JP2001297100A - Device for supporting refining of keyword extraction rule - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device for supporting the refining (strengthening and relieving) of a keyword extraction rule that is used by a keyword extractor. SOLUTION: When the keyword extractor extracts a keyword from an expression having a prescribed relation between a word corresponding to the keyword of a 1st keyword group and a word corresponding to the keyword of a 2nd keyword group, a presentation means presents a rule and an expression adaptive to the rule. When a reception means receives indication from a user not to apply the rule to the expression, a constraint strengthening means changes the rule into an inapplicable one. Meanwhile, a constraint relieving means changes the rule into an applicable one when the presentation means presents an expression and a rule adaptive to only two conditions that are prescribed by the rule and the reception means receives indication from the user to apply the rule to the expression.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a keyword extraction rule refining support apparatus for supporting the refining (enhancement and relaxation) of a keyword extraction rule used in a keyword extraction apparatus for extracting a keyword from a document by, for example, pattern matching of expressions. It relates to a storage medium.

[0002]

2. Description of the Related Art For example, there are the following techniques for generating patterns and keyword extraction rules. As an example, JP-A-7-262189 (hereinafter referred to as Document 1)
The sentence pattern extraction device described in (1) extracts and integrates a sentence pattern from a text. Here, the sentence pattern is a pattern in the order of appearance.

Further, in a method and apparatus for generating a matching pattern described in, for example, Japanese Patent Application Laid-Open No. 9-6796 (hereinafter referred to as Document 2), a user gives a judgment of conformity / non-conformity for a certain purpose, and Automatically extract patterns from a set of documents. Also, in the keyword extraction rule generation method described in Japanese Patent Application Laid-Open No. H10-21249 (hereinafter referred to as Reference 3), a keyword extraction rule used for structuring a document is automatically generated, and a given output format is provided. Automatically generate string conditions to be extracted from definitions.

[0004] Here, the technology described in the above document 1 is as follows.
This is to extract a pattern from a document. It only extracts patterns and does not indicate what they are intended to be. Pattern unification is the merging of multiple rules, not refinement.
Further, the technique described in the above-mentioned document 2 is one in which a user gives a judgment of conformity / non-conformity to a certain viewpoint, and a pattern is automatically generated from a set of conforming documents and a set of non-conforming documents. However, the pattern obtained by this technique is specialized only for a certain viewpoint, there is no guarantee that the pattern is unique only to that viewpoint, and furthermore, no keyword can be extracted.

[0005] The technique described in Document 3 is a technique for generating a keyword extraction rule for structuring a document. The rule depends on the order in which words appear. It is a keyword extraction for each attribute constituting the structure such as a name and a date and time, and does not extract a keyword for each viewpoint such as “problem” or “request”. Also, existing rules cannot be refined for later documents. As described above, there is a technique for automatically generating a pattern or a keyword extraction rule, but there is no technique for supporting correction such that the accuracy is increased as needed.

[0006]

As shown in the above-mentioned conventional example, there is no conventional technique for supporting refinement (enhancement and relaxation) of a keyword extraction rule used in a keyword extraction apparatus. There is a problem that the accuracy of the extraction rule cannot be improved. Specifically, for example, a problem that the keyword extraction device cannot support refining according to the document for the viewpoint-specific and semantic keyword extraction rules. was there.

The present invention has been made in view of the above-mentioned conventional circumstances, and for example, refines (enhances and relaxes) a keyword extraction rule used in a keyword extraction device that extracts a keyword from a document by pattern matching of expression. It is an object of the present invention to provide a keyword extraction rule refining support device and a storage medium that can support (1).

[0008]

In order to achieve the above object, a keyword extraction rule refining support apparatus according to the present invention provides a keyword extraction apparatus for extracting keywords from a document based on a predetermined rule as follows. Help refine the rules. Here, the rule is based on the expression in the document in which a word corresponding to the first keyword included in the first keyword group and a word corresponding to the second keyword included in the second keyword group appear in a predetermined relationship. Specifies that keywords are extracted.

That is, the presenting means presents to the user an expression conforming to the rule and the rule, and accepts from the user an instruction to apply the rule presented to the expression presented by the accepting means, and accepts the instruction. In this case, the rule is changed so that the rule presented to the expression presented by the constraint strengthening means is not applied. Therefore, for example, if the user determines that the presented rule is not applicable to the presented expression, the rule is changed to match such an instruction from the user. This can assist in refining the rule in the direction of strengthening.

In a preferred aspect, in the keyword extraction rule refining support device as described above, the constraint strengthening means includes a keyword corresponding to a word in the presented expression that matches the presented rule, defined by the rule. By deleting from the keyword group corresponding to the word, the presented rule is not applied to the presented expression.

[0011] In another preferred embodiment, in the keyword extraction rule refining support device as described above, the constraint strengthening means converts a keyword corresponding to a word that matches the presented rule in the presented expression into a plurality of subordinate concepts. Converted into keywords and deleted from the keyword group corresponding to the word specified by the rule among the plurality of keywords of these lower concepts, the rule presented for the presented expression Is not applicable.

Further, the keyword extraction rule refining support device according to the present invention supports refining the rule of the keyword extraction device for extracting a keyword from a document based on a predetermined rule as follows. Here, the rule is based on the expression in the document in which a word corresponding to the first keyword included in the first keyword group and a word corresponding to the second keyword included in the second keyword group appear in a predetermined relationship. Specifies that keywords are extracted.

[0013] That is, the presenting means presents to the user an expression that matches only two of the first keyword, the second keyword, and the relation defined by the rule, and the rule,
An instruction to apply the rule presented to the expression presented by the accepting means is received from the user, and when the instruction is accepted, the instruction is applied so that the rule presented to the expression presented by the constraint relaxation means is applied. Change rules. Therefore, for example, when the user determines that the presented rule is to be applied to the presented expression, the rule is changed to match such an instruction from the user, and the rule is relaxed. This can assist in refining the rule in the direction in which it is performed.

[0014] In a preferred embodiment, in the keyword extraction rule refining support device according to the present invention, the constraint relaxation means defines a keyword indicating a concept of a word in the presented expression that does not conform to the presented rule by the rule. By adding it to a keyword group corresponding to the word, the presented rule is applied to the presented expression.

In another preferred embodiment, in the keyword extraction rule refining support device according to the present invention, the constraint relaxation means includes a keyword indicating a concept of a word in the presented expression that does not conform to the presented rule and the rule. By replacing the keyword in the keyword group with a keyword indicating a common superordinate concept with the keyword in the keyword group corresponding to the word specified in, the rule presented to the presented expression is applied. I do.

In another preferred embodiment, in the keyword extraction rule refining support device according to the present invention, the constraint relaxation means converts a word in the presented expression that does not conform to the presented rule into a word specified by the rule. By adding to the keyword group corresponding to the word, the presented rule is applied to the presented expression.

