JPH1185766A - Method and device for extracting keyword and storage medium storing keyword extraction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To eliminate the noise of keyword extraction and to attain improvement in the performance of a retrieval system by extracting a keyword corresponding to the role of a noun phrase to the verb of an input sentence and the modification relation between phrases based on the semantic structure of a syntax analyzed sentence. SOLUTION: When a natural sentence is inputted to a morpheme analytic part 20, at the analytic part 20, parts of speech of respective words are identified by dividing the input sentence into words while referring to contents in a word information storage part 10. A syntax analytic part 30 inputs the morpheme analyzed result and outputs syntax structure while using phrase structure rules, and a semantic analytic part 40 analyzes the deep case of the verb and the modification relation between phrases by applying respective grammatical rules to the syntax structure and outputs the semantic structure expressed by a case frame. Then, an intension comprehending part 60 inputs the semantic structure, reads rules expressing which modification related word is important in the entire sentence out of a keyword extraction rule storage part 50 and extracts the keyword. Therefore, only the required keyword can be extracted without fail.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a keyword extracting method and apparatus, and a storage medium storing a keyword extracting program, and more particularly, to a search key in an information guidance service when an inquiry from a user is a natural sentence. The present invention relates to a keyword extraction method and apparatus in an information search system capable of extracting a keyword used as a keyword from an input sentence, and a storage medium storing a keyword extraction program.

[0002]

2. Description of the Related Art In extracting keywords from natural sentences, words having a specified part of speech and meaning are extracted from a set of words obtained by morphological analysis. As another method, as disclosed in Japanese Patent Application Laid-Open No. 9-114840, a statistical method of extracting words frequently appearing in text sentences such as documents as keywords is used.

[0003]

However, in the method of extracting a word having a specified part of speech and meaning from a word set obtained by morphological analysis, the role of the word in the sentence is not taken into account. There is a problem that insignificant words are extracted as keywords in sentences.

In order to prevent the extraction of such insignificant words, if the specified part of speech and the condition of the meaning are restricted, there is a problem that words to be extracted are not extracted. For example, "I want to buy a car to go to the
"car dealer" in response to the query "office"
Is the subject of guidance, but if the conditions of extracted keywords are verbs and nouns, not only "buy" and "car", which are important as keywords corresponding to the field,
Unimportant "go" and "office" are extracted. By restricting the condition to nouns, on the other hand, the keyword "buy
Is not extracted.

In the statistical method using the appearance frequency,
At most a few, like queries to search systems
A short sentence of about a sentence has a problem that keywords cannot be appropriately extracted because the number of words is small. The present invention
In view of the above, it is possible to provide a keyword extraction method and apparatus, and a storage medium storing a keyword extraction program, which can eliminate noise in keyword extraction and can improve the performance of a search system. With the goal.

[0006]

FIG. 1 is a diagram for explaining the principle of the present invention. According to the present invention, in a keyword extraction method for extracting a keyword from a natural sentence, when a natural sentence is input as an input sentence (step 1), morphological analysis and syntax analysis of the natural sentence are performed (step 2), and the syntax is analyzed. The semantic structure of the sentence is obtained (step 3), and the keyword is extracted based on the role of the noun phrase for the verb of the input sentence parsed based on the semantic structure and the modification relation between the phrases using the keyword extraction rule. (Step 4).

Further, according to the present invention, when extracting a keyword, the type of the keyword to be extracted is specified by using a different keyword extraction rule depending on the target field.
Further, when extracting a keyword, the present invention specifies, from an input sentence, a conditional phrase that is a phrase representing a user request, and when the conditional phrase is a verb (hereinafter, a conditional verb), specifies a conditional verb. A keyword type is specified for a verb and a noun phrase in an input sentence by using a keyword extraction rule including a combination of a role of a noun phrase (hereinafter referred to as a deep case) and a type of a keyword stored in the deep case.

