that may have a relationship with each other from the original data are registered as keys with the original data. The present invention is characterized in that it has a storage means that is stored and a search means that searches the original data registered in the storage section using the input relational expression as a search key.

In the invention according to claim 7, in the relational expression search device according to claim 6, the search means expands at least the input search request into a relational expression that can be conceptually the same. It is characterized in that it includes a relationship expanding means for performing a search.

According to an eighth aspect of the invention, in the relational expression retrieving apparatus according to the seventh aspect, the expression extracting means for extracting a plurality of expressions that may have a relationship with each other from the input data, and the expression extracting means extracts the expressions. And a normalization means for normalizing the relationship extracted by the relationship extraction means to an expression that does not depend on the surface expression based on the relationship extracted by the relationship extraction means. The relational expression extracted and normalized as a plurality of expressions that can have a relationship with each other by the word group extraction means, the relation extraction means, and the normalization means using the data as input data is registered as a key together with the original data. The expanding means uses natural language as input data, and a plurality of expressions that can have a relationship with each other by the word group extracting means, the relationship extracting means, and the normalizing means. It is characterized in that to perform the search by expanding the extracted normalized search request conceptually identical to become obtain relational representation Te.

According to a ninth aspect of the present invention, in the relational expression retrieval device according to the sixth or seventh aspect, the storage means extracts a plurality of expressions that may have a relationship with each other from the original data and converts them into languages. It is characterized in that a relational expression normalized to an independent expression is registered as a key together with the original data.

According to a tenth aspect of the present invention, in the relational expression search device according to the sixth or seventh aspect, a plurality of expressions that can have a relationship with each other from the input data according to the language in which the input data is described. An expression extraction means for extracting the expression, a relationship extraction means for extracting a relationship between the plurality of expressions extracted by the expression extraction means according to a language, and a bilingual word dictionary in which words having equivalent meanings in the plurality of languages are registered, A bilingual relation dictionary in which information indicating equivalent linguistic relations in a plurality of languages is registered, and a relational expression extracted by the relation extracting means according to the language according to the language using the bilingual word dictionary and the bilingual relation dictionary. Has a language normalization means for normalizing the expression into a language-independent expression, and the storage means uses the original data as input data for the expression extraction means and the relation extraction. A relational expression normalized to a language-independent expression by the language normalizing means is registered as a key together with the original data, and the searching means uses natural language as input data, the word group extracting means, and the relation extracting Means for performing a search using a search request normalized as an expression independent of language by the normalizing means.

According to an eleventh aspect of the present invention, in the relational expression retrieving apparatus according to any one of the sixth to tenth aspects, the retrieving means includes the relational expression based on the input retrieval request and the storage means. It is characterized by further comprising certainty factor evaluation means for comparing the relational expressions stored in, to evaluate the certainty factor, and rearranging the search results based on the certainty factor.

[0020]

According to the present invention, in the invention described in claim 1, the expression extracting means extracts from the original data a plurality of expressions which may have a relationship with each other, and gives the relationship to the expression group extracted by the relationship extracting means. As a result, it becomes possible to extract not only one word but a plurality of expressions and the relation between them as a key. As a result, concepts and subjects that do not depend on the surface expression are extracted, and more appropriate keywords can be extracted. Further, as in the invention according to claim 2, the structure having the normalizing means is adopted, and the relation between the expressions extracted by the relation extracting means is normalized by the normalizing means, so that the expression does not depend on the surface expression. Expressions can be keywords. As the processing for extracting the expression in the expression extracting means, for example, as in the invention described in claim 3,
The analysis unit may be configured to analyze the original data and the word group extraction unit may extract a plurality of words that may be related to each other.

According to the fourth aspect of the invention, a language normalization unit is provided, and the relational expression given from the relation extraction unit or the normalization unit is normalized to a language-independent expression, thereby making it dependent on the language. The keyword can be extracted by the expression that does not exist. When the original data is described in a plurality of languages, it is sufficient to analyze each language and normalize it into a language-independent expression. According to the invention described in claim 5, in order to realize this, the expression extraction means is constituted by a language-specific analysis means for analyzing the original data according to the language in which the original data is described, and the relation extraction means is a language. The language-specific relation extracting means extracts a relational expression consisting of a plurality of words that may have a relation to each other and a relation between them. Then, for each of the obtained languages, the language normalization means registers the bilingual word dictionary in which words of equivalent meaning in a plurality of languages are registered, and the information indicating the equivalent linguistic relationship in a plurality of languages. It uses a bilingual relation dictionary and normalizes it into a language-independent expression. As a result, the keyword can be extracted from the original data described in a plurality of languages by a language-independent expression.

In the invention described in claim 6, for example, a relation extracted from the original data as a plurality of relations which can have a relation with each other, such as a key extracted by the relational expression extraction device according to the first aspect. Expression is the key
It is registered in the storage means together with the original data. The original data registered in this storage means is searched by the search means using the input relational expression as a search key. As a result, it is possible to search using not only a single word but a plurality of expressions and the relationship between them, and it is possible to realize an efficient search from concepts and subjects that do not depend on surface expressions.

Among the relational expressions attached to the original data, there are cases where different relational expressions have the same meaning or concept. In the invention according to claim 7, the relationship expansion means expands the input search request into a relational expression that can be conceptually the same, and performs a search with the expanded relational expression, thereby reducing search omissions. It becomes possible. At this time, since only the same semantically and conceptually is searched, unnecessary data is not presented as a search result, and an efficient search can be realized.

In the invention described in claim 8, expression extracting means for extracting from the input data a plurality of expressions that may have a relationship with each other, and relationship extracting means for extracting a relationship between the plurality of expressions extracted by the expression extracting means. , And has a normalizing means for normalizing to an expression that does not depend on the surface expression based on the relationship extracted by the relationship extracting means, and these means are used both for extracting the keyword and for generating the search key. When extracting a keyword, the original data is extracted as input data, and a plurality of expressions that may have a relationship with each other are extracted by the word group extraction means, the relation extraction means, and the normalization means, and the normalized relational expression is used as a key together with the original data. Register in storage means. At the time of search, the natural language is used as input data, and a plurality of expressions that can be related to each other are extracted by the word group extraction means, the relationship extraction means, and the normalization means, and the normalized search request is conceptualized by the relationship expansion means. The search is performed by expanding the relational expression that can be the same as. As a result, the keyword added to the original data and the search key used for the search are generated by the same process, so that a more appropriate search can be performed. Further, since each unit can be used for both keyword extraction and search, it is possible to simplify the device configuration and realize a low-cost search device.

According to the invention described in claim 9, in the storage means, a relational expression extracted from the original data as a plurality of expressions which may have a relation with each other and normalized to a language-independent expression is used as a key. It is registered with the data.
The original data registered in this storage means is searched by the search means using the normalized relational expression independent of language as a search key. This makes it possible to search regardless of the language in which the original data is written.

According to the tenth aspect of the present invention, since the input data is searched regardless of the language, the input data are related to each other according to the language in which the input data is described in the expression extracting means. A plurality of expressions that can have the expression are extracted, and the relationship extraction unit extracts the relationship between the extracted expressions according to the language. For the extracted relations, the language normalization means uses a bilingual word dictionary in which words having equivalent meanings in a plurality of languages are registered, and a bilingual relation dictionary in which information indicating equivalent linguistic relationships in a plurality of languages is registered. , The search request is normalized to a language-independent expression. At this time, as described in claim 7, the relationship can be expanded by the relationship expanding means. On the other hand, in the storage means, the relational expression normalized to the language-independent expression by the expression extracting means, the relation extracting means, and the language normalizing means using the original data as input data is registered as a key together with the original data. . The search means searches using a search request normalized as a language-independent expression. This allows you to enter search requests in various languages, and to search for relevant information regardless of the language used in the search request and the language used in the original data. Becomes

According to an eleventh aspect of the present invention, in the relational expression retrieval device according to any one of the sixth to tenth aspects, a relational expression based on an input retrieval request,
It is possible to compare the retrieved document with the relational expression associated with the retrieved document, evaluate the certainty factor that may have the same relationship, and rearrange the search results based on the certainty factor and present them to the user. For example, the user can refer to the search results in descending order of certainty, and can perform a more efficient search.

[0028]

FIG. 1 is a block diagram showing an embodiment of a relational expression extracting device of the present invention. In the figure, 1 is a data input section,
2 is an expression extraction unit, 3 is a relation extraction unit, 4 is a normalization unit, 5 is a storage unit, 11 is a magnetic disk, 12 is OCR, and 13 is M.
T and 14 are morphological analysis units, 15 is a word group extraction unit, and 16 is a database.

The data input unit 1 provides document data from which relational expressions should be extracted. The data input unit 1 includes, for example, a magnetic disk 11, an OCR 12 and an M that holds an existing electronic document.
It can be configured by T13 or the like. Of course, the document data may be received via another input device, a storage device, or a network.

The expression extracting unit 2 analyzes the input document data and extracts an expression consisting of information such as a word and its part of speech based on a predetermined rule according to a connection pattern that may have a relationship. The expression extraction unit 2 can be configured to include a morpheme analysis unit 14 and a word group extraction unit 15.
The morphological analysis unit 14 analyzes each text in the document data and divides it into word units. The word group extraction unit 15 extracts a plurality of words that may have a relationship with each other from the analysis result by the morpheme analysis unit 14 based on the word group extraction rule.

The relation extraction unit 3 gives a relation to the expressions extracted by the expression extraction unit 2 based on the relation extraction rule. The normalization unit 4 normalizes the relational expression based on the relation given by the relation extraction unit 3. By these processing,
The desired relational expression is obtained.

The storage unit 5 can be composed of, for example, a database 16. The relationships obtained by the processes of the expression extraction unit 2, the relationship extraction unit 3, and the normalization unit 4 are registered together with the input original data.

FIG. 2 is a flow chart showing an example of the entire operation in one embodiment of the relational expression extracting device of the present invention. First, in S21, the document data to be extracted is electronically read from the data input unit 1. In S22, while unprocessed sentences remain in the read document data,
The processing after S23 is repeated.

At S23, one unprocessed sentence is taken out from the read document data. As a unit of morphological analysis in the next step, in Japanese, if even a punctuation mark (.) Is read, it is easy to read one sentence, so this is applied. Of course, if you have enough memory,
Analysis may be performed in longer units.

In step S24, the morpheme analysis unit 14 performs morpheme analysis to divide the text of one sentence into words. Morphological analysis is widely known as a basic technique of natural language processing. For example, Japanese Patent Laid-Open No. 60-20234.
Various known techniques such as the Japanese morphological analysis method described in Japanese Patent Publication can be used.

In S25, the word group extraction unit 15 performs the word group extraction processing. The word group extraction is performed using the word group extraction rule. The basic pattern of the word group extracted in the following description is the following pattern. "Pre-independent word sequence + adjunct word sequence + rear-independent word sequence"
The word group extraction rule is a rule for extracting such a word group pattern. Also, for compound expressions, 1
Extract as one expression. The extraction criterion of the compound expression is basically a series of words that are not used, and specifically, it is based on the condition that "any one of a noun, a sahen, and an adjective is continuous." Of course, it is also possible to extract the word group by other patterns, and the word group extraction rule is determined according to the pattern to be extracted.