In another preferred aspect, in the keyword extraction rule refining support device according to the present invention, the constraint relaxation means includes a word in the presented expression that does not conform to the presented rule and a word defined by the rule. By replacing the superficial keyword with a keyword indicating a common superordinate concept with the superficial keyword in the keyword group corresponding to the word, the presented rule is applied to the presented expression.

Further, in the keyword extraction rule refining support device according to the present invention, the merging means includes one of the first keyword group, the second keyword group, and the relation defined by the rule.
A plurality of different rules are converted into a single rule obtained by summing the different portions. Therefore, for example, these rules can be replaced with one rule without substantially changing the contents of the rules,
The rules can be arranged to reduce the storage capacity of a memory for storing the rules.

The various processes according to the present invention as described above can also be realized by, for example, reading and executing a program stored in a storage medium by a computer. As an example, the storage medium according to the present invention stores a program to be executed by a computer so that the input means of the computer can read the program.
A keyword is extracted from a document based on a rule for extracting a predetermined keyword from an expression in which a word corresponding to a first keyword included in a keyword group and a word corresponding to a second keyword included in a second keyword group appear with a predetermined relationship. Is used to assist in refining the rules of the keyword extraction device that extracts the rules. The program includes an expression that matches the rules, a process of presenting the rules to the user, and a rule that is presented for the presented expressions. The computer causes the computer to execute a process of receiving an instruction to apply the rule from the user and a process of changing the rule so that the rule presented for the expression presented when the instruction is received is not applied.

Similarly, the storage medium according to the present invention comprises:
A program to be executed by the computer is stored readable by the input means of the computer, and a word corresponding to the first keyword included in the first keyword group and a word corresponding to the second keyword included in the second keyword group are stored. Is used to assist in refining the rules of a keyword extraction device that extracts keywords from a document based on rules that extract predetermined keywords from expressions that appear with a predetermined relationship. A process of presenting to the user an expression that matches only two of the first keyword, the second keyword, and the relationship to be performed, and a process of receiving from the user an instruction to apply the rule presented to the presented expression And the rule presented to the expression presented when the instruction is received is applied And a process of changing the urchin the rule is executed on the computer.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment according to the present invention will be described with reference to the drawings. First, an example of a keyword extraction device to which the present invention is applied will be described. In the keyword extraction device, for example, a keyword indicating a characteristic of the document is automatically extracted from a document including text using a keyword extraction rule and a thesaurus prepared in advance. The refinement of such a keyword extraction rule is supported.
Here, the keyword extraction rule is based on an expression pattern, and can use a dependency relationship or the like. In the following, a case where the dependency relationship is used will be described as an example.

FIG. 11 shows an example of the configuration of a keyword extraction device in the case of using a dependency relationship. In this keyword extraction device, the document group holding device 1 stores a document group for which a keyword extraction rule is to be generated. The dependency analysis device 2 performs the dependency analysis process on the document held by the document group holding device 1, and the pattern matching device 3 holds the result of the dependency analysis by the keyword extraction rule holding device 4. Pattern matching with the keyword extraction rules. The keyword extraction rule holding device 4 holds keyword extraction rules prepared in advance, and the thesaurus 5 holds the hierarchical structure of words as data.

Next, a keyword extraction rule refinement support apparatus according to a first embodiment of the present invention will be described. In this example, a device (keyword extraction rule constraint strengthening support device) that supports strengthening of rule constraints is shown. FIG. 1 shows an example of a keyword extraction device incorporating the keyword extraction rule refining support device according to the present embodiment. This device has, for example, a document group holding function similar to that shown in FIG. The apparatus 1 includes a device 1, a dependency analysis device 2, a pattern matching device 3, a keyword extraction rule holding device 4, a thesaurus 5, and a keyword extraction rule constraint strengthening device 6, which is a characteristic component of the present embodiment. Components 1 to 5 having the same functions as those shown in FIG. 11 are denoted by the same reference numerals as those shown in FIG.

The document group holding device 1 has a function of receiving, for example, a document group to be subjected to rule refining and holding it. In this example, the document group consists of multiple documents,
Each document contains text. The text is composed of one or more sentences. Hereinafter, a case will be described as an example in which documents 1 to 4 including the following sentences 1 to 4 are applied to extract a keyword indicating “problem” by an extraction rule. In this case, it is assumed that the element position of each morpheme of each text is as shown in FIG.

Sentence 1: Horizontal resolution is not possible. Sentence 2: Horizontal resolution is possible. Statement 3: There is no gap at the glued portion of the binder. Statement 4: Cannot stack output.

The dependency analysis device 2 has a function of performing a dependency analysis process on a document held by, for example, the document group holding device 1 and holding all obtained dependency relationship data. I have. As the technique of dependency analysis, for example, various techniques may be used as long as it is possible to find an expression that conforms to the keyword extraction rule. For example, “Hirogo Nomura: Nature Basic technology of language processing, IEICE, Corona, 19
88. "can be used.

The elements on the participant side and the receiving side are expressed by morpheme appearance positions in one sentence, and are held in the following pairs. (fp, r, bp) where fp and bp indicates the appearance position in the document relating to the dependent party and the recipient, respectively, and r indicates the relationship between the concerned party and the recipient.

In the case of the sentence 1, the dependency relation obtained by analyzing the dependency relation is as follows. Note that “(No)” indicates that the expression of the attached word (here, “none”) following the morpheme (here, “can”) has a negative meaning. (1, of 3) ・ ・ Relation 1 (3, is 5 (No)) ・ ・ Relation 2

In the case of sentence 2, the dependency relation obtained by analyzing the dependency relation is as follows. (1, of 3) ... Relation 3 (3, 5) ... Relation 4

In the case of sentence 3, the dependency relation obtained by analyzing the dependency relation is as follows. Note that "absence" indicates that there is no word indicating the relationship between two morphemes (here, "paste" and "part"). (1, 3) ・ ・ Relation 5 (3, (No), 4) ・ ・ Relation 6 (4, 2, 8 (No)) ・ ・ Relation 7 (6, 8 (No)) ・ ・Relationship 8

In the case of sentence 4, the dependency relation obtained by analyzing the dependency relation is as follows. (1,, 3) ... Relation 9 (3,, 5 (No)) ... Relation 10 In the same manner as above, such analysis processing is performed on other documents in each document group. Apply.

The keyword extraction rule holding device 4 has, for example, a function of holding a keyword extraction rule prepared in advance and overwriting an extraction rule obtained from the keyword extraction rule constraint strengthening device 6 with an existing keyword extraction rule. . One keyword extraction rule Er is expressed in a format of Er: (F, R, B) → A, and the keyword extraction rule holding device 4 holds this set.