Further, according to the present invention, when extracting a keyword, a condition phrase which is a phrase representing a user request is specified from an input sentence, and the condition phrase is a noun phrase meaning an occupation (hereinafter, a condition noun phrase). In the case of, the type of the keyword is specified for the verb and the noun phrase in the input sentence using the rule of the combination of the type of the keyword that modifies the conditional noun phrase. FIG.
FIG. 1 is a diagram illustrating the principle of the present invention.

The present invention relates to a keyword extracting apparatus for extracting a keyword from a natural sentence, comprising an input means 70 for inputting a natural sentence as an input sentence, a word information storage means 10 for storing grammatical information and semantic information of the word. A morphological analysis means 20 for morphologically analyzing an input sentence with reference to the contents of the word information storage means 10, a syntactic analysis means 30 for obtaining a syntax structure of the morphologically analyzed sentence, and a syntax structure obtained by the syntax analysis means. Based on the semantic structure, the semantic analysis unit 40 for obtaining the semantic structure of the input sentence based on the semantic structure is used. Keyword extracting means 6 for extracting a keyword based on the role of a noun phrase for a verb (hereinafter referred to as a deep case) and the modification relation between the phrases With the door.

The keyword extraction rule storage means 50 stores different keyword extraction rules depending on the target field, and the keyword extraction means 60 specifies the type of keyword to be extracted and refers to the keyword extraction rule. And a first type specifying unit that performs the operation.

When the condition phrase representing the user's request is a verb (hereinafter, the verb is referred to as a conditional verb), the keyword extraction rule storage means 50 stores the type of the deep case of the conditional verb and the The keyword extraction means 60 includes rules for combinations of types of keywords stored in deep cases.
From the input sentence, using a conditional phrase extracting means for specifying a conditional phrase that is a phrase representing the user's request, and a combination rule of the type of the deep case of the conditional verb and the type of the keyword stored in the deep case, A second type specifying means for specifying the type of the keyword for the verb and the noun phrase in the input sentence is included.

When the condition phrase representing the user's request is a noun phrase (condition noun phrase), the keyword extraction rule storage means 50 stores the rule of the combination of the types of keywords for modifying the conditional noun phrase. The keyword extracting unit 60 includes a third type specifying unit that specifies the type of the keyword for the verb and the noun phrase in the input sentence by using the rule of the combination of the type of the keyword that modifies the conditional noun phrase.

[0013] The present invention is a storage medium storing a keyword extraction program for extracting a keyword from a natural sentence, comprising: an input process for inputting a natural sentence as an input sentence;
A morphological analysis process for morphologically analyzing an input sentence using grammatical information and semantic information of a word, a parsing process for obtaining a syntactic structure of the morphologically analyzed sentence, and an input sentence based on a syntactic structure obtained by the parsing process Based on the semantic structure, the role of the noun phrase for the verb (hereinafter referred to as the deep case) and the inter-phrase modification, using the semantic analysis process for finding the semantic structure of A keyword extraction process for extracting a keyword based on the relationship.

[0014] The above-described keyword extraction process includes:
The method includes a first type specifying process of specifying the type of the keyword to be extracted and referring to a different rule depending on the target field. The above keyword extraction process includes a condition phrase extraction process that specifies a condition phrase that is a phrase representing a user's request from an input sentence, and a condition phrase that represents a user's request is a verb (hereinafter, the verb is referred to as a verb). Described as a conditional verb)
A second type specification for specifying a keyword type for a verb and a noun phrase in an input sentence using a rule of a combination of a deep case type of the conditional verb and a keyword type stored in the deep case. Including processes.

[0015] The above-described keyword extraction process includes:
When the conditional phrase representing the user's request is a noun phrase (hereinafter,
The noun phrase is referred to as a conditional noun phrase), using a rule of a combination of types of keywords that modify the conditional noun phrase,
The method includes a third type specifying process for specifying the type of the keyword for the verb and the noun phrase in the input sentence.

