JPH06168129A - Knowledge extracting device - Google Patents

Abstract

PURPOSE:To obtain not a simple associative concept pair of (concept 1)-(concept (i)) but a useful knowledge of (concept 1)-(concept (j))-(concept (i)) by referring to the knowledge stored in a main knowledge base and extracting a concept having prescribed relations to an associated concept. CONSTITUTION:An associating part 2 extracts associative relations of concepts from a text database 1. Knowledge of a subject area of data stored in the text database 1 is stored in a domain knowledge base 3. A knowledge synthesizing part 4 obtains an associated concept pair from the associating part 2 and refers to the domain knowledge base 3 to obtain the relations between these concepts and outputs them as knowledge. The concept pair of concepts 1 and 2 associated by the associating part 2 is extracted from the text database 1, and the domain knowledge base 3 is referred to clarify the relations of this concept pair by the knowledge synthesizing part 4, and new knowledge indicating that concepts 1 and 2 are related to a concept 3 is extracted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge extracting device for extracting an associative relationship between concepts (words) obtained from a text database as knowledge that reveals the relationship between concepts by referring to domain knowledge. It is a thing.

[0002]

2. Description of the Related Art FIG. 16 is a block diagram showing a conventional knowledge extracting apparatus disclosed in, for example, Japanese Patent Laid-Open No. 18962/1993, in which reference numeral 11 indicates learning of an associative concept pair consisting of a concept and its association. Neural network finished with a certain likelihood left by appropriate control, 12 is concept input means for inputting conceptual data to neural network 11, and 13 is an associative output for outputting output from neural network 11 as associative data. It is a means.

Next, the operation will be described. The user of the knowledge extraction device inputs the concept A1 from the concept input means 12 in order to widely study various concepts related to the certain concept A1. The inputted concept A1 is the neural network 11
Input signal. Since the neural network 11 remains in a certain likelihood, the associative concept pair (A1-B
1), the concept A1 is associated with not only the concept B1 but also the concepts B11 ,. ． ． ． ． , B1n are obtained from the associative output means 13. In this way, the user can use the concept B1 having an unexpectedness as the association of the concept A1.
1 ,. ． ． ． B1n can be obtained.

[0004]

Since the conventional knowledge extracting apparatus is constructed as described above, concept 1 and its concept 1
Concepts associated with 2. ． , Concept i ,. ． , The relationship with the concept n is not clear, (1) I do not know why the concept i is associated with the concept 1, and (2) the relationship between the concept 1 and the concept i is not clear. It is difficult to use in another knowledge information processing. (3) When a plurality of concepts are associated with concept 1, the difference between the associations is not clear. (4) Many concepts are associated with concept 1. In that case, there was a problem that it was not possible to prioritize the associated concept from the viewpoint of the purpose of use.

The present invention has been made in order to solve the above-mentioned problems, and includes concept 1 and concepts 2 ,. ． ． , Concept i ,. ． , By providing a "relationship" with the concept n, knowledge extraction that can obtain useful knowledge of (concept 1-relation j-concept i) rather than a simple associative concept pair of (concept 1-concept i) The purpose is to obtain the device.

[0006]

A knowledge extracting apparatus according to the invention of claim 1 refers to knowledge accumulated by a domain knowledge base, and identifies a concept having a predetermined relationship with a concept associated by an association unit. To extract.

The knowledge extracting device according to the invention of claim 2 is adapted to select a part of a text having structure information in addition to the knowledge extracting device of claim 1.

In addition to the knowledge extracting device of claim 1, the knowledge extracting device according to the invention of claim 3 is a representative one of a certain concept.
It is designed to be converted into one concept.

In addition to the knowledge extracting device according to the third aspect, the knowledge extracting device according to the fourth aspect of the present invention adds a certain concept determined by the user to be a synonym to the synonym conversion unit. .

A knowledge extracting device according to a fifth aspect of the present invention is, in addition to the knowledge extracting device of the first aspect, configured to convert one concept consisting of a plurality of words into a synonym for each word. Is.

[0011]

According to the invention of claim 1, the knowledge extracting device extracts a concept pair of concept 1 and concept 2 associated with each other in the associative unit from the text database and determines the relationship between the concept pair and the domain knowledge base. The new knowledge that the concept 1 and the concept 2 have the relationship of the concept 3 is clarified by referring to, and new knowledge is extracted.