When the word group is extracted by the word group extraction unit 15, the relation extraction unit 3 gives a relation in S26. A relation extraction rule is used to extract the relation. A word group and a relationship that the word group has are paired and registered in the relationship extraction rule. Using this relation extraction rule, if there is a registered item corresponding to the extracted word group, the relation corresponding to that word group is extracted, the extracted relation is added to the word group, and the result is output.

When the relation extraction is completed, in S27,
The normalization unit 4 normalizes the relationship based on the extracted relationship. That is, from the relationship extracted by the relationship extracting unit 3, a part that does not depend on surface expression such as utilization or appearance order is extracted. In the following description, the preceding independent word sequence of the pattern when the above-mentioned word group is extracted is defined as expression 1, the subsequent independent word sequence is defined as expression 2, and the relation extracted by the relation extraction unit 3 is defined as [relationship], { [Relationship] A relationship in the form of expression 1 expression 2} is generated. However, the expressions 1 and 2 need to be replaced in accordance with the rules because processing that does not depend on the appearance order as much as possible is required. This is to facilitate matching with similar expressions that are conceptually close at the time of actual search. As for the order of expression 1 and expression 2, basically, the expression 2 is defensive, and as a more specific rule, as shown below, in the case of the pre-independence word sequence only, the syllable stem, the post-independence word The word sequence is expressed as expression 1 and the pre-independent word sequence as expression 2. In addition, the adjective verb stem is abbreviated as Sa, and the adjective verb stem is abbreviated as adjective verb. The same applies to the following description and drawings. (A) Sa-hen + verb to move + noun → {[sul] (noun) (sa-hen)} (B) Sa-hen + is + adjective verb → {[ga case] (adjective verb)
The regular expression is obtained by the above processing.

The relational expressions extracted and normalized by the processes up to S27 do not depend on surface expressions such as adjunct words, inflectional forms, and appearance order, and represent a clearer concept than the word level. I can say. This relational expression is considered to represent the subject of the original sentence.

Thus, the relational expression extraction for one sentence extracted in S23 is completed. The content of the normalized relational expression is written to the result output stream SO, the process returns to S22, and the processing is continued in the same sentence unit as long as the data continues in the input document. Finally, in S29, the relational expression thus obtained is registered in the storage unit 5. As mentioned above,
A normal database system 16 can be used for the storage unit 5. That is, the expression 1, expression 2, and relationship of each regular expression held in the result output stream SO may be registered as transposed keys together with the input document that is the original data.

By the above processing, the subject is extracted as a relational expression from the input document and stored in the database together with the original document. In this way, a subject expression extraction system using the relational expression extraction device is realized by sequentially adding the subject to the document.

FIG. 3 is a flow chart showing an example of the operation of the word group extraction unit 15 in one embodiment of the relational expression extraction device of the present invention. As described above, the word group extraction unit 15 extracts, for example, the pattern “preceding independent word arrangement + adjunct word arrangement + successive independent word arrangement” based on the relational expression extraction rule. Independent word sequence 1 according to this pattern
Storage stream JF, auxiliary word sequence storage stream FZ,
An independent word line 2 storage stream JB is prepared. The morphologically analyzed word string is once read into the word extraction buffer WD, and accordingly, as long as the independent word continues as long as the independent word continues, it is sequentially read into each stream to perform matching with the rule.

In S31, first, the independent word sequence 1 storage stream JF, the adjunct word sequence storage stream FZ, and the independent word sequence 2 storage stream JB are cleared. If the end of the sentence is not reached in S32, one word is extracted from the word string subjected to the morphological analysis in S33, and is read into the word extraction buffer WD. In S34, it is determined whether or not the word read into the word extraction buffer WD is an independent word, and if it is an independent word, it is added to the independent word arrangement 1 storage stream JF in S35. Then, the process returns to S32 and the processing of the next word is repeated. If the word is not an independent word in the determination in S34, the word read in the word extraction buffer WD in S36 is added to the auxiliary word alignment storage stream FZ. Further, in S37, when the word group stored in the independent word sequence 1 storage stream JF is a compound expression, the compound expression alone can be one relational expression, and therefore it is written to the expression storage stream EX.

In S38, the expression extraction rule is referred to based on the pre-independent word sequence and the adjunct word extracted so far, and it is determined whether there is a matching rule. If there is no matching rule, the process returns to S31 to start searching for a new pattern. If there is a matching rule,
The position of the word read in the word extraction buffer WD in S39 is held in the position variable WP.

After determining in S40 whether or not the sentence is at the end, if the sentence is in the middle, the next word is fetched in the word fetch buffer WD in S41. In S42, it is determined whether the extracted word is an independent word. If the extracted word is an independent word, in S43, the extracted word is stored in the independent word arrangement 2 storage stream JB, and S4
Returning to 0, the next word is processed. If the word extracted in S41 is not an independent word in S42, the independent word arrangement 1 storage stream J is determined in S44.
Based on the contents of F, the attached word sequence storage stream FZ, and the independent word sequence 2 storage stream JB, it is determined whether or not there is a matching rule by referring to the word group extraction rule. If they do not match, the independent word arrangement 2 storage stream JB is cleared in S45, the process returns to S40, and a new post independent word arrangement is searched. If there is a matching rule, in S46,
The word group consisting of the independent word sequence 1 storage stream JF, the adjunct word sequence storage stream FZ, and the independent word sequence 2 storage stream JB is written to the expression storage stream EX.

When the end of the sentence is detected in S40, that is, when the pattern cannot be extracted, and S46
When the word group is detected by the steps up to and is written in the expression storage stream EX, the process of extracting the independent word group is ended, the position of the word extraction buffer WD is returned to the position of the position variable WP in S47, and S31 Returning to, processing is resumed from the processing of the next word of the position variable WP.

When the end of the sentence is detected in S32,
If the word group stored in the independent word sequence 1 storage stream JF in S48 is a compound expression, the compound expression is written to the expression storage stream EX, and the word group extraction unit 15
Is completed.

FIG. 4 is a flow chart showing another example of the operation of the word group extraction unit 15 in the embodiment of the relational expression extraction device of the present invention. In the example shown in FIG. 3, in the pattern “pre-autonomous word sequence + adjunct word sequence + post-independent word sequence”, the post-independent word sequence is extracted even if it is not continuous with the adjunct word sequence. It is also possible to extract only when this pattern is continuous, and such an example is shown in FIG.
By doing so, although there are few relationships to be extracted, it is possible to extract only relationships that can be connected more reliably. The range to be extracted may be appropriately selected according to the purpose.

In S51, the independent word sequence 1 storage stream JF, the adjunct word sequence storage stream FZ, and the independent word sequence 2 storage stream JB are cleared, and in S52, the word 1 divided into the word extraction buffer WD by the morpheme analysis unit 14 is used. Take out a word. It is determined whether or not the word extracted in S53 is an independent word. If the word is an independent word, the word is written in the independent word sequence 1 storage stream JF. If it is not an independent word, in S55, the words fetched in the word fetch buffer WD are stored in the auxiliary word sequence storage stream FZ.
Write in.

The independent word sequence 1 storage stream J
F, the word group extraction rule is referred to by the word group written in the attached word sequence storage stream FZ, and it is determined in S56 whether there is a matching word group extraction rule. If there is no matching word group extraction rule, the process returns to S51 to restart the search for a new pattern. If there is a matching word group extraction rule, in S57, if the independent word sequence 1 storage stream JF is a complex expression, it is written to the expression storage stream EX. Then, the acquisition process of the independent word sequence 2 is performed.

In S58, the word extraction buffer W
Extract one word into D. In S59, it is determined whether the extracted word is an independent word, and if it is an independent word,
In S60, the extracted word is written in the independent word arrangement 2 storage stream JB. If it is not an independent word, in S61, the independent word sequence 1 storage stream J extracted so far is stored.
Based on each word group of F, the attached word sequence storage stream FZ, and the independent word sequence 2 storage stream JB, the word group extraction rule is referred to and it is determined whether there is a matching word group extraction rule. If there is a matching word group extraction rule, S6
In 2, the expression storage stream EX outputs an expression composed of the independent word sequence 1 storage stream JF, the adjunct word sequence storage stream FZ, and the independent word sequence 2 storage stream JB.

At this time, the independent word sequence stored in the independent word sequence 2 storage stream JB is the next independent word sequence 1
Can be. Therefore, regardless of whether the word group extraction rule in S61 matches or does not match, in S63,
Independent word sequence 2 Contents of the storage stream JB are independent word sequence 1
It is copied to the storage stream JF, and the adjunct word arrangement storage stream FZ and the independent word arrangement 2 storage stream JB are cleared. Then, after determining in S64 whether or not the sentence is the end,
If it is not the end of the sentence, the process returns to S55 to continue the process, and if the extraction process is completed until the end of the sentence, if the word group in the independent word sequence 1 storage stream JB is a complex expression in S65, the expression storage The data is written to the stream EX, and the processing of the word group extraction unit 15 ends.

FIG. 5 is a flow chart showing an example of the operation of the relation extracting unit 3 in the embodiment of the relation expression extracting apparatus of the present invention. The expressions extracted by the word group extraction unit 15 are stored in the expression storage stream EX and passed to the relationship extraction unit 3. In S71, one expression is extracted from the expression storage stream EX, and in S72, the relationship extraction rule is referenced to check whether there is a match. If there is a match, the expression and the relationship are written to the relationship storage stream RL in S73. S
In 74, it is determined whether or not the expression storage stream EX has reached the end. If there is an unprocessed expression, the process returns to S71 and the processing for the next expression is repeated. When all the expressions in the expression storage stream EX have been processed, the processing of the relation extracting unit 3 ends.

FIG. 6 is a flow chart showing an example of the operation of the normalizing section 4 in the embodiment of the relational expression extracting device of the present invention. The expressions and their relationships extracted by the relationship extraction unit 3 based on the relationship extraction rules are stored in the relationship storage stream R.
It shall be stored in L and passed to the normalization unit 4. S
At 81, one set of expressions and relationships is retrieved from the relationship storage stream RL. In S82, it is determined whether the extracted expressions need to be replaced.
The necessity of replacement occurs, for example, when there is an expression of verbality in Expression 1, as described above. If it is necessary to replace them, the expressions 1 and 2 are replaced in S83. If there is no need for replacement, the process proceeds directly to S84.

At S84, the above-mentioned regular expression {[relationship] expression 1 expression 2} is generated. The generated regular expression is sequentially output to the result output stream NE in S85.
In S86, it is determined whether or not the relation storage stream RL has reached the end. If there is an unprocessed expression-relational pair remaining, the process returns to S81 to process the next expression-relational pair. Repeat. When all the sets of expressions and relations in the relation storage stream RL have been processed, the process of the normalization unit 4 ends.