As described above, the present invention provides, in the text,
Keyword is extracted after considering which dependency word is important in the whole sentence,
Only necessary keywords can be extracted without omission. In addition, by changing the rule to be extracted for each target field, there is the flexibility to extract an appropriate keyword for each field.

[0017]

FIG. 3 shows the configuration of a keyword extracting apparatus according to the present invention. Keyword extraction device 10 shown in FIG.
0 is a word information storage unit 10 that stores grammatical information and semantic information of a word, a morphological analysis unit 20 that divides an input sentence into words using information stored in the word information storage unit 10, and a morphological analysis result. A syntactic analysis unit 30 for obtaining the syntax structure of the input sentence, a semantic analysis unit 40 for obtaining the semantic structure of the input sentence based on the syntax structure, a keyword extraction rule storage unit 50 for storing keyword extraction rules for extracting keywords, and a keyword extraction rule And an intention understanding unit 60 for extracting a keyword based on the semantic structure using the keyword.

Next, the operation of keyword extraction according to the present invention will be described. FIG. 4 is a flowchart showing the operation of keyword extraction according to the present invention. Step 101) When a natural sentence is input, the morphological analysis unit 20 divides the input sentence into words by referring to the word information in the word information storage unit 10. Step 102) Based on the result of the morphological analysis, the syntax analyzer 30 obtains the syntax structure of the input sentence.

Step 103) The semantic analysis unit 40 obtains the semantic structure of the input sentence based on the syntax structure. Step 104) The intention understanding unit 60 extracts a keyword based on the semantic structure using the keyword extraction rule stored in the keyword extraction rule storage unit 50.

[0020]

An embodiment of the present invention will be described below with reference to the drawings. [First Embodiment] The first embodiment will be described below with reference to the flow chart shown in FIG. First, the input sentence is “I want t
o buy a car to go to the office.

When the above-mentioned input sentence is input to the morphological analysis section 20, the input sentence is divided into words by referring to the contents of the word information storage section 10. Since the example of the above input sentence is an English sentence, each word is divided in advance by a space. Therefore, the morphological analysis unit 20 performs a process of identifying the part of speech of each word. As a result, the following morphological analysis result is obtained. I / N want / V to / PREP buy / V a / DE
T car / N to / PREP go / V to / PREP the / DET
The office / N syntax analysis unit 30 receives the above morphological analysis result as input, and outputs a syntax structure of the input sentence using phrase structure rules. FIG. 5 shows an example.

Next, the semantic analysis unit 40 receives the syntax structure shown in FIG. 5 and outputs a semantic structure represented by a case frame using each grammar rule. Analyze the deep case of verbs and the modification relations between phrases. FIG. 6A shows an example thereof. Next, the intention understanding unit 60 extracts the keyword by referring to the contents of the keyword extraction rule storage unit 50 by using the meaning structure shown in FIG.

FIG. 6B is an example of a keyword extraction rule. The rule is that, for the semantic structure shown in FIG. 6B, a verb at a predetermined position (thick frame) indicates a user's request, A noun phrase having the meaning of thing, person, or occupation indicates that the keyword is used as a product keyword. Intention understanding unit 60
Determines that among the keyword extraction rules stored in the keyword extraction rule storage unit 50, the semantic structure of FIG. 6B matches the semantic structure of the input sentence (FIG. 6A),
Based on the extraction rule, “buy” is a verb keyword,
“Car” is extracted as a product keyword. The result is shown in FIG.

[Second Embodiment] Next, as a second embodiment, a method of specifying the type of a keyword to be extracted from an input sentence according to a target field will be described. FIG.
FIG. 9 is a flowchart of an operation of specifying a type of a keyword to be extracted from an input sentence according to a target field of the second embodiment of the present invention.

The following flow chart shows an input sentence
An example will be described in which "I want to go to a shrine tour by train in August" is input. Step 201 to step 203) Morphological analyzer 20
The semantic structure of the input sentence is output through the morphological analysis processing by the syntax analysis section, the syntax analysis processing by the syntax analysis section 30, and the semantic analysis processing by the semantic analysis section. This example is shown in FIG.