Here, F, R, B, and A are a set, and details thereof will be described below. F is a set of expressions related to each other, and each element is represented by [c] or “s”. R is a set of relations, and each element is represented by r. If no relationship is specified, it is described as "(none)". B is a set of expressions of the receiving word, and each element is represented by [c] or "s". A is the content of the keyword extraction operation, and each element indicates the position of the word to be extracted. Specifically, it is f when extracting a dependent word, and b when extracting a received word.
It is. Further, c described above indicates a concept name included in the thesaurus, and s described above indicates a surface layer.

The keyword extraction rule described in the present embodiment corresponds to the rule described in the present invention.
And R correspond to the first keyword group and the second keyword group according to the present invention, respectively, and a predetermined relationship. Each element of F and each element of B according to the present embodiment correspond to the first keyword and the second keyword according to the present invention, respectively. The keyword extracted by A in the present embodiment corresponds to two keywords, and corresponds to the predetermined keyword in the present invention.

The thesaurus 5 has, for example, a function of expressing a hierarchical structure of words. This structure has, for example, a structure in which a node of a concept tree structure is further accompanied by a set of words belonging to the concept. I have. This embodiment shows a case where the thesaurus T ₁ shown in FIG. 3 is used.

The pattern matching device 3 uses, for example, a keyword extraction rule held by the keyword extraction rule holding device 4 to sequentially change all the dependency relationships obtained from the dependency analysis device 2 with the keyword extraction rule. It has a function to perform pattern matching. When the pattern matches (that is, when a dependency expression that matches the keyword extraction rule is found), the pattern matching device 3 uses the dependency at that time as a match expression and applies the keyword extraction rule. And has the function of outputting the same.

[0037] Here, consider an example in which the keyword extraction rule holding unit 4 described above holds the next keyword extraction rule Er _1. Er ₁ : ({[print quality], [service]}, {but, is}, {“
(No) "}) → {f} For example, the relation 2 (" cannot be resolved ") in sentence 1 described above is an expression indicating a problem. Applying Er ₁ to this,
It relates language than thesaurus T ₁ "resolution" belongs to the concept "print quality", the relationship is "is", because the received word is "can (not)", matches Er _1, relates word "solution The image is extracted as a keyword of the concept “print quality”.

Also, for example, the relation 4 (“can be resolved”) of sentence 2 described above is not an expression indicating a problem. On the other hand, when Er ₁ is applied, the relation term “resolution” is obtained from the thesaurus T ₁
Belongs to the concept "print quality" is, but the relationship is "is", because the received word is "possible", does not match the Er _1, it does not extract the "resolution" as a keyword. Further, for example, the relation 8 (“cannot make a gap”) in sentence 3 described above is not an expression indicating a problem. However, when Er ₁ is applied to this, from the thesaurus T ₁ , the suffix “skin” belongs to the concept “print quality”, the relationship is “ga”, and the received word is “can (no)” , Er _1, and extract the word “spacing” as a keyword of the concept “print quality”.

Further, for example, the relation 10 of the above sentence 4
("Stacking not possible") is an expression indicating the problem. Applying Er ₁ contrast, the relationship is "ga", it receives word is "can (not)", since the dependency words from the thesaurus T ₁ "stack" does not belong to the concept "print quality" , Er ₁
And does not extract "stack" as a keyword.

The above-described keyword extraction rule Er ₁
As an example, [print quality] and [service] correspond to the first keyword of the present invention, and the "resolution"
And “spacing” correspond to the word corresponding to the first keyword according to the present invention, and similarly, “do (no)” corresponds to the second keyword according to the present invention, and “do (no)” corresponding thereto "
Corresponds to the word corresponding to the second keyword in the present invention.

As described above, the relation 8 of the above sentence 3
( "Clearance can not") Despite not express indicate a problem, since the result matches the keyword extraction rule Er _1, the "gap" concept "print quality" (in the example of FIG. 3, further " It is extracted as a keyword under “Printing error”. In order to prevent this matching from occurring, it is necessary to strengthen the constraint of Er ₁ applied.

Therefore, in this embodiment, for example, a keyword extraction rule constraint having a function of receiving an input from the pattern matching device 3 and strengthening the constraint of the keyword extraction rule to overwrite the corresponding portion of the keyword extraction rule holding device 4 A reinforcement device 6 is provided. FIG. 4 shows an example of the configuration of the internal operation of the keyword extraction rule constraint strengthening device 6. The keyword extraction rule constraint strengthening device 6 is a constraint strengthening instruction device 11 that constitutes a presentation unit and a reception unit according to the present invention. And a constraint strengthening refining device 12 and a constraint strengthening rule candidate presenting device 13 constituting the constraint strengthening means according to the present invention.
It is composed of

The constraint strengthening instruction device 11 instructs the user whether or not to strengthen the constraint of the keyword extraction rule,
It has a function of receiving the instruction. For example, as shown in FIG. 5, the constraint strengthening instruction device 11 displays the contents of the keyword extraction rule, the expression that matches the rule, and the extraction result of the keyword side by side. In the figure, the keyword “skin” extracted when applying the rule Er ₁ to the relation 8 of the sentence 3 is highlighted.

[0044] Such a display screen to see the user, to determine whether to carry out the constraints strengthening of the rule Er _1, give an indication of the Yes / No. Then, when the constraint strengthening instruction device 11 receives the instruction of Yes, the constraint strengthening refinement device 12 and the constraint strengthening rule candidate presenting device 13 described later perform the constraint strengthening process, and the constraint strengthening instruction device 11 issues the No instruction. If it is received, such a constraint strengthening process is not performed, and the process is terminated as it is, and the process proceeds to, for example, the process by the constraint strengthening instruction device 11 relating to the next keyword.

As described above, when the pattern matching device 3 matches a certain keyword extraction rule with a certain expression and extracts a keyword, the constraint strengthening instruction device 11 obtains a keyword extraction rule obtained from the device 3. , Expressions, and keywords are presented to the user, and the result of the user is determined based on whether the user should actually extract the keywords. When it is instructed that the keyword should not be extracted, the keyword extraction rules, expressions, and keywords are output to the constraint strengthening / purifying device 12. The process ends. Note that the instruction that should not be extracted here corresponds to the instruction that the rule presented in the present invention is not applied.

When the constraint strengthening refining device 12 receives an instruction of Yes from the constraint strengthening instructing device 11,
It has a function of performing refining to strengthen the restrictions on the keyword extraction rules. The restriction-enhancing purification device 12 of the present example
Two different methods (Reinforcement Method 1 and Reinforcement Method 2, respectively)
Has the function of strengthening the restrictions on the keyword extraction rules.

First, enhancement method 1 will be described. That is,
Keyword extraction rule Er: For (F, R, B) → A, for example, if the word that should not be extracted is a dependency, and the concept of the dependency is fc, keyword extraction after strengthening the constraints The rule Era is Era: (F− {fc}, R, B) → A.