In the example of the above-mentioned processing, in order to reduce the load on the computer, the analysis is limited to the range of morphological analysis, and it is treated as two independent words or independent word sequences and the relationship between them. In this case, some relationships such as complex relationships are not extracted, but most effective relationships can be estimated by morphological analysis information, and complex relationships can also be expressed as a combination of the extracted basic relationships. On the other hand, even if the syntactic analysis is used, not all the relationships can be clearly determined, so that the present embodiment can be said to be an efficient method. Of course, when computer resources and the like are available, deep analysis such as syntactic analysis and semantic analysis is also effective.

An example of the above processing will be described based on a specific example. Here, as the processing in the word group extraction unit 15, as shown in FIG. 3, the processing in the case of extracting even distant word groups is used. It is assumed that the document is read from the data input unit 1 in S21, and the following one sentence is read from the document in S23. "The system we have realized realizes a high-speed document search function by applying this method."

In S24, the morpheme analysis unit 14 performs morpheme analysis on this one sentence, and as a result, the following morpheme analysis result is obtained. We (pronouns) realized (case particles) realized (sa change) verb ending / union system (noun) is (adverb), (symbol) book (prefix) method (noun) applied (sa case) (Particle equivalent) Document (noun) (case particle) High-speed (adjective verb) Fast (adjective verb ending / union) Search (sahen) Function (sahen) (case particle) realize (sahen) ). (symbol)

The words that have been morphologically analyzed and divided as described above are extracted as word groups by the word group extraction unit 15 in S25. FIG. 7 is an explanatory diagram of an example of the word group extraction rule. The word group extraction unit 15 uses a word group extraction rule, a part of which is shown in FIG. 7, to extract a word group having a pattern of “preceding independent word arrangement + adjunct word arrangement + after independent word arrangement”. The word group extraction rule is composed of columns of independent word arrangement 1, adjunct word arrangement, and independent word arrangement 2 corresponding to the above pattern. For example, the rule of the first line is noun + "ni" + sahen, sahen + "ni" + sahen, adjective verb +
It indicates that the pattern "ni" + sahen is extracted as an expression.

First, words are read in order from the result of morphological analysis, and pattern matching with rules is performed. The morphologically analyzed word string is once read into the word extraction buffer WD, and if it is an independent word, therefore, it is sequentially read into each stream as long as the independent word continues, and matching with rules is performed.

In S31 of FIG. 3, independent word sequence 1 storage stream JF, adjunct word sequence storage stream FZ, independent word sequence 2
After clearing the stored stream JB, first in S33,
"We" (pronoun) is read into the word fetch buffer WD. Since this word is an independent word, S34 to S
35, "we" are independent words and one storage stream J
Written to F. Returning to S33, the next word "ga" (case particle) is read into the word extraction buffer WD. Since this word is not an independent word, the process proceeds from S34 to S36, and the word "ga" is written in the auxiliary word sequence storage stream FZ. Since the content of the independent word sequence 1 storage stream JF is not a compound expression, S37 is skipped. And
In S38, it is determined whether or not the pattern "pronoun + ga (case particle)" matches within the word group extraction rule as shown in FIG. 7, for example. This pattern does not exist in FIG. Therefore, the process returns to S31 and the contents of each stream are cleared.

Then, similarly, "realization" (sa change) is written in the independent word sequence 1 storage stream JF, and "done" (the verb ending / union) is written in the auxiliary word sequence storage stream FZ. Then, the word group extraction rule is referred to in S38. At this time, since the pattern of "sahen + wa (do verb)" corresponds to the rule shown in the bottom line of the word group extraction rules shown in FIG. 7, it is held as it is. Further, in S39, the position of "done" is held in the position variable WP.

Further, in S41, "system"
(Noun) is read into the word extraction buffer WD, and S
At 42, it is determined whether the word is an independent word. Since "system" is an independent word, it is written as independent word arrangement 2 in independent word arrangement 2 storage stream JB in S43. The next word "ha"
Although (sub particle) is read into the word extraction buffer WD, but it is not an independent word, the process proceeds from S42 to S44, and the word group extraction rule is referred to. From each stream at this time, it is determined whether or not the pattern “sahen + da + noun” is in the word group extraction rule. Since the rule shown in the bottom line of the word group extraction rule shown in FIG. 7 is applicable, the expression "realized / executed / system" is extracted, and in S46,
It is output to the expression storage stream EX. Hereinafter, "/" is attached between the independent word and the attached word.

In S47, the position held in the position variable WP, that is, the position of the word "do" is returned to S47.
Each stream is cleared at 31 and the processing is further continued. The next word "system" is read and the independent word sequence 1
It is written in the storage stream JF, and "ha" is written in the auxiliary word sequence storage stream FZ. Since the pattern of “system” (noun) + “wa” (sub particle) does not exist in the word group extraction rule shown in FIG. 7, each stream is cleared. The subsequent "," does not match the word group extraction rule because it is a symbol, and the processing for the next word is started. Next, "book" (prefix) is read, but this is also ignored because it is not an independent word.

This time, the "method" is written in the independent word sequence 1 storage stream JF. Since the subsequent word "apply" is also an independent word, it is written in the independent word sequence 1 storage stream JF. This "method application" is a so-called compound word.
An independent word sequence without an accompanying word is also a target for extraction as a compound expression. At this time, the compound words are collectively treated as a single noun. This is because, as described above, since only two independent word sequences and the relationship between them are handled, it is necessary to handle the independent word sequence 1 as a unit. For the same reason,
Even when extracting three or more independent words arranged independently, a relationship other than the direct connection of each word is not specified.

Next, the next word "by" is read, but at this time, the preceding independent word sequence is treated as a compound noun. Since the word "by" is a case particle equivalent word, it is written in the adjunct word storage stream FZ as an adjunct word list. In S37, the independent word sequence 1 storage stream JF
Since the complex expression is written in this, this sequence of independent words "method | application" is extracted and written to the expression storage stream EX. Hereinafter, '|' is attached between each word in the independent word sequence. At this point, "applying method" + "by"
The pattern (noun + case particle equivalent) is held. This pattern corresponds to the pattern on the fifth line of the word group extraction rule in FIG.

Next, the "document" (noun) is a sequence of independent words 2
It is written in the storage stream JB, and the next word “NO” is read in the word fetch buffer WD. Since it is not an independent word, the process proceeds to S44. In S44, it is determined whether the pattern "noun + by + noun" matches the word group extraction rule. In the pattern on the 5th line of the word group extraction rule of FIG. 7, the words that can be connected as the independent word sequence 2 are limited to the sahen or adjective verbs. Therefore, the following word "document" (noun) does not match and this pattern is not extracted. S4
In step 5, the independent word line 2 storage stream JB is cleared, and S4
Returning to 0, matching with distant words is performed. At this time, the patterns up to the independent word sequence 1 and the adjunct word sequence are fixed and considered.

This time, "high speed" (adjective verb) is written in the independent word array 2 storage stream JB, but "noun + by + adjective verb" does not match the word group extraction rule. Furthermore, "search" (sa-hen) and "function" (sa-hen) are written in the independent word array 2 storage stream JB as a compound expression, but since the part of speech of the compound expression is treated as a noun, "noun +
Therefore, the pattern of "+ noun" does not match the word group extraction rule. Next, "realization" (sa-hen) is written in the independent word array 2 storage stream JB. The pattern of “noun + by + sahen” matches the word group extraction rule. for that reason,
In S46, the word group "method | application / by / realization" is extracted and output to the expression storage stream EX.

At this point, the position variable WP holds the position of "due". In S47, the position of the word extraction buffer WD is returned to this word, and the extraction process is restarted from the next word. Similarly to the above-mentioned processing, "document" is written in the independent word sequence 1 storage stream JF, and subsequently "no" is written in the auxiliary word sequence storage stream FZ.
The pattern is "document" + "no" (noun + case particle). The position "WP" is held in the position variable WP, and the independent word sequence 2 is searched. First, "high speed" (adjective verb) is written in the independent word sequence 2 storage stream JB, and "na"
(Adjective verb ending) is read into the word extraction buffer WD. However, the pattern of "noun + no-adjective verb" does not match the word group extraction rule and is skipped.
This time, since "search (sahen)" and "function (sahen)" follow, these are collectively stored in the independent word array 2 storage stream JB as compound nouns. As a result, the pattern matching with the word group extraction rule was successful in the pattern of "noun + (case particle) + (composite) noun", the "document / no / search | function" was extracted, and the expression storage stream EX was extracted. Is output.

The pointer is returned to the position "NO" held in the position variable WP, and from the next word, again,
By performing the extraction processing, "high speed / na / search / function" further performs extraction processing from the word next to "na", and the independent word sequence "search / function" and "search / function / "/ Realization" is extracted.

By the above processing, the following word groups are extracted from the expression storage stream EX. Realization / Achieved / System (Sahen + "Shi" + Noun) Method | Apply (Compound Noun (Noun + Sahen)) Method | Apply / By / Realize (Compound Noun (Noun + Sahen))
+ “By” + Sahen) Document / no / search | function (noun + “no” + compound noun (sahen + sahen)) High-speed / na / search | function (adjective verb + “na” + compound noun (sahen + sahen) ) Search | Function (Compound noun (Sahen + Sahen)) Search | Function / Achieve (Compound noun (Sahen + Sahen) +
"Wa" + sa strange)

When the word group extraction unit 15 extracts the word group as described above, the relation extraction unit 3 adds a relation. FIG. 8 is an explanatory diagram of an example of the relation extraction rule. The relationship extraction unit uses, for example, a relationship extraction rule partially shown in FIG. The relation extraction rule is the word group extraction unit 1.
The independent word sequence 1, the independent word sequence 2, and the adjunct word sequence extracted in 5 are associated with the relationship to be given. For example, the first line in FIG. 8 shows that a relation [ga case] is given to the pattern "noun or sahen +" ga "+ sahen or adjective verb". The relationship extracting unit 3 extracts a word group from the expression storage stream EX, extracts a word that matches the relationship extraction rule, adds a relationship, and outputs the relationship storage stream RL.

As described above, in this example, since the relation between words based on the morpheme analysis is handled, the added relation is at a level obtained from the morpheme information. However, it is possible to obtain a more detailed relationship by performing a deep analysis.

In the above example, first, in S71 of FIG. 5, the first expression "realized system" is extracted from the expression storage stream EX. In S72, the relation extraction rule is referenced. In this case, independent word sequence 1 and independent word sequence 2 are
Since they are sa-hen and noun, respectively, and "do" is an inflectional form of "suru verb", the relation [suru] is obtained from the relation extraction rule shown in FIG. Therefore, in S73, the relation [suru] is added to the expression and is output to the relation storage stream RL.

The following expression "applying method" is a compound expression consisting of two independent words. In a compound expression consisting of only independent words, in the case of two words, a relationship of "direct connection" is given between both parts of speech.