Step 204) In the intention understanding process in the intention understanding unit 60, keywords of goods, verbs, and occupations, which are keywords independent of the field, are extracted. That is, with the semantic structure as an input, the keyword extraction rule storage unit 50 searches the keyword extraction rule (FIG. 8B) irrespective of the field and matches the semantic structure (FIG. 8A) to extract the keyword. The result (FIG. 8C) is output.

Step 205) The field inference process in the intention understanding unit 60 is a keyword extraction result (FIG. 8C).
Is used as an input to infer a field name for classifying a target to be guided by the search system. FIG. 8D shows the inference result. Step 206) In the intention understanding process 2 in the intention understanding unit 60, keywords depending on the field are extracted. That is,
The field inference result (FIG. 8D) is input from the keyword extraction rule storage unit 50 with the field inference result (FIG. 8D) as input.
For the keyword extraction rule corresponding to the field "Tourist information". This processing is performed by using the semantic structure shown in FIG.
By matching the keyword extraction rule 2 shown in (b), a keyword extraction result shown in FIG. 9C is output.

Third Embodiment Next, as a third embodiment, when the conditional phrase is a verb (conditional verb), the type of the deep case of the conditional verb and the keyword stored in the deep case are described. An example will be described in which a keyword type is specified for a verb or a noun phrase of an input sentence using a keyword extraction rule that stores a rule of a combination of types.

FIG. 10 shows the structure of the intention understanding section of the third embodiment of the present invention. The intention understanding unit 60 shown in FIG. 1 includes a conditional phrase extraction unit 61 and a keyword extraction unit 62.
Here, the conditional phrase is a word that forms the center of the question sentence, and serves as a base point for processing for extracting other keywords. The operation of the condition phrase extraction unit 61 will be described below.

FIG. 11 is a flowchart of the operation of the conditional phrase extraction unit according to the third embodiment of the present invention. In the following conditional phrase extraction process, a process for extracting a conditional phrase when Japanese is input is shown. Here, a verb used to request information of a store from the search system, such as “I want to know” or “I want to teach”, is defined as a request verb.

Step 301) First, it is determined whether or not the verb of the last phrase of the input sentence is a request verb. If the verb is not a request verb, the process proceeds to step 302, and if it is a request verb, the process proceeds to step 303. Step 302) If the verb is not a request verb (for example, "want to buy" or "want to eat"), the verb of the last phrase is used as a conditional phrase.

Step 303) If the verb is a request verb, the meaning of the target case of the verb in the final phrase is determined to be an object, a thing, or a person. ), The process proceeds to step 304. Otherwise, the process proceeds to step 305. Step 304) In the case of a thing, thing, or person, the verb of the last clause is used as a conditional phrase.

Step 305) It is determined whether the target case of the verb in the final phrase is a profession or a profession-equivalent pronoun (store, place, trader, etc.). Tell me), go to step 306, otherwise go to step 30
Move to 7. Step 306) If the occupation is a pronoun equivalent to the occupation, the occupation is an important phrase, so the occupation is set as the conditional phrase. Here, an item representing the attribute of the store, such as "address" or "telephone number" of "I want to know the address of the restaurant", is the subject.

Step 307) It is determined whether or not the target case of the verb in the final phrase is the subject, and if there is an occupation that modifies the subject, the process proceeds to step 308. If not, the process proceeds to step 309. I do. Step 308) If there is an occupation that modifies the subject, since the occupation that modifies the subject is an important phrase, the occupation is used as the conditional phrase.

Step 309) Otherwise,
There is no conditional phrase. In the above flowchart, like "TV"
If the last phrase of the input sentence is not a verb but a product (thing, thing, person), it is assumed that the input sentence is a question to the information guide system. And output the semantic structure. As a result, in the conditional phrase extraction unit 61 of the intention understanding unit 60, since “shop” is a pronoun equivalent to the occupation,
At 6, "store" is the conditional phrase.