The text structure selection unit in the invention of claim 2 selects a part of the text from the text having the structure.

The synonym conversion unit in the invention of claim 3 is
Convert a concept into one representative concept.

The synonym relation registration unit in the invention of claim 4 adds a certain concept determined by the user as a synonym to the synonym conversion unit.

The synonymous expression conversion unit in the invention of claim 5 converts one concept composed of a plurality of words into a synonym for each word.

[0016]

【Example】

Example 1. An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is an overall configuration diagram showing an embodiment of a knowledge extracting device according to the invention of claim 1. In the figure, 1 is a text database that is a source of knowledge to be extracted, 2 is an associative unit that extracts an associative relationship of concepts from the text database 1, and 3 is knowledge of a topic area of data accumulated in the text database 1. Domain knowledge base, 4
Is a knowledge synthesizing unit that obtains an associative concept pair from the associative unit 2, obtains the relationship between the concepts by referring to the domain knowledge base 3, and outputs it as knowledge.

Next, the operation will be described. FIG. 2 is an example of the domain knowledge base 3. FIG. 2A shows a knowledge representation format, and domain knowledge is represented by a triplet of (concept 1, relationship r, concept 2). This means that concept 1 has a relationship of r with concept 2. As the relation r, there is an isa relation that the concept 2 is a superordinate concept of the concept 1 and an eq relation that the concept 1 and the concept 2 are synonymous. Figure 2
FIG. 2B shows the domain knowledge of FIG. 2A in a graph structure in order to facilitate understanding.

FIG. 3 shows an example of the associative unit 2. The association section 2
It is composed of a concept association relation extraction unit 21, a concept association network 22, and an association control unit 23. Conceptual association relation extraction unit 21
Extracts an associative relationship between concepts included in the text according to the contents of the text database 1. One way to find the association of concepts is to use the co-occurrence frequency between concepts. If concept i and concept j are in the same text data, concept i and concept j are said to co-occur. If concept i and concept j co-occur for many text data,
It can be said that the concept i and the concept j have an associative relationship. The concept associative relationship obtained from the concept associative relationship extracting unit 21 is expressed by the concept associative network 22. Concept i and concept j are linked by a link of association strength Wij. As Wij, for example, a normalized frequency of co-occurrence of concept i and concept j can be used. The associative control unit 23 uses the concept associative network 22 to specify a concept a, which is associated with the concept 1, concept 2 ,. ． ． , Concept i ,. ． ． , Get concept n. Search for associated concepts 1
As one of the methods, there is a method in which, among the concepts connected to the concept a by a link of co-occurrence relation, the strength of association is equal to or higher than a certain threshold.

FIG. 4 is a flow chart for explaining the operation of the knowledge synthesizing section 4. The operation of the knowledge synthesizing unit 4 will be described with reference to FIG. First, in step ST1, the concept a from the associative unit 2
Concept 1, concept 2 ,. ． ． , Concept k ,. ． ． , Get concept n. Next, the processes from step ST2 to step ST9 are sequentially performed from concept 1 to concept n. The operation of the concept i, which is the i-th concept, will be described below. In step ST2, the concept i is converted to the concept b
And In step ST3, the domain knowledge base having the concept b in the first term of the triple is searched. In step ST4, it is determined whether the triplet is in the domain knowledge base. If yes, go to step ST5; otherwise, go to step ST9. In step ST5, the triplet is defined as (concept b, relation r, concept c). Relation r in step ST6
Find out what kind of. When the relation r is the equivalence relation equ, the concept c is set as the concept b in step ST8, and the process returns to step ST3. When the relation r is the hierarchical relation isa, (concept a, concept c, concept i) is stored as the extracted knowledge in step ST7. If the triplet is not found in step ST4, (concept a, association, concept i) is stored as extracted knowledge in step ST9. In step ST10, the knowledge extraction results from concept 1 to concept n are summarized as the knowledge extraction result from concept a.

As an example, from the concept of the expert system (corresponding to concept a in the above description), the concepts of Lisp (corresponding to concept 1 in the above description) and if-then rule (corresponding to concept 2 in the above description) are associated. Assuming that the domain knowledge base has the contents shown in FIG. 2, the extracted knowledge is (expert system, programming language, Lisp), (expert system, knowledge representation,
if-then rule). This means that the expert system is written in a programming language called Lisp and often uses if-then rules as a knowledge expression.