Similarly, for example, if the word to be not extracted is a receiver word, and the concept of the receiver word is bc, the keyword extraction rule Erb after the constraint reinforcement is Erb: (F, R, B − {Bc}) → A.

More specifically, for example, the keyword extraction rule Er ₁ held by the keyword extraction rule holding device 4 is matched with the above sentence 3 by matching with the relation 8 of the sentence 3 (“cannot make a gap”). Is performed, the relation 8 matches Er ₁ and the relation word “skin” is extracted as a keyword of the concept “print quality”. However, since the expression "cannot make a gap" is not an expression that actually indicates a problem, the user wants to strengthen Er ₁ so as not to perform the extraction, and gives an instruction to that effect. In this case, since the word that should not be extracted is a related word, the constraint is strengthened so that Er is changed to Era. That is, the keyword extraction rule Era ₁ after the constraint strengthening obtained from Er ₁ is Era ₁ : ({[service]}, {but, is}, {“can (no)”}) →
{f}.

The above-mentioned keyword extraction rule Er₁
The above keyword extraction rule Era ₁To change to
As an example, the “spacing” described above refers to the present invention.
Word that matches the rule presented in the expression
"Print quality" corresponds to the keyword corresponding to the word
And the above-mentioned keyword extraction rule Er₁Section specified in
The keyword group on the word side is the
Equivalent to the keyword group corresponding to the word specified in the
And the keyword “print quality” from the keyword group
Has been removed. Here, in the present example,
However, the same applies to the receiving side.

As described above, in the strengthening method 1, the constraint strengthening refining device 12 receives the keyword extraction rule, the expression, and the keyword from the constraint strengthening instruction device 11, and receives the keyword portion (as in this example, the dependency portion). In such a case, the constraint of the rule is strengthened by a method of reducing the concept of the linguistic or receiving word) from the keyword extraction rule.

Next, enhancement method 2 will be described. That is,
The keyword extraction rule Er, when word you want to avoid such as extraction is dependency word, fcu ₁ of m subgeneric concepts fc of the dependency word, ..., and fcu _m, these subgeneric In fcu _i (1 ≦ i
≤ m), the keyword extraction rule after strengthening Er constraints
Erc is Erc: (F− {fc} + {fcu ₁ ,…, fcu _i−1 , fcu _{i + 1} ,…, fc
u _m }, R, B) → A.

[0053] Similarly, for example, when word you do not want to extract is subjected word, bcu ₁ n pieces of subordinate concepts of the receiving language concepts bc, ..., and bcu _n, of these subgeneric Let bcu _j (1 ≦ j ≦ n) be the concept including the received word in Er, the keyword extraction rule Erd after strengthening the constraint of Er is Erd: (F, R, B− {bc} + {bcu ₁ , …, Bcu _j-1 , bcu _{j + 1} ,…,
bcu _n }) → A.

[0054] As with the reinforcing method 1 described above, for example, it holds the keyword extraction rule holding unit 4 is a keyword extraction rule Er _1, can not be related 8 ( "gap of the statement 3 by matching between sentences 3 described above )), The word that should not be extracted is the linguistic term “spacing”.
Enhance the constraint to change to Specifically, the subordinate concepts of the concept “print quality” are “print state” and “print error”, and the “clearance” belongs to the “print error”, so the concept “print error” is deleted. That is, Er ₁
Keyword extraction rule Erc after strengthening constraints obtained from
₁ is Erc ₁ : ({[printing status], [service]}, {but, is}, {“
(No) ”}) → {f}.

The above-described keyword extraction rule Er₁
The above keyword extraction rule Erc ₁To change to
For example, the “print status” and “print error”
Corresponds to a plurality of keywords of the subordinate concept referred to in the present invention,
"Print error" in multiple keywords of these subordinate concepts
Among the keywords of the subordinate concept referred to in the present invention
Keyword that matches the word).
The original keyword "print quality" corresponding to the word is
Replaced with "printing state" which does not correspond to the word
Have been. Here, in this example, the example of the hang-up side was shown.
However, the same applies to the receiving side.

As described above, in the strengthening method 2, the constraint strengthening refining device 12 receives the keyword extraction rule, the expression, and the keyword from the constraint strengthening instructing device 11, and receives the keyword portion (as in the present example, the dependency portion). In this case, the concept of the related word or the received word) is expanded into lower-level concepts using a thesaurus, and then the constraints of the rules are strengthened by a method of reducing only the lower-level concepts to which the part belongs from the keyword extraction rules.

The constraint-strengthening rule candidate presentation device 13 generates a keyword extraction rule candidate after constraint strengthening refined by the above-described constraint-strengthening refinement device 12 (in this example, the rule generated by the reinforcement method 1 and the keyword extraction rule by the reinforcement method 2). (A generated rule) is presented to the user, and when the user selects one of the presented candidates, the user has a function of performing an overwrite process in accordance with the selection instruction. That is, in the constraint strengthening rule candidate presentation device 13, the rule selected by the user is determined as the keyword extraction rule after the constraint strengthening, and the rule is overwritten with the rule before the constraint strengthening in the keyword extracting rule holding device 4.

[0058] Similar to the above, for example, the keyword extraction rule Er ₁ as an example, a candidate in this case becomes Erc ₁ were Era ₁ and the mentioned above, for example, something like what is shown in FIG. 6 is displayed to the user You. In the example of the figure, highlighting the changes to the original rule Er ₁ of each candidate. For example, when the user selects “1” in the figure, Era ₁ is overwritten on Er ₁ in the keyword extraction rule holding device 4 and, for example, the user selects “original rule” in the figure. In this case, Er ₁ remains in the keyword extraction rule holding device 4 without performing the overwriting process.

As described above, in the keyword extraction rule refining support device of the present embodiment, for example, upon receiving the result of the matching performed by the pattern matching device 3 of the keyword extraction device, the constraint of the keyword extraction rule is strengthened in accordance with the judgment and selection of the user. Thus, for example, the refining of the existing keyword extraction rule can be realized by repeatedly performing the process of strengthening the constraint of the existing keyword extraction rule according to the document.

Next, a keyword extraction rule refining support apparatus according to a second embodiment of the present invention will be described. Note that, in this example, a device (keyword extraction rule constraint relaxation support device) that supports relaxation of rule constraints is shown. FIG. 7 shows an example of a keyword extraction device incorporating the keyword extraction rule refining support device according to the present embodiment. This device has, for example, the same function as that shown in FIG. 1 of the first embodiment. Group holding device 1, dependency analysis device 2, pattern matching device 3, keyword extraction rule holding device 4, thesaurus 5, and similar rule selection device 21 and keyword extraction rule constraint which are characteristic components of this example. A mitigation device 22 is provided.