If the last sentence of the input sentence is not a verb but a profession like "restaurant", or if it is a subject like "restaurant address", it is also assumed that the question is to the information guidance system. Then, the semantic analysis unit 40 complements “I want to know” after the input sentence and outputs a semantic structure. As a result, in the conditional phrase extracting unit 61 of the intention understanding unit 60, step 305 or step 307 matches, and the occupation becomes the conditional phrase.

Next, the operation of the keyword extracting unit 62 will be described. FIG. 12 is an example of a keyword extraction rule according to the third embodiment of the present invention. The rule is stored in the keyword extraction rule storage unit 50. It is an example of the rule of the combination of the type of the deep case of the conditional verb and the type of the keyword stored in the deep case. Here, FIG. 13 shows a semantic structure when the input sentence “I want to have a party at a restaurant” is input.

The intention understanding unit 60 specifies the verb “do” as a conditional phrase by the conditional phrase extracting unit 61. The keyword extraction unit 62 sequentially compares the extraction rule whose conditional phrase is a verb with the semantic structure of the input. Rule 1 in FIG. 12 indicates that if the target case of the conditional verb is a noun phrase having the meaning of thing, thing, person, or occupation, the noun phrase is used as a product keyword, and the conditional verb is used as a verb keyword. Intention understanding unit 60
Is because the semantic structure of the target case matches the structure of rule 1,
“Party” is a product keyword, and “do” is a verb keyword.

Rule 2 in FIG. 12 indicates that if the place verb of the conditional verb is a noun phrase having the meaning of occupation, the noun phrase is used as the occupation keyword. Intention understanding unit 60
Is because the semantic structure of the locative matches the structure of Rule 2,
“Restaurant” is used as the occupation keyword. FIG. 14 shows the semantic structure of the input sentence “I want to buy a device that can use the Internet”. The intention understanding unit 60 includes the conditional phrase extraction unit 6
1, the verb “buy” is specified as a conditional phrase. Rule 3 in FIG. 12 is a noun phrase in which the target case of the conditional verb has the meaning of a story tail or a word tail, and the noun phrase has a verb to modify,
Further, it indicates a specific rule of the type of keyword when the target case of the verb is a noun phrase having the meaning of thing, thing, person, or occupation.

Since the meaning structure of the target case matches the structure of Rule 3 in FIG. 12, the intention understanding unit 60 uses “(verb) device (story tail) using the Internet (thing)” as a product keyword and “ "Buy" as a verb keyword. In this case, the product keyword has a structure of “Internet” + “use”, and in the field inference process (step 205), what “internet” + “use” is further inferred.

Next, the processing of the conditional phrase extraction unit 61 when English is input will be described. FIG. 15 is a flowchart of the conditional phrase extraction processing when English is input according to the third embodiment of this invention. FIG. 16 shows an example of a keyword extraction rule in the following example, and FIG. 17 shows an example of a semantic structure. For English input sentence, deep case of verb, modification relation between phrases, head relation between phrases in semantic analysis (the head is a phrase included in the phrase that plays a central grammatical role) ) Is output. For example, FIG. 17 shows the input sentence “I wa
nt to buy a car to go to the office ”, but the head of the entire input sentence is the verb“ want ”.

The processing flow will be described below using the input sentence of FIG. First, the verb "want" of the head of the input sentence
Is the target of the conditional phrase extraction processing (step 401).
It is determined whether the target verb is a request verb. Here, the target verb “want” is a request verb because it can represent a request to the search system (step 40).
2). If the target verb is not a request verb, the target verb is used as a conditional phrase (step 403).

It is determined whether or not the target case of the target verb is a verb. In this case, the target case of "want" is the verb "buy"
(Step 404), the new verb "buy"
Is set as a new target (step 405). Since the target verb "buy" is not a request verb, "buy" is used as a conditional phrase (step 403). The processing after step 406 is the same as the processing after step 303 in FIG.