Example 2. FIG. 5 is an overall configuration diagram showing an embodiment of a knowledge extracting device according to the invention of claim 2. In the figure, when the text stored in the text database 1 has structure information, 5 is a text in which a part of the structure is selected and cut out from the text data and a part of the cut out text data is input to the associative unit 2. It is a structure selection unit.

Next, the operation will be described. FIG. 9 is an example of text data having structure information. There is a structure information table that holds the structure information corresponding to each text data. In the structure information table, partial names of text data and corresponding start positions and end positions of text data are described in correspondence with each other. When the text data is a technical document, generally, the title of the document, the author of the document, the affiliation of the author, the outline of the document, Chapters 1 and 2 of the text.
chapter,. ． ． , Conclusions, and references are part names.

FIG. 10 is a flow chart for explaining the operation of the text structure selection unit 5. The operation of the text structure selection unit 5 will be described with reference to FIG. The partial name of the text selected in step ST91 is obtained from the user and is set as P. P is, for example, “outline”. Then step S
At T92, the text T is obtained from the text database 1. In step ST93, it is determined whether the text T is empty.
If it is empty, the process ends. If it is not empty, the line of the partial name corresponding to P is searched from the structure information table of the text T in step ST94, and the start position S and the end position E of the P part in the text T are obtained. Next, in step ST95, the portion of the text T from the position S to the position E is cut out and sent to the associative unit 2 as a part of the selected text. Then, it returns to step ST92.

Example 3. FIG. 6 is an overall configuration diagram showing an embodiment of the knowledge extraction device according to the invention of claim 3. In the figure, 6 is a synonym conversion unit for converting a plurality of synonymous words into one standard word.

Next, the operation will be described. FIG. 11 is an example of the synonym conversion unit 6. When a word is given, the index part uses the features that can be easily and automatically obtained, such as the character type and the number of characters of the word, to create a synonym dictionary in which the word may be registered. Find the range.
The synonym dictionary describes the words of the heading and the standard words for the words.

FIG. 12 is a flow chart for explaining the operation of the synonym conversion unit 6. The operation of the synonym conversion unit 6 is shown in FIG.
Follow the instructions below. First, in step ST111, the associative unit 2
The word W to be converted is obtained. Next step ST
At 112, the character type of W is determined and it is set as JS. The types of characters include katakana such as "neural network", kanji such as "neural network", hiragana such as "distortion", alphanumeric characters such as "newal network", and "C language" or " There is a mixture of various character types such as "threshold". Then step S
At T113, the number of W characters is calculated, and is set as CN. For example, when W is “neural network”, the number of characters CN is 5.
Next, in step ST114, the index part is searched using JS and CN as keys, and the index number obtained therefrom is I.
Let N. Then, in step ST115, the line of the index number IN in the synonym dictionary is searched, and the next word number of that line is set to J.
N, the word of the heading is EW, and the standard word is CW. Next, in step ST116, the word W and the heading word E
It is determined whether W is the same, and if they are the same, step S
At T117, the processing ends with CW as the converted word. If they are different, the next word number JN in step ST118.
Determine if there is. For example, when the numerical value at the next word number JN in the synonym dictionary is 0, it is determined that there is no JN. If the next word number JN does not exist, the process ends with the word W converted in step ST119. If there is, JN is set as a new IN in step ST112 and the process returns to step ST115.

Example 4. FIG. 7 is an overall configuration diagram showing an embodiment of the knowledge extracting device according to the invention of claim 4. Reference numeral 7 denotes a synonym relation registration unit that determines whether the word pair obtained from the association unit 2 is a word pair having a new synonym relation that is not registered in the synonym dictionary of the synonym conversion unit 6.

Next, the operation will be described. FIG. 13 is a flowchart for explaining the operation of the synonym relationship registration unit 7.
First, in step ST701, the associative word pairs obtained from the associative unit 2 are defined as the word W1 and the word W2. Next in step ST7
In 02, W1 and W2 are presented to the user, and W1 and W are presented to the user.
It is determined whether or not 2 have a synonymous relationship. Step ST
When the user determines in step 703 that W1 and W2 have a synonymous relationship, in step ST704 W1 and W2 are registered in the synonym dictionary of the synonym conversion unit 6, the index unit is updated, and the process ends. When the user determines in step ST703 that W1 and W2 are not synonymous with each other, the process ends.