The components 1 to 5 having the same functions as those shown in FIG. 1 are denoted by the same reference numerals as those shown in FIG. Further, for the sake of convenience of explanation, the description of the same configuration as that shown in the first embodiment will be omitted in this example.

[0062] In this example, a case where the keyword extracting rule holding unit 4 holds the keyword extraction rule Er ₂ as follows as an example. Er ₂ : ({"collation, [service]}, {but, is}, {"
(Not) "}) → {f} In this case, when trying to match the relationship 10 sentence 4 a pattern matching device 3, for example above (the" can not stack ") and a keyword extraction rule Er _2, the relationship" but "and received word about the" can not "is to match the corresponding portion of the Er _2, but because relates word for the" stack "does not match the corresponding part of the Er _2, keyword is not extracted.

If none of the other keyword extraction rules held by the keyword extraction rule holding device 4 match the above-mentioned relation 10, no keyword is extracted from the relation 10. However, "stacking is not possible" is an expression indicating a problem, and "stacking" should be extracted as a keyword.

Therefore, in the present embodiment, a similar rule selection device 21 and a keyword extraction rule constraint relaxation device 22 are provided as components for realizing such keyword extraction. For example, when all of the keyword extraction rules held by the keyword extraction rule holding device 4 do not match a certain expression, the similar rule selection device 21 converts the keyword extraction rule close to the pattern of the expression into a keyword extraction rule holding device. 4 has a function of selecting from among the keyword extraction rules held by the device.

When the dependency relation is used as in this example, the keyword extraction rule selected here is one of the following Uma, Umb, and Umc. Uma is a keyword extraction rule in which the terms do not match and the relation and the receiving word match. Umb is a keyword extraction rule in which the relationship does not match and the dependency and the receiving word match. Umc is a keyword extraction rule in which the receiving word does not match and the dependency word and the relationship match.
Specifically, in the case of Er ₂ described above and the relationship 10 described above, the dependency “stack” does not match the dependency portion of Er ₂ , but the relationship “ga” and the receiving word “cannot” match, respectively. , Corresponds to the above Uma.

As described above, the similar rule selecting device 21
For example, if there is no keyword extraction rule that is matched by the pattern matching device 3, one place between the expression and the keyword extraction rule (in the case of a dependency relation as in this example, the relation word, relation, ), All other matches (that is, the other two locations match)
Such a keyword extraction rule is selected from among the keyword extraction rules held in the keyword extraction rule holding device 4, and the selected keyword extraction rule and the expression and the keyword that is not extracted because it did not match are extracted from the keyword extraction rule constraint relaxation device. 22.

Keyword extraction rule constraint relaxation device 22
Has a function of, for example, receiving the result from the similar pattern selection device 21 and relaxing the restriction on the keyword extraction rule to overwrite the corresponding part of the keyword extraction rule holding device 4. FIG. 8 shows a configuration example of the internal operation of the keyword extraction rule constraint relaxation device 22. The keyword extraction rule constraint relaxation device 22 constitutes a presentation means according to the present invention together with, for example, the similar rule selection device 21 described above. In addition, it is composed of a constraint relaxation instructing device 31 constituting a receiving means according to the present invention, a constraint relaxation refining device 32 and a constraint relaxation rule candidate presenting device 33 constituting a constraint relaxing means according to the present invention.

The constraint relaxation instructing device 31 instructs the user whether or not to relax the constraint of the keyword extraction rule,
It has a function of receiving the instruction. For example, as shown in FIG. 9, the constraint relaxation instructing device 31 outputs the expression and the extraction result that did not match the keyword extraction rule because the content of the keyword extraction rule and, for example, either the lingo word or the received word did not match. Are displayed side by side.

[0069] Such a display screen as viewed user determines whether to perform the constraint relaxation this rule Er ₂ Indeed, Yes
/ No instruction is given. Then, the constraint relaxation instructing device 31
When the instruction of es is received, the constraint relaxation purification device 3 described later
2 and the constraint relaxation rule candidate presentation device 33 perform the constraint relaxation process, and when the constraint relaxation instructing device 31 receives an instruction of No, such a constraint relaxation process is not performed, and the process is terminated as it is. For example, the processing shifts to the processing by the similar rule selection device 21 and the constraint relaxation instructing device 31 for the next keyword.

As described above, the constraint relaxation instructing device 31 presents to the user the document obtained by the similar rule selecting device 21, the keyword extraction rule, the expression, and the word of the unmatched portion, and actually extracts this word. Let the user decide whether to do so and output the result. Then, when it is instructed to be extracted, the rule, expression, and keyword are output to the constraint relaxation refining device 32. When it is instructed not to be extracted, a series of processes related to extraction of this keyword is ended. I do. The instruction to be extracted here corresponds to the instruction to apply the presented rule according to the present invention.

When the constraint relaxation refining device 32 receives an instruction of Yes from the constraint relaxation instructing device 31,
It has a function of performing refining to relax the restrictions on the keyword extraction rules. The constraint relaxation purifying device 32 of this example has 4
It has a function to relax the restrictions on the keyword extraction rules by two different methods (referred to as relaxation method 1, relaxation method 2, relaxation method 3, and relaxation method 4, respectively).

First, the relaxation method 1 will be described. That is,
Taking the case of Uma described above as an example, for example, if a dependency exists in the thesaurus, the constraint can be relaxed on the keyword extraction rule Er. In this case, the keyword extraction rule Ere after the constraint is relaxed is If the concept of the word is fc, then Ere: (F + {fc}, R, B) → A.

Similarly, taking the case of Umc as an example, if the received word exists in the thesaurus, the constraint can be relaxed. In this case, the keyword extraction rule Erf after the constraint is relaxed is the received word. If the concept of is bc, then Erf: (F, R, B + {bc}) → A.

Specifically, for example, the above-mentioned Er_TwoWas mentioned above
When relation 10 is detected as a constraint relaxation target
Corresponds to the case of Uma described above, and
-Las T ₁The concept of the stake "stack" because it exists in
Keyword after the relaxation of the following constraints with "Sort" added
E extraction rule_TwoIs obtained. Ere_Two: ({"Collation", [service], [sort]}, {but,
Is}, {“can (no)”}) → {f}

The above-described keyword extraction rule Er_Two
The above keyword extraction rule Ere _TwoTo change to
As an example, the "stack" described above is the presentation
Equivalent to a word in the expression that did not conform to the rules presented
Then, the above [sort] describes the concept of the word in the present invention.
Keyword, and the keyword [sort]
Keyword extraction rules have been added. Here, this example
In the above, an example in which the stakeholders do not match was shown.
The same applies when the words do not match.

As described above, in the relaxation method 1, the constraint relaxation refining device 32 receives the keyword extraction rule, the expression, and the keyword from the constraint relaxation instructing device 31, and receives the keyword portion (as in this example, In this case, the concept of the rule is relaxed by a method of adding the concept of the linguistic word or the receiving word) to the corresponding part of the keyword extraction rule.