Similarly, a relationship is obtained for each expression in the expression storage stream EX, and as shown below, the relationship is added to the expression and output to the relationship storage stream RL. Realize / Done / System (Sahen + “do” + noun) [Sul] Method / Apply (Noun + Sahen) [Direct connection] Method | Apply / By / Realize (Compound noun + “By”)
+ Sa-hen) [De-case] Document / no / search | function (noun + "no" + compound noun) [noun connection] Fast / na / search | function (adjective verb + "na" + compound noun)
[Na case] Search | Function (Sahen + Sahen) [Direct connection] Search | Function / Achieve / (Compound noun + "wa" + Sahen) [wo case]

When the relation extracting and relation giving process in the relation extracting unit 3 is completed, the normalizing unit 4 normalizes the relation in the normalizing unit 4 based on the pair of expression and relation output to the relation storing stream RL. Sequential result output stream N
Output to E. That is, by normalization, parts that do not depend on surface expression such as utilization and appearance order are extracted from the extracted relation. As described above, in the normalization processing, the independent word sequence 1 and the independent word sequence 2 are exchanged as necessary to generate a regular expression of the form {[relationship] expression 1 expression 2}.

In the above-mentioned example, first, in S81 of FIG. 6, the first expression-relationship pair is extracted from the relation storage stream RL. The extracted "realization / done / system (sahen +" do "+ noun) [sul]" is a pattern that requires replacement of expressions. In S83, the expression 1 and the expression 2 are replaced with each other, and in S84, the regular expression {[sul] system realization} is generated. The obtained regular expression is sequentially written in the result output stream NE in S85.

Subsequently, since "method / application (noun + sahen) [direct connection]" is a direct connection expression, the regular expression {[direct connection] method application} is generated as it is and sequentially output to the result output stream NE. . Thereafter, the normalization process is similarly performed, and the following result is output to the sequential result output stream NE. {[Through] system realization} {[direct connection] method application} {[de-case] method | application realization} {[noun connection] document search | function} {[na case] high-speed search | function} {[direct connection] search function } {Search for [Wo] | Function realization}

This is the end of the relational expression extraction processing for one sentence shown in the example. The content of the relational expression that has been normalized and sequentially output to the result output stream NE is written to the result output stream SO in S28 of FIG. 2, returns to S22, and as long as data continues in the input document, 1 Continue processing sentence by sentence. Finally, in S29, the relational expression thus obtained is registered in the storage unit 5. As a result, the relationship among the regular expressions held in the result output stream SO, Expression 1 and Expression 2, are registered as keys into the storage unit 5 together with the input document which is the original data.

When a document registered in the storage unit 5 is searched, the search is performed using Expression 1, Expression 2, and the relation as search keys. As a result, it becomes possible to search from concepts and subjects that do not depend on surface representation.
By obtaining semantically matching items in the search, the matching rate is improved and efficient search is possible.

In the above-described specific example, the processing shown in FIG. 3 is used as the processing of the word group extraction unit 15. However, the process shown in FIG. 4 can be used. In this case, in the case of the above-mentioned example sentence, the word group extraction unit 15 outputs the following expression. Realization / achieved / system (sahen + “ta” + noun) Method | Apply (compound noun (noun + sahen)) High-speed / na / search | Function (adjective verb + “na” + compound noun (sahen + sahen)) Search ｜ Function (Compound noun (Sahen + Sahen)) Search | Function / Implementation (Compound noun (Sahen + Sahen) +
"Wa" + sa strange)

Based on the extracted expressions, the relation extraction unit 3 gives the relations, and the expressions having the following relations are obtained. Realization / achieved / system (sahen + “do” + noun) [sul] Method / application (noun + sahen) [direct connection] Fast / na / search | function (adjective verb + na) + compound noun)
[Na case] Search | Function (Sahen + Sahen) [Direct connection] Search | Function / Achieve / (Compound noun + "wa" + Sahen) [wo case]

Further, the normalizing section 4 performs normalization to obtain the following regular expression. {[Through] system realization} {[direct connection] method application} {[na case] high-speed search | function} {[direct connection] search function} {[wo case] search | function realization}

The regular expression thus obtained is stored as a key in the storage unit 5 together with the document.

In one embodiment of the above-described relational expression extracting device,
Although the configuration including the normalization unit 4 has been described, the present invention
It is also possible to adopt a configuration that does not include normalizing means. That is, if the system is not a system based on the above-described search but is, for example, a system that only extracts a concept from the original data, the normalizing means is unnecessary. In this case, the configuration is realized by removing the normalization unit 4 from the configuration shown in FIG. 1 and connecting the relationship extraction unit 3 to the storage unit 5. As for the flow of processing, only the normalization processing performed in step S27 of FIG. 2 is omitted. Specifically, the result of extracting the relational expression from the relational expression storage stream RL may be output to the result storage stream SO as it is.

The expression as it is extracted by the relation extracting unit 3 is
It is considered that it represents the concept of the original data, and by appropriately editing these, it is possible to create an abstract of a document or an OHP presentation material.
By applying this, a summary creation support system can be realized.

It goes without saying that the relational expression extraction device of the present invention is applicable to general keyword extraction from text data and is not limited to the above-mentioned embodiment. For example, it is possible to extract the relationship from the user's search request instead of extracting it from the document.

Conventionally, as a method of giving a search request, there are a method of using a search language such as SQL and a so-called natural language interface. However, these are eventually extensions of the keyword search and result in logical operation of keywords in word units such as AND search.

Therefore, it is conceivable to use the relational expression extracting device of the present invention for inputting a search request. That is, a pattern of "independent word + adjunct word + independent word" is given as a search request, and a relational expression is extracted from the pattern.

For example, even when the user gives a natural expression such as “create a document”, “create from a document”, or “created document”, a search key such as “document & create” is not distinguished by the conventional natural language interface. Was generally generated. On the other hand, by using the relational expression extraction device of the present invention, it is possible to “create a [Wo”] document respectively.
Expressions with relations such as "[color] document creation" and "[through] document creation" can be extracted. As a result, the user can easily give his / her intention as a search request using a very natural notation.

As an implementation method, in the configuration shown in FIG. 1 described above, the output from the normalization section 4 or the relation extraction section 3 may be provided to the search means. At this time, the relational expression needs to be added as a key to the searched database, as in the above-described embodiment.

As described above, the relational expression extracting device of the present invention is applied to the scene of generating the key to be added to the information registered in the database and the scene of generating the search key based on the search request from the user. be able to. That is, it is possible to realize a relational expression search device configured to perform the same relational expression extraction process when information is registered in the database and when information in the database is searched. Hereinafter, such a relational expression search device will be described.

FIG. 9 shows a first example of the relational expression retrieval device of the present invention.
It is a block diagram showing an example of. In the figure, the same parts as those in FIG. Reference numeral 6 is a search request input unit, 7 is a relationship expansion unit, 8 is an output unit, 17 is a keyboard, 18 is a mouse, and 19 is a CRT display. The search request input unit 6 includes a keyboard 17 and a mouse 18.
And various other well-known input means. The user inputs a search request based on the relationship from the search request input unit 6. The expression extraction unit 2 analyzes the text input from the current data input unit or the search request input unit 6, performs morpheme analysis by the morpheme analysis unit 14, divides into words, and the word group extraction unit 15 performs word group extraction. Information such as a plurality of words and their parts of speech that may have a relationship with each other based on the extraction rule is extracted as an expression. The relation extracting unit 3 gives a relation to the expressions extracted by the expression extracting unit 2. The normalization unit 4 normalizes the relation-added expression based on the relation given by the relation extraction unit 3, and obtains a relational expression to be obtained.
The relational expression obtained by the normalization unit 4 is
It is registered in the database in the storage unit 5 together with the input original data. On the other hand, at the time of search, the regular expression is passed to the relationship expansion unit 7. The relationship expansion unit 7 expands the regular expression obtained by the normalization unit 4, generates an expression that can be conceptually the same, and searches the storage unit 5. The output unit 8 is, for example, a CRT.
It is composed of display means such as the display 19 and outputs the search result obtained by the search.

Next, an example of the operation of the first embodiment of the relational expression search device of the present invention will be described. Here, the registration of the relationship and the search from the relationship will be described separately.

First, the operation of extracting a relationship from original data such as a document and storing it in the storage unit 5 will be described. The operation in this case is similar to the operation in the above-described embodiment of the relational expression extraction device. That is, the document data to be extracted is electronically read from the data input unit 1, and one unprocessed sentence is extracted from the read document data. The extracted one sentence is divided into words by morphological analysis. Then, the word group extraction unit 15 extracts the word group that matches the rule based on the word group extraction rule. The basic pattern of the extracted word group can be, for example, a pattern of "preceding independent word arrangement + adjunct word arrangement + rear independent word arrangement" and a complex expression. The expressions extracted by the word group extraction unit 15 are given a relationship by the relationship extraction unit 3 based on the relationship extraction rule. Then, the normalization unit 4 replaces the word groups as necessary to generate a regular expression. As a regular expression, for example, {[relationship] expression 1
Expression 2}.

This is the end of the relational expression extraction for one sentence. Further, as long as the data continues in the input document, the processing is similarly continued for each sentence. Finally, the relational expression thus obtained is registered in the storage unit 5 together with the input document. At this time, the expression 1, expression 2, and relationship of each regular expression may be registered as transposed keys together with the input document that is the original data.

Through the above processing, the subject is extracted as a relational expression from the input document and stored in the database together with the original document.

Next, the search based on the relationship extracted in advance will be described. The search is mainly performed by inputting a search request, extracting a relational expression, developing a relation, and inquiring a database. FIG. 10 is a flowchart showing an example of a search process in the first embodiment of the relational expression search device of the present invention. First, in S91,
The search request input unit 6 inputs the search request as a natural language.

When a search request is input, S92 to S92
By the processing of 95, the normalized relational expression is extracted as in the case of registration. For the processing at this time, the same algorithm as when extracting the relational expression from one sentence at the time of registration can be applied. For example, a regular expression can be obtained by using the processing shown in FIGS.

When the regular expression is obtained, in S96,
The relationship expansion unit 7 refers to the relationship expansion rule and determines whether there is a matching relationship expansion rule. If there is a relation expansion rule, the relation expansion rule is used in S97 to generate an expression that can be conceptually identical to the regular expression. In S98, the storage unit 5 is searched using the regular expression and the expanded expression generated in S97 as a search key. The obtained search result is output to the output unit 8 in S99.

An example of the above operation will be described based on a specific example. As a specific example, suppose that the user wants to search for documents related to "what realized a system". In S91, the search request is input from the search request input unit 6. At this time, as an actual search request, for example, an expression "a system is realized" which is very close to the original search intention is given in a natural language.