Example 5. FIG. 8 is an overall configuration diagram showing an embodiment of the knowledge extraction device according to the invention of claim 5. Reference numeral 8 denotes a synonymous expression conversion unit that converts a concept composed of a plurality of synonymous words into one standard expression.

Next, the operation will be described. FIG. 14 is a configuration example of the synonymous expression conversion unit 8. A segmentation unit 81 divides a concept expression composed of a plurality of words into each constituent element, and the synonym conversion unit 6 divides each constituent element.
Is converted into a standard expression with the synonymous expression synthesizing unit 83, and the constituent elements converted into the standard expression are combined to obtain a standard expression of a concept composed of a plurality of words.

FIG. 15 is a flow chart for explaining the operation of the synonymous expression conversion section 8. First, in step ST801, the associative unit 2 obtains a concept expression CC composed of a plurality of words to be converted. Next, in step ST802, CC
Is divided into components, and each divided component is SC
1 ,. ． ． SCi ,. ． ． SCn. Then, in step ST803, SCi is converted into a synonym from SC1 to SCn, and the standardized expression HCi is obtained. Then, in step ST804, HC1 ,. ． ． HCi ,. ． ． HCN is synthesized and used as the standard expression HCC. An example of an expression of a concept composed of a plurality of words is a reference document such as a technical document. For example, "Yamada, Suzuki: Model for ○○, Theory of Science (B), J62-B, 1, 12-22, 1
The expression "979" refers to the author name (Yamada, Suzuki), title (model for XX), journal name (faculty theory (B)), volume (J62-B), number (1), page (12- 22), the year of issue (1979)
Etc. is divided into each component of the concept of references, and then each component is converted into its standard representation. In this example, for example, the author name (Yamada, Suzuki),
Title (model related to ○○), published journal name (IEICE Transactions (B)), volume (Vol. J6
2-B), number (No. 1), page (PP.12)
-22), is converted to the year of issue (1979). Next, each component is synthesized and "Yamada, Suzuki: model about ○○,
IEICE Transactions (B), Vol. J62-B,
No. 1, PP. 12-22, 1979 "is a standard reference expression.

[0032]

As described above, according to the first aspect of the invention, the knowledge extraction apparatus is provided with a text database for accumulating a large amount of text, an associative unit for associating a concept pair from the text, and the concept pair. Since the knowledge synthesis unit refers to the domain knowledge base to check if there is a relation, it is possible to obtain a concept pair (extract knowledge) for which the relation between concepts is known, and There is an effect that it can be used in another knowledge information processing

According to the second aspect of the invention, in addition to the effect of the first aspect of the invention, the text structure selection portion can detect important parts in the text which can be judged from the structure of the text.
Since the selection is made by the user, there is an effect that new knowledge can be extracted from a part of the text that the user thinks is important.

According to the invention of claim 3, in addition to the effect of the invention of claim 1, the synonym conversion unit is configured to convert a plurality of words having a synonym relationship into one representative word. Therefore, there is an effect that knowledge can be extracted even from words that are not in the domain knowledge base.

According to the invention of claim 4, in addition to the effect of the invention of claim 3, the synonym relation registration unit is configured to add the word determined to be synonymous to the synonym dictionary of the synonym conversion unit. Therefore, synonyms that match the target text can be added to the synonym dictionary, which has the effect of improving the performance of the knowledge extraction device.

According to the invention of claim 5, in addition to the effect of the invention of claim 1, the synonymous expression conversion unit converts a concept composed of a plurality of synonymous words into one standard expression. Since it is configured as described above, there is an effect that knowledge can be extracted for a concept composed of a plurality of words.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of a knowledge extraction device according to the invention of claim 1.

FIG. 2 is a diagram showing an example of a domain knowledge base.

FIG. 3 is an explanatory diagram showing an example of an associative unit.

FIG. 4 is a flowchart illustrating an operation of a knowledge synthesizing unit.