Next, relaxation method 2 will be described. That is,
Taking the case of Uma as an example, if, for example, a dependency is present in the thesaurus and there is a common superordinate concept between the concept fc of the dependency and a certain concept fec of the dependency part of the keyword extraction rule Er, , The keyword extraction rule Erg after the constraint relaxation in this case for all the common superordinate concepts fcc
Becomes Erg: (F− {fec} + {fcc}, R, B) → A.

Similarly, taking the case of Umc as an example, for example, a receiving word exists in a thesaurus, and a common superordinate concept of the concept bc of the receiving word and a certain concept bec of the receiving word portion of Er If there is, the keyword extraction rule Er after the constraint is relaxed in this case for all the common superordinate concepts bcc
h becomes Erh: (F, R, B− {bec} + {bcc}) → A.

More specifically, for example, when the above-described keyword extraction rule Er ₂ is detected as a constraint mitigation object with respect to the above-described relationship 10, it corresponds to the above-mentioned Uma,
And it relates words present in the thesaurus T _1. Then, the concept of having a common higher-level concepts and concept "sort" of relates word "stack", "service" is present in the relates word part of the Er _2, their common higher-level concepts and two of the "symptoms" and "problem" Become. Accordingly, the following two keyword extraction rules Erg ₂ and Erg ₂ ′ after the restriction is relaxed are obtained. Erg ₂ : ({"collation", [symptom]}, {but, is}, {"
(No) ”)) → {f} Erg ₂ ': ({“ collation ”, [problem]}, {but, is}, {“
(No) ”}) → {f}

The above-described keyword extraction rule Er_Two
The above keyword extraction rule Erg _Two, Erg_TwoChange to '
As an example, the “service” described above is applied to the present invention.
(Proposed)
Keywords in keyword group corresponding to (matched) words
The above [symptoms] and the above [problems] correspond to the book.
Say to the invention (keyword indicating the concept of the word and the key
Indicates a common superordinate concept (with the keyword in the word group)
The keyword corresponds to the keyword, and the keyword in the keyword group
The keyword that the code "service" indicates the common high-level concept
Are replaced by [symptoms] and [problems]. here
In this example, an example was shown in which the terms did not match.
However, the same applies to the case where the received words do not match.

As described above, in the mitigation method 2, the constraint mitigation refinement device 32 performs constraint mitigation by folding the superordinate concept. Specifically, the constraint mitigation instruction device 31 inputs the keyword extraction rules, expressions, and keywords. Then, the constraint of the rule is replaced by a method that replaces the concept of the keyword part (in the case of a dependency as in the present example, the dependent word or the received word) with a common superordinate concept of the concept of the corresponding part of the keyword extraction rule. ease.

Next, relaxation method 3 will be described. That is,
Taking the case of Uma described above as an example, for example, if the stakeholder is fs, the keyword extraction rule after the constraint is relaxed in this case
Eri becomes Eri: (F + {“fs”}, R, B) → A.

Similarly, taking the case of the above Umb as an example, for example, assuming that the relationship is r, the keyword extraction rule Erj after the constraint is relaxed in this case is Erj: (F, R + {r}, B) → A Becomes Further, taking the case of Umc as an example, for example, if the receiving word is bs, the keyword extraction rule Erk after constraint relaxation in this case is Erk: (F, R, B + {“bs”}) → A Become.

[0084] Specifically, if the keyword extraction rule Er ₂ for example described above were detected as constraint relaxation object with respect to the relationship 10 described above, the case for Uma described above,
The following keyword extraction rule Eri ₂ after constraint relaxation is obtained. Eri ₂ : ({“collate”, [service], “stack”},
{But, is}, {“can (no)”}) → {f} Here, this example shows the case where the linguistic terms do not match, but the same applies to the case where the receiving terms do not match .

As described above, in the relaxation method 3, the constraint relaxation refining device 32 receives the keyword extraction rule, the expression, and the keyword from the constraint relaxation instructing device 31, and receives the keyword portion (as in this example, In such a case, the constraint of the rule is relaxed by a method of adding the surface layer of the related word or the received word) to the corresponding part of the keyword extraction rule.

Next, relaxation method 4 will be described. That is,
Taking the case of the above Uma as an example, for example, a certain surface layer fe in the dependency part of the dependency fs and the keyword extraction rule Er
are present in the thesaurus, and if there is a common (superordinate) concept thereof, the keyword extraction rule Erl after the constraint relaxation for all the common superordinate concepts fcc is Erl: (F− {“ fes ”} + {[fcc]}, R, B) → A.

Similarly, taking the case of Umc as an example, for example, both the receiving word bs and a certain surface layer bes in the receiving word portion of the keyword extraction rule Er exist in the thesaurus, and their common (higher) If there is a concept, for all the common superordinate concepts bcc, the keyword extraction rule Erm after constraint relaxation is Erm: (F, R, B − {“bes”} + {[bcc]}) → A Become.

More specifically, for example, when the above-described keyword extraction rule Er ₂ is detected as a constraint relaxation target with respect to the above-described relationship 10, it corresponds to the above-mentioned Uma,
Both relates words and "collating" the surface layer of the relates word part of the "stack" and Er ₂ is present in the thesaurus T _1, their common (upper) concept "sort", "Operation", "symptom", " because there is a problem ", keyword extraction rule Erm ₂ after four constraint relaxation such as the _{following, Erm 2 ', Erm 2'} ', Erm 2''' can be obtained. Erm ₂ : ({[sort], [service]}, {but, is}, {“
(No) ”)) → {f} Erm ₂ ': ({[action], [service]}, {but, is}, {“
(No) ”)) → {f} Erm ₂ ″: ({[symptoms], [service]}, {but, is}, {“
(No) ”)) → {f} Erm ₂ ''': ({[problem], [service]}, {but, is}, {“
(No) ”}) → {f}

The above-mentioned keyword extraction rule Er_Two
The above keyword extraction rule Erm _Two, Erm_Two', Er
m_Two'', Erm_TwoIf you change to '' 'as an example,
"Collation" is defined by the rules stated (presented) in the present invention.
Key that corresponds to the word that did not comply with the rule
Corresponds to the surface keyword in the word group,
The word "collage" is common with this word "stack".
Keywords such as "sort" and "action"
It has been replaced by "symptoms" and "problems." Where the book
In the example, the case where the stakeholders do not match is shown.
The same applies to the case where the words do not match.

As described above, in the relaxation method 4, the constraint relaxation refining device 32 performs constraint relaxation by conversion from the surface layer to the concept. More specifically, the constraint relaxation instructing device 31 transmits the keyword extraction rules, expressions, and keywords. It receives the input and replaces it with the common concept of the surface of the keyword part (in the case of the dependency, in the case of the dependency, the received word) and the surface of the corresponding part of the keyword extraction rule, the restriction of the rule is applied. To relax.