When the search request is input, in S92, the search request "system realized" is first issued.
The morphological analysis unit 14 performs morphological analysis. The results of morphological analysis are shown below. The system (noun) is realized (case particle) (sa strange) (is a verb ending / union)

Then, in S93, the word group extraction unit 1
In step 5, the word group is extracted. As the pattern to be extracted at this time, “preceding independent word arrangement + adjunct word arrangement + sufficient independent word arrangement” is used, and if the word group extraction rule as shown in FIG. 7 is used, the following one expression is obtained. System / Achieve / Realize (Noun + "wo" + Sahen)

Further, in S94, the relation extraction unit 3 gives a relation. At this time, for example, if a relation extraction rule as shown in FIG. 8 is used, [wo case] is given to the above-mentioned expression extracted by the word group extraction unit 15, and an expression with the following relation is given. can get. System / Implement / Realize (Noun +
"Wa" + Sa strange) [Wo]

Further, in S95, the normalizing section 4 generates a regular expression. Assuming that the regular expression generated at this time is {[relationship] expression 1 expression 2}, the following regular expression is obtained from the expressions to which the above-mentioned relationship is given.
In this example, it is not necessary to switch the independent word sequence, so
As it is, the front independent word sequence is represented as expression 1 and the rear independent word sequence is represented as expression 2. {[Wo] system realization}

In this way, a normalized relational expression was obtained. In the case of a search request, when the normalized relational expression is obtained, it is sent to the relation expansion unit 7. Relationship development department 7
In, the normalized relational expressions are expanded to generate what can be conceptually the same. The relation expansion rule is used for expansion. FIG. 11 shows a first example of the relational expression search device of the present invention.
6 is an explanatory diagram of an example of a relationship expansion rule in the embodiment of FIG.
As shown in FIG. 11, in the relation expansion rule, the relation in the regular expression obtained by the normalization unit 4, the set of the expressions 1 and 2 and the expandable relation are registered as a pair. . In the above example, the relation is o, the expression 1 is a noun, and the expression 2 is sa, so it matches the rule on the second line of FIG. 11. for that reason,
The process proceeds from S96 to S97. The expression obtained at this time is
It can be seen that it can be expanded to [No Case], [Direct Connection], and [Through] in addition to [Wo Case]. Accordingly, the following relational expression is generated in S97. {[Wo] system realization} {[No case] system realization} {[Direct connection] system realization} {[Thru] system realization}

Then, in S98, an inquiry is made to the database system 16 based on the expanded relational expression. In this case, a search is performed using each obtained relational expression as a key. The search result is the CRT in S99.
It is output to the display 19.

The candidates obtained as a result of the search can be conceptually the same. For example, the document extracted in the above-described specific example of the relational expression extracting device and registered in the storage unit 5 has the relational expression {[Sul] system realization}, and is searched. This document is a good candidate, since it describes at least the implementation of a system.

On the other hand, for example, another document is assumed to have the phrase "function realized by this system". The relational expression extracted from this phrase is as follows. In this case, Expression 1 and Expression 2 are exactly the same, but since the relation [ga case] is not included in the expansion result, it is not a candidate unless there is a relational expression that matches other parts in the document. In fact, “function realized by this system” represents a concept different from “function realized by the system”, and therefore it is appropriate that it is not searched.

As described above, in this embodiment, the concept required by the user can be easily retrieved using the natural language without worrying about the surface structure. Further, by performing a search through the relational expression, the appearance of unnecessary candidates is reduced.

In the first embodiment of the relational expression search device described above, the natural language is input as the search request, but the relational expression search device of the present invention is not limited to the input of the natural language as the search request. . For example, it is of course possible that the user gives the relation, the expression 1, and the expression 2 output from the normalization unit 4 and retrieves them. In this case, the input from the search request input unit 6 is directly input to the relationship expansion unit 7. Alternatively, the expression extracting unit 2, the relationship extracting unit 3, and the normalizing unit 4 may be configured to be input to the relationship expanding unit 7 without performing processing.

FIG. 12 is an explanatory diagram showing an example of a search request input screen using the GUI in the first embodiment of the relational expression search device of the present invention. In the figure, 101 is an expression 1 input section,
102 is an expression 2 input unit, 103 is a relation selection unit, 104 is a search instruction button, 105 is a search result list subwindow, and 106 is a content display subwindow. In recent years, G
UI (Graphical User Interface)
ce) is remarkable. So, using the GUI,
FIG. 12 shows a window image for inputting the relation, expression 1, and expression 2 as a search request. here,
An example of a case where the display for input in the search request input unit 6 and the display of the search result in the output unit 8 are displayed in one window is shown.

The user selects the expression 1 and the expression 2 in the expression 1 input section 101 and the expression 2 input section 102 at the upper left of the window.
Enter in word units. As for the relation, a button indicating each relation arranged in the relation selection unit 103 on the upper right of the window is selected by the mouse. In FIG. 12, “system” is input as the expression 1, “realization” is input as the expression 2, and the case is selected as the relationship. If you enter like this,
Since the search request can be handled as a relational expression as it is, the relational expression can be expanded and searched by passing it to the relational expansion unit. To develop the relational expression and execute the search, the search instruction button 104 at the upper left of the window may be selected with the mouse.

When the search is successful, the titles of the searched candidates are displayed in the search result list subwindow 105 at the lower left of the window together with related information. The user
You can easily select the desired result by referring to this title. In FIG. 12, the candidates searched by the search key expanded to the relation [Suru] are selected on the list displayed in the search result list subwindow 105. By this selection, the contents of the document are displayed in the contents display subwindow 106 at the lower right.

As described above, the method not relying on the natural language interface is particularly effective when it is difficult to incorporate a complicated process such as a portable search device. At this time, the original data input unit 1, the expression extraction unit 2, the relation extraction unit 3, and the normalization unit 4 are necessary only when the relational expression and the document are registered in the storage unit 5 in association with each other. It does not need to be installed in the case of a type search device. In the storage unit 5, for example, the relational expression and the document may be registered in advance in association with each other using the relational expression extraction device described above. Of course, the original data input unit 1, the expression extraction unit 2, the relation extraction unit 3, and the normalization unit 4 may be installed for registering and updating the document.

The relational expression search device described above can also be configured without using the normalization unit 4.

According to the above-described relational expression search device, it is possible to search using the relational expression input as a search request or the relational expression extracted from the search request input in natural language as a key. At that time, it is possible to expand the search request into a relational expression that can be conceptually the same and perform the search. For example, if a search request of {[no case] system realization} is input as a search request, the relational expressions that can be conceptually the same are {[no case] system realization}, {[wo case] system realization. }, {[Direct connection] system realization}, etc., and search for all that match any of them.

However, not all the developed relational expressions always represent the same relation as the search request. In some cases, it may represent a relationship different from the search request. In other words, the degree of certainty that they are conceptually the same differs depending on the expression. In the above-mentioned device, the user can know what kind of relationship in the search result is matched, but the user must judge whether it is really conceptually the same or not. However, it is difficult to efficiently obtain search results when there are many search results. As an apparatus that solves this problem, an example of an apparatus that rearranges search results using a certainty factor and presents them to the user to perform an efficient search will be described below.

FIG. 13 is a block diagram showing a second embodiment of the relational expression search device of the present invention. In the figure, parts similar to those in FIG. 9 are designated by the same reference numerals, and description thereof will be omitted. Reference numeral 20 is a certainty factor evaluation unit.

The certainty factor evaluation unit 20 compares the expression used for the search with the regular expression to evaluate the certainty factor, and rearranges the search results based on this. The output unit 8 uses the certainty factor evaluation unit 2
The search results sorted by 0 are output.

Next, an example of the operation of the second embodiment of the relational expression search device of the present invention will be described. Also in the relational expression retrieval device of this embodiment, the relational expression needs to be stored as a key in the storage unit 5 at the time of retrieval.
The registration of the relationship is the same as that of the above-mentioned embodiment, and the description thereof is omitted.

The search based on the relationship extracted in advance will be described. FIG. 14 is a flowchart showing an example of search processing in the second embodiment of the relational expression search device of the present invention. The same processes as those in FIG. 10 are denoted by the same numbers. The process from the input of the search request in S91 to the search instruction and execution of the search in S98 is the same as in the above-described embodiment.

In S98, the expanded expression generated by the relationship expansion unit 7 is used as a search key to search the storage unit 5. When the search result is obtained, the certainty factor evaluation unit 20
In S111, the expanded expression used for the search is compared with the regular expression, and the certainty factor is evaluated based on the relation evaluation rule. Subsequently, in S112, the search results are rearranged based on this evaluation result, and in S113, the search results are output to the output unit 8.

An example of the operation of the above-described relational expression search device in the second embodiment will be described based on a specific example. As a specific example, it is assumed that the user wants to search for documents related to "what realized some system", as in the above-described embodiment. In S91, the search request is input from the search request input unit 6. After the morphological analysis in S92, the word group extraction in S93, and the relation extraction in S94, the normalized relation expression {[wo case] system realization} is obtained in S95.

Next, the relationship expanding unit 7 expands the normalized relationship expression to generate what can be conceptually the same. FIG. 15 is an explanatory diagram of an example of a relation expansion rule in the second embodiment of the relational expression search device of the present invention. In the relationship expansion rule shown in FIG. 15, only the relationships in the regular expression obtained by the normalization unit 4 and the expandable relationships are registered as a pair. In this way, the relation expansion rule shown in FIG. 11 can be simplified. In the above example, since the relation is wo case, it matches the rule on the second line in FIG.
Therefore, it can be seen that the regular expression can be expanded to [No case], [Direct connection], and [Thru] in addition to [Wo case]. Accordingly, the following relational expression is generated in S97. {[Wo] system realization} {[No case] system realization} {[Direct connection] system realization} {[Thru] system realization}

Then, in S98, an inquiry is made to the database system 16 based on the expanded relational expression. In this case, a search is performed using each obtained relational expression as a key. As a result, for example, when documents having the relational expressions of {[wo case] system realization}, {[no case] system realization}, and {[suru] system realization} are registered in the storage unit 5, respectively. The document is searched.

When the search result is obtained, the certainty factor evaluation unit 20
Compares the relationship used in the search in S111 with the relationship of the original regular expression to evaluate the certainty factor. For example,
When a document with a relational expression of {[no case] system realization} is searched, [no case] and the original relationship [wo case]
To compare. The relational expressions developed in S97 can be conceptually the same as the original expressions, but they cannot be said to be the same expressions, so the same certainty factor is given as the evaluation value.

The certainty factor can be evaluated by using, for example, a relation evaluation rule. FIG. 16 is an explanatory diagram of an example of the relation evaluation rule in the second embodiment of the relational expression search device of the present invention. In the example shown in FIG. 16, whether or not the relationships are exactly the same, for example, [wo case] and [wo case], more specifically, the relationship is more concrete, for example, from [no case] to [wo case]. The evaluation value is determined in advance depending on whether the relationship is expanded in the direction of, or the relationship is expanded in the direction of abstraction from [wo case] to [no case].
As described above, from the normalized relational expression {[wo case] system realization} and the expanded relational expression, {[wo case] system realization}, {[no case] system realization}, {[thru] When a document having a relation expression "system implementation" is retrieved, based on FIG. 16, each relation expression {[wo case] system implementation},
The {[no case] system realization} and {[thru] system realization} are given confidence evaluation values of 100, 80, and 80, respectively.