FIG. 5 is an overall configuration diagram showing an embodiment of a knowledge extracting device according to the invention of claim 2;

FIG. 6 is an overall configuration diagram showing an embodiment of a knowledge extracting device according to the invention of claim 3;

FIG. 7 is an overall configuration diagram showing an embodiment of a knowledge extraction device according to the invention of claim 4;

FIG. 8 is an overall configuration diagram showing an embodiment of a knowledge extraction device according to the invention of claim 5;

FIG. 9 is a diagram showing an example of text data having structure information.

FIG. 10 is a flowchart illustrating an operation of a text structure selection unit.

FIG. 11 is an explanatory diagram illustrating an example of a synonym conversion unit.

FIG. 12 is a flowchart illustrating an operation of a synonym conversion unit.

FIG. 13 is a flowchart illustrating an operation of a synonym relationship registration unit.

FIG. 14 is a diagram illustrating a configuration example of a synonymous expression conversion unit.

FIG. 15 is a flowchart illustrating an operation of a synonymous expression conversion unit.

FIG. 16 is a block diagram showing a conventional knowledge extraction device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 text database 2 associative unit 3 domain knowledge base 4 knowledge synthesis unit 5 text structure selection unit 6 synonym conversion unit 7 synonym relation registration unit 8 synonymous expression conversion unit 11 neural network 12 concept input unit 13 associative output unit

Claims (5)

[Claims] 1. A text database for accumulating text data including concepts, and an associative relationship between each concept included in the text data accumulated by the text database is obtained, and concepts having an associative relationship with a certain concept are extracted. And a domain knowledge base that accumulates, as knowledge, concepts that have a predetermined relationship to a certain concept in advance,
A knowledge extracting device comprising: a knowledge synthesizing unit that refers to the knowledge accumulated by the domain knowledge base and extracts a concept having a predetermined relationship with the concept having the associative relationship extracted by the associative unit. 2. A text database that stores text data including a concept, a text structure selection unit that selects a part of the text data stored by the text database, and text data selected by the text structure selection unit. Associative part that obtains the association relation between each concept included in the, and extracts the concept that has the association relation to a certain concept, and the domain knowledge base that accumulates the concepts that have a certain relation to the certain concept as knowledge in advance. And refer to the knowledge accumulated by the above domain knowledge base,
A knowledge extracting device comprising: a knowledge synthesizing unit that extracts a concept having a predetermined relationship with the concept having the associative relationship extracted by the associative unit. 3. A text database for accumulating text data including concepts, a synonym conversion unit for converting synonyms of the concepts contained in the text data accumulated by the text database into standard concepts, and An associative unit that obtains an associative relationship between standard concepts converted by the synonym conversion unit and extracts a concept that has an associative relationship with a concept, and a concept that has a predetermined relationship with the concept in advance. A domain knowledge base that accumulates as knowledge, and a knowledge synthesizing unit that refers to the knowledge accumulated by the domain knowledge base and extracts a concept having a predetermined relationship with the concept having the associative relationship extracted by the associative unit. Knowledge extraction device. 4. A text database that stores text data including a concept, a synonym conversion unit that converts synonyms of the concepts contained in the text data stored by the text database into standard concepts, and An associative unit that obtains an associative relationship between standard concepts converted by the synonym conversion unit and extracts a concept that is in an associative relationship with a certain concept, and a result that is associated by the associative unit as a result When a pair has a synonymous relationship, a synonym relation registration unit that registers the concept pair in the synonym conversion unit, a domain knowledge base that stores in advance a concept having a predetermined relationship to a certain concept as knowledge, A knowledge synthesizing unit that refers to the knowledge accumulated by the domain knowledge base and extracts a concept having a predetermined relationship with the concept having the associative relationship extracted by the associative unit. Knowledge extractor equipped with and. 5. A text database for accumulating text data containing a concept composed of a plurality of words, and a synonym for converting synonyms of the words constituting the concept accumulated by the text database into a standard concept. An expression conversion unit, an association unit that obtains an associative relationship between standard concepts converted by the synonymous expression conversion unit, and extracts a concept that is in an associative relationship with a certain concept; Domain knowledge base for accumulating concepts in the relationship of knowledge as knowledge, and knowledge accumulated by the domain knowledge base, to refer to a concept having a predetermined relationship with the concept in the associative relationship extracted by the associative unit. A knowledge extracting device having a knowledge synthesizing unit for extracting.