The constraint mitigation rule candidate presentation device 33 is, for example, a keyword extraction rule candidate after the constraint mitigation refined by the constraint mitigation refinement device 32 (in this example, the rule generated by the mitigation method 1 and the relaxation method 2). And the rule generated by the relaxation method 3 and the rule generated by the relaxation method 4) to the user, and when the user selects one of the presented candidates, It has a function to perform overwrite processing according to instructions. That is, in the restriction mitigation rule candidate presentation device 33, the rule selected by the user is determined as the keyword extraction rule after the restriction mitigation, and the rule is overwritten with the rule in the keyword extraction rule holding device 4 before the restriction mitigation.

Similarly to the above, taking the keyword extraction rule Er ₂ as an example, the candidates in this case are Ere ₂ , E
rg ₂ , Erg ₂ ′, Eri ₂ , Erm ₂ , Erm ₂ ′, Erm ₂ ″, Erm ₂ ′ ″, for example, the contents shown in FIG. 10 are displayed to the user. For example, if the user selects “1” in the figure, Ere ₂ in the keyword extraction rule holding device 4
₂ is overwritten, and for example, when the user selects the “original rule” in FIG.
Er ₂ remains in the keyword extraction rule holding device 4 as it is.

As described above, in the keyword extraction rule refining support device of the present example, upon receiving the result of the matching performed by the pattern matching device 3 of the keyword extraction device, for example, the restrictions on the keyword extraction rule are relaxed according to the user's judgment and selection. Thus, for example, the refining of the existing keyword extraction rule can be realized by repeatedly performing the constraint relaxation processing of the existing keyword extraction rule according to the document.

Next, a keyword extraction rule refinement support device according to a third embodiment of the present invention will be described. The configuration of the keyword extraction rule refining support device of this example is almost the same as that shown in FIG. 1 of the first embodiment and FIG. 7 of the second embodiment, for example. A merging function (not shown) constituting the merging means is provided, and this merging function has a function of merging (unifying) keyword extraction rules.

Note that components 1 to 5 having the same functions as those shown in FIGS. 1 and 7 are denoted by the same reference numerals as those shown in FIG. Further, for the sake of convenience of explanation, in the present embodiment, the description of the same configuration as that shown in the first embodiment and the second embodiment is omitted.

The merging function of the present example has a function of merging keyword extraction rules held in the keyword extraction rule holding device 4 at any time, for example. Will be described. As an example, it is assumed that the following two keyword extraction rules Er ₃ and Er ₄ are included in the keyword extraction rules held by the keyword extraction rule holding device 4. Er ₃ : ({[printing status]}, {but, is}, {“yes (no)”, “yes
(No) ”)) → {f} Er ₄ : ({[printing state]}, {but, wo}, {“ na (i) ”}) → {f}

At this time, the two keyword extraction rules
Merged into a single keyword extraction rule
Can improve the readability of the
Speed up the extraction rule and refine the extraction rules.
Obtain the effect of increasing the degree
Can be. Specifically, the above Er_Three, Er_FourMerge
Keyword extraction rule Er obtained ₃₄Is as follows. Er₃₄: ({[Printing status]}, {but, is}, {“yes (no)”, “
(No) ”,“ na (i) ”}) → {f}

Incidentally, taking this case as an example, the above-mentioned 2
Keyword extraction rules Er_Three,, Er _FourIs said in the present invention
Keywords and second keywords defined by rules
Only one of the groups and relationships is equivalent to different rules
And the above-mentioned keyword extraction rule Er₃₄Says to the present invention
This corresponds to one rule in which the different parts are summed.

As another example, the following two keyword extraction rules Er ₅ , E ₅ are included in the keyword extraction rules held by the keyword extraction rule holding device 4.
and there is an r _6. Er ₅ : ({[ease of use]}, {is, is}, {“evil”)} → {f} Er ₆ : ({[service]}, {is, is}, {“evil ) ”}) → {f}

At this time, the thesaurus T ₁ shown in FIG.
"Ease of use", because the higher-level concepts that are common to two of the concept of "service" is a "symptom" in, Er ₅ described above,
The keyword extraction rule Er ₅₆ obtained by merging the Er _6, as follows are assumed as an example. Er ₅₆ : ({[symptoms]}, {but, is}, {“evil”)} → {f}

However, when such a merge is executed, the concepts of "failure", "operation", and "quality" are included in addition to the "usability" and "service" originally included. Therefore, for example, when the user recognizes this, according to the instruction from the user, the above Er ₅₆
Is the merged keyword extraction rule. On the other hand, if the user does not recognize this, according to the instruction from the user, for example, the next keyword extraction rule Er ₅₆ ′ is set as the merged keyword extraction rule. Er ₅₆ ': ({[Convenience], [Service]}, {G, is}, {“Evil
(Yes) ”}) → {f}

Incidentally, taking this case as an example, the above-mentioned 2
Keyword extraction pieces of keyword extraction rule Er _5, Er ₆ corresponds to only one different rules of the first keyword group and the second keyword group and relationships defined by the rules referred to in the present invention, the above-mentioned The rule Er ₅₆ and the above-described keyword extraction rule Er ₅₆ ′ correspond to one rule summing up the different parts according to the present invention.

Here, the configuration of the keyword extraction rule refining support device and the keyword extraction device according to the present invention is as follows.
The configuration is not necessarily limited to the above, and various configurations may be used. For example, in the above embodiment, for the sake of convenience of explanation, while the device having the function of strengthening the constraint of the keyword extraction rule in the first embodiment is shown, the constraint of the keyword extraction rule is relaxed in the second embodiment. Although the device having the function has been described, it is also a preferable embodiment to use a configuration having both such a constraint strengthening function and such a constraint relaxing function.

Also, for example, in the above embodiment, as a preferred mode, various processes performed by the keyword extraction rule refining support device and the keyword extraction device according to the present invention include, for example, a hardware including a processor and a memory. Although the configuration is such that the processor is controlled by executing the control program stored in the ROM in the resources, for example, each functional means for executing the processing may be configured as an independent hardware circuit.

The present invention also relates to a floppy (registered trademark) disk or CD-ROM storing the above-described control program.
It can be understood as a computer-readable storage medium such as a ROM, and the processing according to the present invention can be performed by inputting the control program from the storage medium to the computer and causing the processor to execute the control program.

[0106]

As described above, according to the keyword extraction rule refining support apparatus and the storage medium of the present invention, for example, when a keyword is not to be extracted but matches a certain keyword extraction rule. Even so,
It is possible to strengthen the constraint of the rule so that it does not match, and even if, for example, a keyword should be extracted but there is no matching keyword extraction rule, a keyword close to its expression The constraint of the extraction rule can be relaxed so as to match. For this reason, for example, the keyword extraction rule can be refined by repeatedly performing the constraint strengthening process and the constraint relaxation process of the existing keyword extraction rule according to the document.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating an example of a keyword extraction device to which a keyword extraction rule refinement support device according to a first embodiment of the present invention is applied.