When the head of the input sentence is not a verb but a product, such as “television”, a supplementary process such as “know the shop of (product)” is performed in the semantic analysis unit 40 in the same manner as in Japanese. And output the semantic structure. As a result, in the conditional phrase extraction unit 61 of the intention understanding unit 60, "shop" is a noun equivalent to the occupation, so it is matched in step 408, and "shop" is the conditional phrase.

As in the case of "restaurant", the last phrase of the input sentence is not a verb but is a occupation, or "the address of ar
Even in the case of a subject such as "estaurant", in the same way as in Japanese, the semantic analysis unit 40 precedes the input sentence with "know"
And output the semantic structure. As a result, the conditional phrase extraction unit 61 of the intention understanding unit 60 matches in step 408 and step 410, and the occupation becomes the conditional phrase.

Next, the processing of the keyword extracting unit 62 when English is input will be described. Description will be made using the semantic structure shown in FIG. The intention understanding unit 60 specifies the verb “buy” as a conditional phrase by the conditional phrase extracting unit 61.
In the keyword extraction unit 62, the extraction rules stored in the keyword extraction rule storage unit 50 and whose conditional phrases are verbs are sequentially collated with the semantic structure of the input. Since the meaning structure of the target case matches the structure of the keyword extraction rule shown in FIG. 16, the intention understanding unit 60 uses “car” as the product keyword and “buy” as the verb keyword. Since there is no extraction rule in the keyword extraction rule storage unit 50 that uses a verb as a target case of a conditional verb as a keyword, “go” is not a keyword.

[Fourth Embodiment] Next, as a fourth embodiment, when the conditional phrase is a noun phrase meaning an occupation, the rule of the combination of the types of keywords for modifying the conditional noun phrase is used. An example in which the type of keyword is specified for a verb or a noun phrase in an input sentence will be described. FIG. 18 shows an example of a keyword extraction rule according to the fourth embodiment of the present invention.

The keyword extraction rule shown in FIG. 6 is an example of a rule of a combination of types of keywords to be modified into conditional noun phrases stored in the keyword extraction rule storage unit 50. FIG. 19 shows the meaning structure of the input sentence “restaurant where steak can be eaten”. The occupation “restaurant” is specified as a conditional phrase by the conditional phrase extracting unit 61 of the intention understanding unit 60. The keyword extraction unit 62 sequentially compares the keyword extraction rules whose conditional phrase is occupation with the semantic structure of the input. In rule 2 of FIG. 18, there is a verb to be modified to a conditional noun phrase, and further, a specific rule of the type of keyword when the target case of the verb is a noun phrase having the meaning of thing, thing, person, or occupation is shown. ing. Since the meaning structure of the input matches the structure of rule 2 in FIG. 18, the intention understanding unit 60 specifies “steak” as a product keyword, “eats” as a verb keyword,
“Restaurant” is used as the occupation keyword.

Input sentence "restaurant famous for steak"
Is shown in FIG. 19 (b). The intention understanding unit 60
The occupation “restaurant” is specified as a conditional phrase by the conditional phrase extracting unit 61. In rule 3 of FIG. 18, there is an adjective to modify the conditional noun phrase, and further, a thing to modify to the adjective,
This represents a specific rule for the type of keyword when there is a noun phrase having the meaning of thing, person, or occupation. Intention understanding unit 60
Since the semantic structure of the input matches the structure of rule 3 in FIG. 18, "steak" is used as a product keyword and "restaurant"
Is the occupation keyword.

As described above, according to the present invention, a morphological analysis, a syntactic analysis, and a semantic analysis process are performed on an input sentence, thereby obtaining a semantic structure representing a deep case of a verb and a modification relation between phrases. . Using the semantic structure as an input, a keyword can be extracted based on a keyword extraction rule that takes into account the dependency relationship between words. Further, the present invention is constructed by a program (software) and stored in a portable storage medium such as a disk device, a floppy disk, or a CD-ROM connected to a computer for extracting a keyword.
By installing as needed, general-purpose keywords can be extracted. It should be noted that the present invention is not limited to the above-described embodiment, but can be variously modified and applied within the scope of the claims.