At S112, the certainty factor evaluation unit 20
For example, the search results are sorted in descending order based on the evaluation points. Then, in S113, the sorted search results are output to the CRT display 19.

As described above, according to the present embodiment, the search results can be presented to the user in descending order of certainty, and the user can efficiently refer to the results.

The evaluation of the certainty factor is not limited to the above method, and various methods can be used. For example, a certain amount of data may be used to calculate the probability that one relational expression and another relational expression have the same relationship, and the certainty factor may be determined based on this. Further, instead of giving the certainty factor between the respective relations as in the above-described embodiment, the relational abstraction level or the like is given to each relation, and the certainty factor is calculated based on the difference. Good. Further, in displaying the search results, not only may the images be sorted and displayed based on the certainty factors, but the certainty factors may be displayed together.

In the second embodiment of the relational expression retrieval device described above, the retrieval request is not limited to the input of the natural language as in the previous embodiment, and the relation output from the normalization unit 4 is used, for example. Of course, it is also possible for the user to search by giving Expression 1 and Expression 2. In this case, the input from the search request input unit 6 is directly input to the relationship expansion unit 7. Alternatively, the expression extracting unit 2, the relationship extracting unit 3, and the normalizing unit 4 may be configured to be input to the relationship expanding unit 7 without performing processing.

FIG. 17 is an explanatory diagram showing an example of a search request input screen using a GUI in the second embodiment of the relational expression search device of the present invention. Reference numerals in the figure are the same as those in FIG. Also in the second embodiment of the relational expression search device described above, the search request can be input using the GUI as shown in FIG. Expression 1 input unit 101, Expression 2
Using the input unit 102 and the relation selection unit 103, the expressions 1 and 2 and their relations are input. Then, the search is performed by selecting the search instruction button 104. If the search is successful, the titles of the searched candidates are sorted in descending order of certainty in the search result list subwindow 105 at the lower left of the window, and as shown in FIG. 17, displayed together with information on the certainty and relationship. The user can refer to this title and easily select the desired result.

As described above, the method not relying on the natural language interface is particularly effective when it is difficult to incorporate a complicated process such as a portable search device, as in the case of FIG. The relational expressions extracted in advance may be registered in the storage unit 5 in association with the document. At this time, it is possible to configure without using the original data input unit 1, the expression extraction unit 2, the relation extraction unit 3, and the normalization unit 4. Alternatively, only the original data input unit 1 may be installed, and the relational expressions and the original data extracted in advance may be input to the data input unit 1 and stored in the storage unit 5 as they are.

In the second embodiment of the relational expression search device described above, the relational expansion unit 7 develops and expands the regular expression obtained by the normalization unit 4 into an expression that can be conceptually the same. You are searching for documents that have a relational expression that matches the specified expression as a key. However, the present invention is not limited to this, and a search unit may be provided in place of the relationship expansion unit 7 to directly search for an expression that can be conceptually the same even if it does not match the regular expression. In this case, the certainty factor evaluation unit 20 may evaluate the certainty factor by comparing the regular expression with the relational expression attached as a key to the document.

FIG. 18 is a block diagram showing a third embodiment of the relational expression search device of the present invention. In the figure, FIG. 9 and FIG.
The same parts as those in 3 are assigned the same reference numerals and explanations thereof will be omitted.
121 is an analysis switching unit, 122 is a language analysis unit, 12
3 is a language-specific relation extraction unit, 124 is a bilingual word dictionary, and 125.
Is a bilingual relation dictionary, 126 is a language normalization unit, 127 is a registration unit, and 128 is a search unit.

The above-described embodiments are applicable only to a single language. That is, there is no problem when the original data described in Japanese is stored in the database and they are searched in Japanese. However, if the original data having similar contents is described in another language such as English or German and stored in the database, these data cannot be retrieved. It is also possible that the search request is input in various languages. In that case, regardless of the input language, it is desirable to be able to retrieve data described in another language as long as the data has similar contents. This third
In this embodiment, an example of searching a database including such multi-languages is shown.

The data input unit 1 is a part for inputting document data which is a search target and from which a relational expression serving as a key for the search is to be extracted. The input document data may be described in other languages such as English, German, French, etc. in addition to Japanese. Further, the document data may include pictures and figures as long as it includes at least text described in natural language.

The expression extraction unit 2 analyzes the input data in accordance with the language in which the text in the input data is described, and extracts a relational expression consisting of a plurality of words that may have a relationship with each other and the relationship between them. . The expression extraction unit 2 can be configured to include an analysis switching unit 121 and a language-specific analysis unit 122.

The analysis switching section 121 selects a corresponding one from the plurality of language-specific analysis sections 122 according to the language in which the text in the input data is described. The type of language in which the text is described is recorded in the input data in advance, or is input together with the input data from the data input unit 1 or the search request input unit 6 when the input data is input, or the characteristics of the text are analyzed. It is obtained by doing. As a method of analyzing a text, there is a method of determining by obtaining each word that constitutes the text using a bilingual dictionary or the like. The bilingual dictionary used at this time is sufficient as long as it can discriminate each language, so that it can be sufficiently used even if it has about 10 words. Here, it is assumed that the type of language is recorded in advance in the document data in the data input from the data input unit 1. As a method of describing the type of language, there are a method of describing it separately from the text of the text like a title and a creation date, and a method of embedding it in the text by using a tag or the like. The latter method can deal with the case where a part of the text is described in another language. In this embodiment, for simplicity, a method described separately from the text of the former text is adopted.

The language-specific analysis unit 122 analyzes the text in the document data. A plurality of language-specific analysis units 122 are prepared for each language type, such as Japanese, English, German, and French. The analysis unit that analyzes the corresponding language is selected according to the determination result of the analysis switching unit 121. Each analysis unit has a grammar rule, a dictionary, etc. dedicated to that language, and has a configuration suitable for analyzing each language.
However, a common analysis engine may be used. On the other hand, it is also possible to change the analysis level for each language.
For example, English and German can be configured by using a common language analysis unit, and Japanese can be configured by using a morphological analysis unit as in each of the above-described embodiments.

The language-specific relation extraction unit 123 extracts a relational expression consisting of a plurality of words that may have a relation with each other and the relation between them, from the analysis result of the language-specific analysis unit 122. As with the language-specific analysis unit 122, a plurality of language-specific relation extraction units 123 are also prepared for each language type, and the corresponding language is selected. Further, like the language-specific analysis unit 122, the language-specific relationship extraction rule and other information are provided, and the configuration is suitable for each language.

In the bilingual word dictionary 124, words having equivalent meanings in a plurality of languages are registered. Further, the bilingual word dictionary 124 can be considered as an extension of a synonym dictionary in which words having similar meanings to a certain word are expanded to a plurality of languages. FIG. 19 is an explanatory diagram of an example of the data of the bilingual word dictionary. For example, in a Japanese synonym dictionary,
With respect to the word "production", "manufacturing", "production", "production", etc. are registered as synonyms. In the bilingual word dictionary 124, words having similar meanings are recorded not only in the same language but also in a plurality of languages. For example,
As shown in FIG. 19, with respect to the same word “production”, “produce” and “manufact” in English.
"ure", "create", etc. are "s" in German.
“Chaffen”, “produzieren”, etc. are registered. In the bilingual word dictionary 124 shown in FIG. 19, these words are held as a list in pairs with symbols indicating the respective languages. When you input, the equivalent word in a predetermined language is output. When there are multiple equivalent words in the output language, for example, output the first word of the list or a list of words. With this bilingual word dictionary 124, it is possible to input words in various languages and obtain similar words unified in a predetermined language.

In the bilingual relation dictionary 125, relation names representing equivalent linguistic relations in a plurality of languages are registered.
FIG. 20 is an explanatory diagram of an example of data in the bilingual relation dictionary. Here, "ga" and "wo" in the above examples
"GA-NOMINATIVE", "WO-ACC"
"USATIVE" or the like. The relationship between languages is, for example, Japanese “WO-ACCUSATIVE”.
On the other hand, the relationship "ACC" is registered in English, German and French. When you input a relation in a certain language, the equivalent relation name in a predetermined output language is output. This bilingual dictionary 125
By using, the relationship in each language can be unified to the relationship in a predetermined language.

The language-based normalization unit 126 normalizes the relational expressions extracted by the language-based relation extraction unit 123 into language-independent expressions. As a language-independent expression, it is converted into a predetermined language, for example, a Japanese expression, regardless of the language in which the original text is written. The bilingual word dictionary 124 and the bilingual relation dictionary 125 are used for language normalization. The words constituting the relational expression are translated into a predetermined language using the bilingual word dictionary 124, and the symbols representing the relation are converted into symbols in a predetermined language using the bilingual relation dictionary 125. The language-based normalization unit 126 can obtain words and relationships unified in a certain language.

The registration unit 127 registers in the storage unit 5 together with the document data input from the data input unit 1, using the normalized relational expression independent of the language obtained from the language normalization unit 26 as a key. As a result, the language-independent relational expression is added as a key to the storage unit 5, and the document data is stored.

The search unit 128 searches the document data registered in the storage unit 3 using the expanded relational expression obtained by the relationship expansion unit 7 as a key. The relational expression given from the relational expansion unit 7 is normalized by the language normalization unit 127 into a language-independent expression. Also, the storage unit 5
The key added to the document data stored in is also normalized to a language-independent expression. Therefore, the search unit 128 can perform a search using a language-independent expression.

Although the registration unit 127 and the search unit 128 are explicitly shown in this embodiment, they are naturally provided in each of the above embodiments. In addition, the normalized search request that does not depend on the language and that is output from the language-based normalization unit 126 does not go through the relationship expansion unit 7, but directly.
Although input to the certainty factor evaluation unit 20, it does not matter whether the relation expansion unit 7 is used or not. Also in the above-described second embodiment, the normalized search request may be directly input to the certainty factor evaluation unit 20 without going through the relationship expansion unit 7.

An example of the operation of the third embodiment of the present invention will be described below. In the following description, the operation at the time of registering the relational expression and the operation at the time of searching from the relational expression will be described separately.

FIG. 21 is a flow chart showing an example of the operation at the time of registration of a relational expression in the third embodiment of the relational expression search device of the present invention. When registering a relational expression as a key for searching, the data input unit 1, the expression extracting unit 2,
The language-specific relationship extraction unit 123, the language-specific normalization unit 126, the registration unit 127, and the storage unit 5 operate.

In S131, the document data is read from the data input unit 1 and sent to the analysis switching unit 121 and the registration unit 127. In S132, the analysis switching unit 121 reads the information about the type of language in which the text is described, which is described in the document data, and the analysis unit 122 by language and the relationship extraction unit by language 12 according to the language.
3. Select the language-specific normalization unit 126. In S133, the language-specific analysis unit 122 corresponding to the language selected by the analysis switching unit 121 analyzes the text in the document data. In S134, the language-specific relation extraction unit 123 corresponding to the language selected by the analysis switching unit 121 extracts a relation according to the selected language based on the text analysis result. In S135, the language-based normalization unit 126 corresponding to the language selected in the analysis switching unit 121 uses the bilingual word dictionary 124 and the bilingual relation dictionary 125 to set the relation extracted in S134 to a predetermined language relation. Normalize to. Then, in S136, the registration unit 12
In step 7, the relationship normalized in S135 and the document data read in step S131 are paired and registered in the storage unit 5.