FIG. 2 is a diagram showing an example of element positions of a morpheme.

FIG. 3 is a diagram showing an example of data of a thesaurus.

FIG. 4 is a diagram illustrating a configuration example of a keyword extraction rule constraint strengthening device.

FIG. 5 is a diagram illustrating a display example by the constraint strengthening instruction device.

FIG. 6 is a diagram illustrating a display example of the constraint strengthening rule candidate presentation device.

FIG. 7 is a diagram illustrating an example of a keyword extraction device to which a keyword extraction rule refinement support device according to a second embodiment of the present invention is applied.

FIG. 8 is a diagram illustrating a configuration example of a keyword extraction rule constraint relaxation device.

FIG. 9 is a diagram illustrating a display example by the constraint relaxation instructing device.

FIG. 10 is a diagram showing a display example by the constraint relaxation rule candidate presentation device.

FIG. 11 is a diagram illustrating an example of a keyword extraction device.

[Explanation of symbols]

1. Document group holding device 2. Dependency analysis device 3,
··· Pattern matching device, 4 ·· Keyword extraction rule holding device, 5 ·· thesaurus, 6 ·· Keyword extraction rule constraint strengthening device, 11 ·· Constraint strengthening instruction device, 12 •• Constraint strengthening refinement device, 13 •• Constraint Reinforcement rule candidate presentation device, 21 ... Similar rule selection device, 2
2. Keyword extraction rule constraint relaxation device, 31
Constraint relaxation instruction device, 32 ... constraint relaxation refinement device, 33
..Constrained relaxation rule candidate presentation device

Claims (11)

[Claims] 1. A predetermined keyword is extracted from an expression in which a word corresponding to a first keyword included in a first keyword group and a word corresponding to a second keyword included in a second keyword group appear in a predetermined relationship. A keyword extraction rule refining support device that supports refining the rule of a keyword extraction device that extracts a keyword from a document based on the rule, and a presentation unit that presents an expression that conforms to the rule and the rule to a user, A receiving means for receiving from the user an instruction to apply the rule presented to the presented expression; and accepting the instruction so that, when the instruction is accepted, the rule presented to the presented expression is not applied. A keyword extraction rule refining support device, comprising: a constraint strengthening means for changing. 2. The keyword extraction rule refining support device according to claim 1, wherein the constraint strengthening unit replaces a keyword corresponding to a word that matches the presented rule in the presented expression with the keyword specified by the rule. A keyword extraction rule refining support device characterized in that a presented rule is made unapplicable to a presented expression by deleting the keyword from a corresponding keyword group. 3. The keyword extraction rule refining support device according to claim 1, wherein the constraint strengthening means converts a keyword corresponding to a word that matches the presented rule in the presented expression into a plurality of subordinate concepts. , By removing the keyword corresponding to the word from the plurality of keywords of these lower concepts from the keyword group corresponding to the word specified in the rule, the rule presented for the presented expression is not applied. A keyword extraction rule refining support device characterized in that: 4. A predetermined keyword is extracted from an expression in which a word corresponding to a first keyword included in a first keyword group and a word corresponding to a second keyword included in a second keyword group appear in a predetermined relationship. A keyword extraction rule refinement support device that assists a keyword extraction device that extracts a keyword from a document based on a rule to refine the rule, comprising: a first keyword and a second keyword defined by the rule; A presentation means for presenting to the user an expression that matches only two and the rule; a receiving means for receiving from the user an instruction to apply the rule presented to the presented expression; and a presentation means for receiving the instruction when the instruction is received. A constraint easing means for changing the rule so that the rule presented for the expression is applied. Keyword extraction rule refining assisting apparatus characterized. 5. The keyword extraction rule refining support device according to claim 4, wherein the constraint relaxation unit defines a keyword indicating a concept of a word in the presented expression that does not conform to the presented rule in the rule. A keyword extraction rule refining support device, characterized in that a rule presented is applied to a presented expression by adding it to a keyword group corresponding to the word. 6. The keyword extraction rule refining support device according to claim 4, wherein the constraint relaxation unit is defined by a keyword indicating a concept of a word in the presented expression that does not conform to the presented rule and the rule. By replacing the keyword in the keyword group with a keyword indicating a common broad concept with the keyword in the keyword group corresponding to the word, the presented rule is applied to the presented expression. Keyword extraction rule refinement support device. 7. The keyword extraction rule refining support device according to claim 4, wherein the constraint relaxation unit converts a word in the presented expression that does not conform to the presented rule to the word defined by the rule. A keyword extraction rule refining support apparatus characterized in that a rule presented is applied to a presented expression by adding it to a keyword group. 8. The keyword extraction rule refining support device according to claim 4, wherein the constraint easing means corresponds to a word in the presented expression that does not conform to the presented rule and the word defined by the rule. Keyword extraction characterized by applying the presented rule to the presented expression by replacing the superficial keyword with a keyword indicating a common superordinate concept with the superficial keyword in the keyword group. Rule refinement support device. 9. The keyword extraction rule refining support device according to claim 1, wherein only one of the first keyword group, the second keyword group, and the relation defined by the rule is provided. Different rules,
A refining support system for keyword extraction rules, comprising a merging means for converting the different parts into one rule that is the sum of the different parts. 10. A storage medium in which a program to be executed by a computer is stored readable by input means of the computer, wherein a word corresponding to the first keyword included in the first keyword group and a second word included in the second keyword group are included. A keyword extraction device that extracts a keyword from a document based on a rule that extracts a predetermined keyword from an expression in which a word corresponding to the keyword appears with a predetermined relationship is used to assist in refining the rule. The program includes a process of presenting an expression that conforms to the rule and the rule to the user, a process of receiving from the user an instruction to apply the rule presented to the presented expression, and a process of presenting the instruction when the instruction is received. To change the rule so that the rule presented for the specified expression is not applicable Storage medium, characterized in that to execute the door to the computer. 11. A storage medium storing a program to be executed by a computer in a manner readable by an input means of the computer, wherein a word corresponding to the first keyword included in the first keyword group and a second word included in the second keyword group are included. A keyword extraction device that extracts a keyword from a document based on a rule that extracts a predetermined keyword from an expression in which a word corresponding to the keyword appears with a predetermined relationship is used to assist in refining the rule. The program presents an expression that matches only two of the first keyword, the second keyword, and the relationship defined by the rule and a process of presenting the rule to the user, and applies the presented rule to the presented expression. The process of receiving an instruction from the user and, when the instruction is received, Storage medium and a process of changing the rule to the presented rule is applied, characterized in that to execute the computer to.