[0051]

As described above, according to the present invention, a keyword is extracted by using a rule that expresses which dependency word in a sentence is a word that is important in the entire sentence. Only keywords can be extracted without omission. For this reason, noise of keyword extraction can be eliminated, and the performance of the search system can be improved.

Also, by changing the rule to be extracted according to the target field, an appropriate keyword can be extracted for each field, thereby providing flexibility in maintenance of the search system.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of a keyword extraction device of the present invention.

FIG. 4 is a flowchart illustrating an operation of keyword extraction according to the present invention.

FIG. 5 is an example of a syntax structure analysis result of the first embodiment of the present invention.

FIG. 6 is a diagram for explaining keyword extraction processing according to the first embodiment of this invention.

FIG. 7 is a flowchart of an operation of specifying a type of a keyword to be extracted from an input sentence according to a target field of the second embodiment of the present invention.

FIG. 8 is a diagram (part 1) for describing a keyword extraction process according to the second embodiment of the present invention;

FIG. 9 is a diagram (part 2) for explaining the keyword extraction process according to the second embodiment of the present invention;

FIG. 10 is a configuration diagram of an intention understanding unit according to a third embodiment of the present invention.

FIG. 11 is a flowchart illustrating the operation of a conditional phrase extraction unit according to the third embodiment of this invention.

FIG. 12 is an example (part 1) of a keyword extraction rule according to the third embodiment of the present invention.

FIG. 13 is an example (part 1) of a semantic structure according to the third embodiment of the present invention.

FIG. 14 is an example (part 2) of the semantic structure of the third embodiment of the present invention.

FIG. 15 is a flowchart of a conditional phrase extraction process when English is input according to the third embodiment of this invention.

FIG. 16 is an example (part 2) of a keyword extraction rule according to the third embodiment of this invention.

FIG. 17 is an example (part 3) of a semantic structure according to the third embodiment of the present invention.

FIG. 18 is an example of a keyword extraction rule according to the fourth embodiment of the present invention.

FIG. 19 is an example of a semantic structure of the fourth embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 Word information storage means, word information storage part 20 Morphological analysis means, morphological analysis part 30 Syntax analysis means, syntax analysis part 40 Semantic analysis means, semantic analysis part 50 Keyword extraction rule storage means, Keyword extraction rule storage part 60 Keyword extraction means , Intention analysis unit 61 conditional phrase extraction unit 62 keyword extraction unit 70 input means

Claims (12)