The operation at the time of registration will be described using a specific example. Here, the text in the document data input from the data input unit 1 is described in English, and the normalized language is Japanese. In S131, the document data including the text described in English is read from the data input unit 1 and sent to the analysis switching unit 121 and the registration unit 127. In S132, the analysis switching unit 121 determines that the text in the document data is described in English, and the language-specific analysis unit 122 for English and the language-specific relationship extraction unit 1
23, the language-based normalization unit 126 is selected. Here, these are analyzed by the English analysis unit 122E, the English relation extraction unit 123E,
The English normalization unit 126E is used.

In step S133, the English analysis unit 122E.
Analyzes the text in the document data. Here, it is assumed that syntax analysis is performed as the analysis level. FIG.
FIG. 9 is an explanatory diagram showing an example of an analysis result of an English analysis unit in the third embodiment of the relational expression search device of the present invention. In FIG. 22, "The ABC Ltd. will prod"
uce wine. It is an example of the analysis result of the sentence ".
Here, the analysis result is shown as a tree structure together with attributes.

At S134, the English relationship extraction unit 123
In E, the tree structure of the analysis result as shown in FIG. 22 is collated with the English relational expression extraction rule to extract a plurality of words that may have a relationship with each other and the relationship between them.
FIG. 23 is an explanatory diagram of an example of the English relational expression extraction rule. Here, an example is shown in which two words and the relationship between them are extracted as the relationship expression. In FIG. 23,
The left side of "->", that is, the left side is a pattern of words. Among them, (# 1 # 2 # 3 # 4) represents a tree structure in which # 1 is a parent, and three children # 2, # 3, and # 4 are under it. The part enclosed by "<" and ">" is the designation of the attribute of the node. “Cat” is a category and indicates a syntactic role. “Form” indicates the word itself, and “infl” indicates the inflection. "Pre
“Sp” indicates the present participle (ing form), and “past-p” indicates the past participle. The right side is the relational expression to be extracted,
The first term is the relation name, and the second and third terms are the related words. "# 2.head" indicates the modified noun of the noun phrase,
“# 3.main” indicates the main verb (excluding the auxiliary verb) of the verb phrase. Although not shown in the rules, verbs are basically kept in their original forms and nouns are kept in the singular form in relational expressions.

When the analysis result shown in FIG. 22 and the English relational expression extraction rule shown in FIG. 23 are collated, the first rule and the second rule are applied, and the following relational expressions are extracted. (NOM "ABC Ltd.""product
e ”) (ACC“ WINE ”“ produce ”)

The relational expression extracted by the English relation extracting unit 123E is normalized in S135 by the English normalizing unit 126E so as not to depend on the language. Here, it is assumed that the bilingual word dictionary 124 and the desired translation relation dictionary 125 store the contents shown in FIGS. 19 and 20, respectively. At this time, for example, the relational expression (ACC “W
INE "" produce "), first, using the bilingual word dictionary 124, the word" WIN "forming the relational expression.
Translate E "and" produce "into words in a predetermined language. Here, the output language is Japanese, and if there are multiple Japanese words, the first word in the list is output. Assuming that, from FIG. 19, "WIN
E "and" produce "are" wine "and
Translated as "production". Next, using the bilingual relation dictionary 25, the relation name "ACC" is converted into the Japanese relation name "WO-ACC".
Convert to "Usative". As a result, it is normalized to a language-independent relational expression (WO-ACCUSATIVE "wine""production"). The relational expression thus normalized is associated with the document data containing the original text and registered in the storage unit 5 by the registration unit 127.

Next, the operation when the text of the document data is described in Japanese will be described. The analysis switching unit 121 determines that the text of the document data is described in Japanese, and the analysis unit 122 for each language corresponding to Japanese, the extraction unit 123 for each language, and the normalization unit 12 for each language.
Select 6. Here, these are analyzed by the Japanese analysis unit 122.
J, Japanese relation extraction unit 123J, Japanese normalization unit 126
Let's say J.

The Japanese analysis unit 122J analyzes the text in the document data as in the above-described embodiments. Here, morphological analysis is performed as the analysis level. The Japanese-relationship extraction unit 123J extracts a plurality of words that may have a relationship with each other and the relationship between them by collating the morphological analysis result with the Japanese-relationship expression extraction rule.

FIG. 24 is an explanatory diagram of an example of Japanese relational expression extraction rules in the third embodiment of the relational expression retrieval device of the present invention. This relational expression extraction rule is basically the same as the relational extraction rule shown in FIG. Here, as in the case of English, two words and the relationship between them are extracted as the relationship expression. Match these rules with the results of morphological analysis (independent word 1, adjunct word,
If there is a sequence of independent words 2), the relationship name in the relationship column is given as the relationship between independent word 1 and independent word 2. For example, "ABC
The company produces wine. 24, the result of morphological analysis is collated with FIG. 24, and the following relational expression is extracted. (GA-NOMINATIVE "ABC""manufacturing") (WO-ACCUSATIVE "wine""manufacturing")

Since the relational expression extracted by the Japanese-language relation extracting unit 123J is Japanese, it is not necessary to normalize it into a language-independent form. However, the Japanese-language normalizing unit 126E uses the bilingual word dictionary 124. Normalize synonyms. Thereby, for example, the above-mentioned relational expression (WO-ACCUS
ATIVE "wine""manufacturing") is a relational expression (W
O-ACCUSATIVE "Wine""Production"). The relational expression thus normalized is sent to the registration unit 127, and the normalized relational expression is registered in the storage unit 5 together with the document data as a key of the document data sent from the data input unit 1. This completes the operation of registering the relational expression.

As described above, "The" written in English
ABC Ltd. will product win
e. ", And" ABC written in Japanese
The company produces wine. From the sentence "," a relational expression (WO-ACCUSATIVE
"Wine" and "Production") are generated and registered as keys. In this way, the language-independent relational expression can be registered as a key.

Next, the operation at the time of search will be described. FIG. 25 is a flowchart showing an example of an operation at the time of searching for a relational expression in the third embodiment of the relational expression search device of the present invention. At the time of searching for a relational expression as a key for retrieval, a search request input unit 6, an expression extraction unit 2, a language-specific relation extraction unit 123, a language-specific normalization unit 126, a relation expansion unit 7, a search unit 128, a confidence factor evaluation. The unit 20 and the storage unit 5 operate. Here, it is assumed that the search request is input in a natural language using a certain language.

In S141, the user inputs a search request from the search request input unit 6 in natural language. The search request input by the search request input unit 5 is sent to the expression extraction unit 2. In the expression extraction unit 2, in the same manner as when the relational expression is extracted from the document data, in S142, the analysis switching unit 12
The type of the language in which the search request is described is determined in 1 and S
In 143, the language-specific analysis unit 122 performs analysis according to the language in which the search request is described, and in S144, the language-specific relationship extraction unit 123 extracts a plurality of words that may have a relationship with each other and the relationship between them. In S145,
The language-specific normalization unit 126 normalizes the language-independent expression.

The relational expression obtained by the language-specific normalization unit 126 is sent to the relation expansion unit 7. In S146, the relationship expansion unit 7 expands the relational expression obtained by the language-based normalization unit 126 into a relational expression that can be conceptually the same. A relational expression expansion rule is used for this expansion. The expanded relational expression is sent to the search unit 128. In S147, the search unit 128 compares the expanded relational expression with the relational expression that is the key of the document data registered in the storage unit 5, and determines the matching document data together with the key as the search result. Send to 20.

At S148, the certainty factor evaluation unit 20
The relational expression obtained by the language-based normalization unit 126 is compared with the relational expression matched at the time of search to evaluate the certainty factor. The confidence can be evaluated using, for example, a relation evaluation rule. The certainty factor evaluation unit 20 determines in S149 that
For example, the search results are sorted in descending order based on the certainty factor evaluation value of each search result. The search results sorted by the certainty factor evaluation unit 20 are displayed in the sorted order, for example, on the CRT display 19 of the output unit 8 in S150.

The above-mentioned search operation will be described with reference to a specific example. For example, as a search request, "To product
Consider the case where the expression "e wine" is entered.
The analysis switching unit 121 determines that this search request is described in English, and switches the language-specific analysis unit 122, the language-specific relationship extraction unit 123, and the language-specific normalization unit 126 for English. Similar to the time of registration, the English analysis unit 122E, the English relation extraction unit 123E, and the English normalization unit 126E
A relational expression of (WO-ACCUSIVE "Wine""Production") is extracted.

The relational expression obtained by the English normalization unit 126E is sent to the relation expansion unit 7. In the relationship expansion unit 7,
The relational expression obtained by the language-based normalization unit 126 is developed into a relational expression that can be conceptually identical using a relational expression expansion rule. FIG. 26 shows a third example of the relational expression search device according to the present invention.
5 is an explanatory diagram of an example of a relational expression expansion rule in the embodiment of FIG. This relational expression expansion rule is basically the same as that shown in FIG.
This is a rule similar to the relation expansion rule shown in. This FIG.
The normalized relational expression extracted from the search request according to the relational expression expansion rule shown in (1) is expanded as follows. (WO-ACCUSIVE "Wine""Production") (HA-THEME "Wine""Production") (MO-ADDITION "Wine""Production") (CASE-UNSPEC "Wine""Production") (UNSPEC "Wine""production")

The search unit 128 compares the relational expression developed by the relationship developing unit 7 with the relational expression given as a key to the document data registered in the storage unit 5, and has a matching key. The document data is sent to the certainty factor evaluation unit 20 as a search result together with the key. Here, assuming that the document data described in English and the document data described in Japanese used in the example of the operation at the time of registration are registered in the storage unit 5, these document data are searched. To be done. Then, the relational expression (WO-ACCUSATIVE "Wine") given as a key to these documents.
"Production") together with these documents
Sent to 0.

The certainty factor evaluation unit 20 includes an English normalization unit 126.
The relational expression obtained from E and the relational expression matched at the time of search are compared to evaluate the certainty factor. FIG. 27 is an explanatory diagram of an example of the relation evaluation rule in the third embodiment of the relational expression search device of the present invention. The relationship evaluation rule shown in FIG.
It is basically the same as the relation evaluation rule shown in FIG. Here, since the two retrieved documents both have the relational expression (WO-ACCUSATIVE “wine” “production”), the two documents are matched with the fifth relational evaluation rule in FIG. Is obtained as 100.
The certainty factor evaluation unit 20, along with other search results, sorts the search results, for example, in descending order based on the certainty factor evaluation value, and displays them on the CRT display 19 in the sorted order. At this time, similarly to the second embodiment, the reliability evaluation value can be displayed. In addition,
Various methods as described in the second embodiment can be used to evaluate the certainty factor.