[Claims] In a keyword extraction method for extracting a keyword from a natural sentence, when the natural sentence is input as an input sentence, morphological analysis and syntax analysis of the natural sentence are performed, and the semantic structure of the parsed sentence is determined. A keyword extraction method comprising extracting a keyword based on a role of a noun phrase for a verb of an input sentence parsed based on a semantic structure using a keyword extraction rule and a modification relation between the phrases. 2. The keyword extracting method according to claim 1, wherein when extracting the keyword, the type of keyword to be extracted is specified using a different keyword extraction rule depending on a target field. 3. When extracting a keyword, a conditional phrase that is a phrase representing a user request is specified from the input sentence. If the conditional phrase is a verb (hereinafter, a conditional verb), the conditional verb A keyword type for a verb and a noun phrase in the input sentence using a keyword extraction rule composed of a combination of a role of a noun phrase (hereinafter referred to as a deep case) and a type of a keyword stored in the deep case The keyword extracting method according to claim 1, wherein 4. When extracting a keyword, a condition phrase that is a phrase representing a user's request is specified from the input sentence, and the condition phrase is a noun phrase meaning a job (hereinafter, a condition noun phrase). 2. The keyword extraction method according to claim 1, wherein the keyword type is specified for the verb and the noun phrase in the input sentence by using a rule of a combination of the keyword types for modifying the conditional noun phrase. 5. A keyword extracting device for extracting a keyword from a natural sentence, an input unit for inputting a natural sentence as an input sentence, a word information storage unit for storing grammatical information and semantic information of the word, and the word information Morphological analysis means for morphologically analyzing the input sentence with reference to information in the storage means; syntactic analysis means for obtaining a syntax structure of the morphologically analyzed sentence; and A semantic analysis unit for obtaining a semantic structure of the input sentence; a keyword extraction rule storage unit for storing a keyword extraction rule that is a rule for extracting a keyword; a verb based on the semantic structure using the keyword extraction rule; And a keyword extracting means for extracting a keyword based on a modification relation between the phrases and a role of a noun phrase (hereinafter referred to as a deep case) for the phrase. Keyword extracting device, characterized in that. 6. The keyword extraction rule storage means stores different keyword extraction rules depending on a target field, and the keyword extraction means specifies a type of a keyword to be extracted and refers to the keyword extraction rule. 6. The keyword extracting device according to claim 5, further comprising a first type specifying unit. 7. The keyword extraction rule storage means, when a conditional phrase representing a user request is a verb (hereinafter, the verb is referred to as a conditional verb), the type of the deep case of the conditional verb and the deep case The keyword extraction means includes: a condition phrase extraction means for specifying, from the input sentence, a condition phrase that is a phrase representing a user request; and a deep case of the conditional verb. 6. The method according to claim 5, further comprising a second type specifying unit that specifies a type of a keyword for a verb and a noun phrase in the input sentence using a combination rule of the type of the keyword and the type of the keyword stored in the deep case. Keyword extraction device. 8. The keyword extraction rule storage means according to claim 1, wherein the condition phrase representing the user's request is a noun phrase (hereinafter, referred to as a noun phrase).
The term noun phrase is referred to as a conditional noun phrase), and the rule of the combination of the types of keywords that modify the conditional noun phrase is included. 6. The keyword extracting apparatus according to claim 5, further comprising a third type specifying unit that specifies a type of a keyword for a verb and a noun phrase in the input sentence. 9. A storage medium storing a keyword extraction program for extracting a keyword from a natural sentence, comprising: an input process for inputting a natural sentence as an input sentence; and the input sentence using grammatical information and semantic information of a word. A morphological analysis process for performing morphological analysis, a syntactic analysis process for obtaining a syntactic structure of the morphologically analyzed sentence, a semantic analysis process for obtaining a semantic structure of the input sentence based on the syntactic structure obtained by the syntactic analysis process, and a keyword. Keyword extraction for extracting a keyword based on the role of a noun phrase for a verb (hereinafter, deep case) based on the semantic structure and a modification relationship between phrases using a keyword extraction rule that is a rule for extracting And a storage medium storing a keyword extraction program characterized by having a process. 10. The keyword extracting program according to claim 9, wherein the keyword extracting process includes a first type specifying process for specifying a type of a keyword to be extracted and referring to a different rule depending on a target field. Storage media. 11. The keyword extraction process includes: a condition phrase extraction process that specifies a condition phrase that is a phrase representing a user request from the input sentence; and a condition phrase that represents a user request is a verb (hereinafter, referred to as a verb). The verb is referred to as a conditional verb), and the rules of the combination of the type of the deep case of the conditional verb and the type of the keyword stored in the deep case are used for the verb and the noun phrase in the input sentence. The storage medium storing the keyword extraction program according to claim 9, further comprising a second type specifying process for specifying a type. 12. The keyword extraction process according to claim 1, wherein a condition phrase representing a user request is a noun phrase (hereinafter, referred to as a noun phrase).
The noun phrase is referred to as a conditional noun phrase), using a rule of a combination of types of keywords that modify the conditional noun phrase,
The storage medium according to claim 9, further comprising a third type specifying process for specifying a type of a keyword for a verb and a noun phrase in the input sentence.