In this way, the search request is not limited to Japanese, but English or other languages can be used for the search. Further, as described above, even in the case of a search request in English, it is possible to search for document data having similar contents in which texts are described not only in English but also in Japanese or other languages. In this way, the search can be performed regardless of the search request and the description language of the text of the document data.

In the third embodiment, the case where a Japanese relational expression is used as a normalized relational expression independent of language and the relational expressions of other languages are unified to the Japanese relational expression is shown. However, the present invention is not limited to this, and may be unified to a relational expression of another language, or a new relational expression may be defined and unified to the relational expression.

Further, at the time of the search, the text is described only in the document data in which the text is described in the same language as the language used in the search request, or the text is described in several designated languages. It is also possible to configure to search only the document data.

In the above-mentioned third embodiment, the certainty factor evaluation unit 2
Although the configuration having 0 is shown, it is also possible to configure without using the certainty factor evaluation unit 20 as in the first embodiment described above. When one language is selected as the language for requesting the search, the configuration and operation at the time of the search are the same as those of the first and second embodiments described above, but the language is stored in the storage unit 5. By registering a normalized relational expression that does not depend on the key as a key, it is possible to search for document data that does not depend on the description language of the text in the document data. Also, the data input unit 1 and the analysis switching unit 1
21, language-specific analysis unit 122, language-specific relationship extraction unit 123,
With the bilingual word dictionary 124, the bilingual relation dictionary 125, the language-based normalization unit 126, and the storage unit 5, it is possible to obtain a device in which the relational expression extraction device illustrated in FIG. 1 is expanded into multiple languages. [Brief description of drawings]

[0175]

As is apparent from the above description, according to the relational expression extracting apparatus of the present invention, it is possible to extract not only a single word but a plurality of expressions and the relationship between them as a key. As a result, concepts and subjects that do not depend on the surface expression are extracted, and more appropriate keywords can be extracted.

Further, according to the relational expression retrieval apparatus of the present invention, it is possible to retrieve not only one word but a plurality of expressions and the relation between them as a key. As a result, it is possible to realize an efficient search from concepts and subjects that do not depend on the surface representation.

Further, by displaying the result based on the certainty factor of the result for the search request, the user can obtain the desired search result more efficiently.

Furthermore, the language in which the search request is made, and
There is an effect that it is possible to search for document data containing desired contents regardless of the description language of the text of the document data.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a relational expression extraction device of the present invention.

FIG. 2 is a flowchart showing an example of the overall operation in an embodiment of the relational expression extraction device of the present invention.

FIG. 3 is a flowchart showing an example of the operation of a word group extraction unit 15 in one embodiment of the relational expression extraction device of the present invention.

FIG. 4 is a flowchart showing another example of the operation of the word group extraction unit 15 in the embodiment of the relational expression extraction device of the present invention.

FIG. 5 is a flowchart showing an example of the operation of the relation extracting unit 3 in the embodiment of the relation expression extracting device of the present invention.

FIG. 6 is a flowchart showing an example of the operation of the normalization unit 4 in the embodiment of the relational expression extraction device of the present invention.

FIG. 7 is an explanatory diagram of an example of a word group extraction rule.

FIG. 8 is an explanatory diagram of an example of a relation extraction rule.

FIG. 9 is a block diagram showing a first embodiment of the relational expression search device of the present invention.

FIG. 10 is a flowchart showing an example of a search process in the first embodiment of the relational expression search device of the present invention.

FIG. 11 is an explanatory diagram of an example of a relation expansion rule in the first embodiment of the relational expression search device of the present invention.

FIG. 12 is an explanatory diagram of an example of a search request input screen using a GUI in the first embodiment of the relational expression search device of the present invention.

FIG. 13 is a block diagram showing a second embodiment of the relational expression search device of the present invention.

FIG. 14 is a flowchart showing an example of search processing in the second embodiment of the relational expression search device of the present invention.

FIG. 15 is an explanatory diagram of an example of a relation expansion rule in the second embodiment of the relational expression search device of the present invention.

FIG. 16 is an explanatory diagram of an example of a relation evaluation rule in the second embodiment of the relational expression search device of the present invention.

FIG. 17 is an explanatory diagram showing an example of a search request input screen using a GUI in the second embodiment of the relational expression search device of the present invention.

FIG. 18 is a block diagram showing a third embodiment of the relational expression search device of the present invention.

FIG. 19 is an explanatory diagram of an example of bilingual word dictionary data.

FIG. 20 is an explanatory diagram of an example of data of a bilingual relation dictionary.

FIG. 21 is a flowchart showing an example of an operation at the time of registering a relational expression in the third embodiment of the relational expression search device of the present invention.

FIG. 22 is an explanatory diagram showing an example of the analysis result of the English analysis unit in the third embodiment of the relational expression search device of the present invention.

FIG. 23 is an explanatory diagram of an example of English-related expression extraction rules.

FIG. 24 is an explanatory diagram of an example of Japanese relational expression extraction rules in the third embodiment of the relational expression search device of the present invention.

FIG. 25 is a flowchart showing an example of an operation at the time of searching for a relational expression in the third embodiment of the relational expression search device of the present invention.

FIG. 26 is an explanatory diagram of an example of a relational expression expansion rule in the third embodiment of the relational expression search device of the present invention.

FIG. 27 is an explanatory diagram of an example of a relation evaluation rule in the third embodiment of the relational expression search device of the present invention.

[Explanation of symbols]

1 ... Data input unit, 2 ... Expression extraction unit, 3 ... Relationship extraction unit,
4 ... Normalization unit, 5 ... Storage unit, 6 ... Search request input unit, 7 ...
Relationship developing unit, 8 ... Output unit, 11 ... Magnetic disk, 12 ...
OCR, 13 ... MT, 14 ... Morphological analysis unit, 15 ... Word group extraction unit, 16 ... Database, 17 ... Keyboard, 1
8 ... Mouse, 19 ... CRT display, 20 ... Certainty factor evaluation part, 121 ... Analysis switching part, 122 ... Language analysis part, 123 ... Language relationship extraction part, 124 ... Bilingual word dictionary, 125 ... Bilingual relationship dictionary, 126 … Language normalization section,
127 ... registration unit, 128 ... search unit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yoshihiro Ueda 134 Kobe-cho, Hodogaya-ku, Yokohama-shi, Kanagawa Yokohama Business Park East Tower Fuji Xerox Co., Ltd. (72) Tadanobu Miyauchi Kobe-cho, Hodogaya-ku, Yokohama, Kanagawa Address 134, Yokohama Business Park East Tower, Fuji Xerox Co., Ltd. (72) Inventor: Juhei Nakagaki, 134, Kobe-cho, Hodogaya-ku, Yokohama-shi, Kanagawa Yokohama Business Park East Tower, Fuji Xerox Co., Ltd.

Claims (11)

[Claims] 1. In a relational expression extraction device for extracting a relational expression that can be a keyword from original data, an expression extraction means for extracting a plurality of expressions that may have a relationship with each other from the original data, and an expression extraction means for extracting the expression. A relational expression extracting device having a relation extracting means for giving a relation to an expression group. 2. The relational expression extracting device according to claim 1, further comprising a normalization unit that normalizes into an expression that does not depend on a surface expression based on the relationship given by the relationship extraction unit. . 3. The expression extracting means includes an analyzing means for analyzing the original data, and a word group extracting means for extracting a plurality of words which may have a relationship with each other from an analysis result by the analyzing means. Item 1. The relational expression extraction device according to Item 1 or 2. 4. The language normalization means for normalizing the relational expression given by the relational extraction means or the normalization means into a language-independent expression. A relational expression extraction device according to item. 5. The expression extraction means comprises language-specific analysis means for analyzing the original data according to the language in which the original data is described, and the relationship extraction means can have a plurality of relationships with each other according to the language. It consists of language-specific relation extracting means for extracting relational expressions consisting of words and their relations, and represents a bilingual word dictionary in which words of equivalent meaning in a plurality of languages are registered and an equivalent linguistic relation in a plurality of languages. A bilingual relation dictionary in which information is registered, and a language normalizing means for normalizing the relational expression extracted by the relation extracting means according to the language into a language-independent expression using the bilingual word dictionary and the bilingual relational dictionary. The relational expression extraction device according to any one of claims 1 to 3, further comprising: 6. A search device for performing a search using a relational expression as a search key, and storage means in which the relational expressions extracted from the original data as a plurality of expressions having a relationship with each other are registered as keys together with the original data, and input. A relational expression retrieving apparatus comprising a retrieval means for retrieving original data registered in the storage section using the relational expression as a retrieval key. 7. The search unit according to claim 6, further comprising at least a relationship expansion unit that expands an input search request into a relational expression that can be conceptually the same and performs a search. Relational expression retrieval device. 8. An expression extracting means for extracting a plurality of expressions that may have a relationship with each other from input data, a relationship extracting means for extracting a relationship between the plurality of expressions extracted by the expression extracting means, and an extracting by the relationship extracting means. Has a normalizing means for normalizing to an expression that does not depend on a surface expression based on the above relationship, and the storage means uses the original data as input data to extract the word group, the relationship extracting means, and the normalizing means. A relational expression extracted and normalized as a plurality of expressions that can be related to each other is registered as a key together with the original data, and the relation expansion means uses the natural language as input data and the word group extraction means and the relational extraction means. , A search request that is extracted and normalized as a plurality of expressions that may have a relationship with each other by the normalizing means is expanded into a relationship expression that can be conceptually the same 8. The method according to claim 7, wherein
The relational expression retrieval device described in. 9. The relational expression extracted from the original data as a plurality of expressions that may have a relationship with each other and normalized to a language-independent expression is registered as a key with the original data in the storage means. The relational expression search device according to claim 6 or 7. 10. An expression extraction unit for extracting a plurality of expressions that may have a relationship from each other from the input data according to a language in which the input data is described, and a relationship between the plurality of expressions extracted by the expression extraction unit. A relationship extraction means for extracting according to, a bilingual word dictionary in which words having equivalent meanings in a plurality of languages are registered, and a bilingual relationship dictionary in which information indicating equivalent linguistic relationships in a plurality of languages is registered,
The storage means includes a language normalization means for normalizing the relational expression extracted according to the language by the relation extracting means according to the language using the bilingual word dictionary and the bilingual relation dictionary into a language-independent expression. In, the relational expression normalized to the language-independent expression by the expression extracting means, the relation extracting means, and the language normalizing means using the original data as input data is registered together with the original data as a key.
The search means performs a search using a search request normalized with natural language as input data by the word group extraction means, the relation extraction means, and the normalization means as a language-independent expression. Item 6. The relational expression search device according to Item 6 or 7. 11. The search means compares a relational expression based on an input search request with a relational expression stored in the storage means to evaluate a certainty factor, and rearranges search results based on the certainty factor. 11. The relational expression search device according to claim 6, further comprising a certainty factor evaluation unit.