INFORMATION EXTRACTION DEVICE AND INFORMATION EXTRACTION SYSTEM

• US 20110202545A1
• Filed: 01/06/2009
• Published: 08/18/2011
• Est. Priority Date: 01/07/2008
• Status: Abandoned Application

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Kawai, Takao, Ando, Shinichi
• Application Number:
  US12/811,861
• Publication Number:
  US 20110202545A1
• Time in Patent Office: Days
• US Class Current: 707/755
• CPC Class Codes: G06F 16/24564 Applying rules; Deductive queries
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition

Abstract

The information extraction device for extracting specific information using information extraction rules comprises a case candidate extraction means for extracting new specific information that is not extracted by the information extraction rules as novel case candidates based on extraction results obtained from extraction target text data; a rule candidate generation means for generating multiple extraction rule candidates based on the novel case candidates; a relation analysis means for analyzing the derivational relation between the novel case candidates and the extraction rule candidates and the overlapping relation between the multiple extraction rule candidates to generate relation analysis results; and a case candidate selection means for calculating the priorities of the novel case candidates based on the relation analysis results and previously prepared case information and selecting the novel case candidates according to the priority.

Claims

1. An information extraction device for extracting specific information using information extraction rules, comprising:

a case candidate extraction unit for extracting new specific information that is not extracted by said information extraction rules as novel case candidates based on extraction results obtained from extraction target text data;

a rule candidate generation unit for generating multiple extraction rule candidates based on said novel case candidates;

a relation analysis unit for analyzing the derivational relation between said novel case candidates and said extraction rule candidates and the overlapping relation between said multiple extraction rule candidates to generate relation analysis results; and

a case candidate selection unit for calculating the priorities of said novel case candidates based on said relation analysis results and previously prepared case information and selecting said novel case candidates according to the priority.

2. The information extraction device according to claim 1 wherein said case candidate extraction unit generates extraction conditions for extracting said novel case candidates from said text data based on said extraction results.

3. The information extraction device according to claim 2 wherein said extraction conditions are attribute values of one or multiple morphemes to which a character string obtained as said extraction results applies or a combination of such attribute values.

4. The information extraction device according to claim 3 wherein said case information includes correct/incorrect information indicating whether or not the content of case information is information suitable for extraction;

and

said case candidate extraction unit excludes a specified portion in said text data from said novel case candidates when the specified portion conforms with any case information of which the correct/incorrect information is incorrect.

5. The information extraction device according to claim 1 wherein said rule candidate generation unit generates said derivational relation by associating said novel case candidates with each of said generated extraction rule candidates.

6. The information extraction device according to claim 4 wherein said overlapping relation is a relation indicating whether or not at least part of the extraction results by one extraction rule candidate includes the extraction results by the other extraction rule candidate;

and

said information extraction device further comprises an information extraction unit associating the extraction results extracted from said text data according to said extraction rule candidates given by said rule candidate generation unit with each of said extraction rule candidates to generate said overlapping relation.

7. The information extraction device according to claim 6 wherein said relation analysis unit generates relation network information linking between said novel case candidates and extraction rule candidates satisfying said derivational relation and between said extraction rule candidates satisfying said overlapping relation.

8. The information extraction device according to claim 7 wherein said relation network information includes a first set consisting of multiple extraction rule candidates satisfying said derivational relation and overlapping relation;

and

said case candidate selection unit generates a second set by excluding the extraction rule candidates of which the extraction results extracted by said information extraction unit conform with any case information of which said correct/incorrect information is incorrect from multiple extraction rule candidates included in said first set, and calculates said priorities using said second set.

9. The information extraction device according to claim 8 wherein said case candidate selection unit calculates said priorities using the number of said extraction rule candidates or the number of extraction results extracted from said text data according to said extraction rule candidates in said second set.

10. The information extraction device according to claim 8 wherein said case candidate selection unit calculates said priorities using the number of links or the largest number of passing links in said second set.

11. An information extraction system comprising an information extraction device connected to a user terminal via communication lines for extracting specific information using information extraction rules, wherein

said information extraction device comprises:

a case candidate extraction unit for extracting new specific information that is not extracted by said information extraction rules as novel case candidates based on extraction results obtained from extraction target text data;

a rule candidate generation unit for generating multiple extraction rule candidates based on said novel case candidates;

a relation analysis unit for analyzing the derivational relation between said novel case candidates and said extraction rule candidates and the overlapping relation between said multiple extraction rule candidates to generate relation analysis results;

a case candidate selection unit for calculating the priorities of said novel case candidates based on said relation analysis results and previously prepared case information and selecting said novel case candidates according to the priority; and

a case candidate inquiry unit for inquiring of said user terminal about the correct/incorrect of novel case candidates selected by said case candidate selection unit and giving the determination results from said user terminal to said case candidate selection unit;

said case candidate selection unit determines the correct/incorrect of said selected novel case candidates based on said determination results given by said case candidate inquiry unit.

12. An information extraction method for extracting specific information using information extraction rules, comprising the flowing steps:

extracting new specific information that is not extracted by said information extraction rules as novel case candidates based on extraction results obtained from extraction target text data;

generating multiple extraction rule candidates based on said novel case candidates;

analyzing the derivational relation between said novel case candidates and said extraction rule candidates and the overlapping relation between said multiple extraction rule candidates to generate relation analysis results; and

calculating the priorities of said novel case candidates based on said relation analysis results and previously prepared case information and selecting said novel case candidates according to the priority.

13. The information extraction method according to claim 12 wherein said case information includes correct/incorrect information indicating whether or not the content of case information is information suitable for extraction;

and

in said step of extracting, a specified portion in said text data is excluded from said novel case candidates when the specified portion conforms with any case information of which the correct/incorrect information is incorrect.

14. The information extraction method according to claim 13 wherein in said step of generating relation analysis results, relation network information linking between said novel case candidates and extraction rule candidates satisfying said derivational relation and between said extraction rule candidates satisfying said overlapping relation is generated;

and

said relation network information includes a first set consisting of multiple extraction rule candidates satisfying said derivational relation and overlapping relation, and

in said step of selecting said novel case candidates, a second set is generated by excluding the extraction rule candidates of which the extraction results include any case information of which said correct/incorrect information is incorrect from multiple extraction rule candidates included in said first set and said second set is used to calculate said priorities.

15. The information extraction method according to claim 12 wherein said information extraction method further comprises the following steps:

inquiring of a user terminal about the correct/incorrect determination of said selected novel case candidates; and

receiving the determination results indicating said correct/incorrect determination from said user terminal and determining the correct/incorrect of said selected novel case candidates based on said determination results.

16. A recording medium storing an information extraction program for an information extraction device provided with a computer and extracting specific information using information extraction rules, wherein said program allows said computer to perform the following procedures:

extracting new specific information that is not extracted by said information extraction rules as novel case candidates based on extraction results obtained from extraction target text data;

generating multiple extraction rule candidates based on said novel case candidates;

analyzing the derivational relation between said novel case candidates and said extraction rule candidates and the overlapping relation between said multiple extraction rule candidates to generate relation analysis results; and

calculating the priorities of said novel case candidates based on said relation analysis results and previously prepared case information and selecting said novel case candidates according to the priority.

17. The recording medium according to claim 16 wherein said case information includes correct/incorrect information indicating whether or not the content of case information is information suitable for extraction;

and

in said procedure of extracting, a specified portion in said text data is excluded from said novel case candidates when the specified portion conforms with any case information of which the correct/incorrect information is incorrect.

18. The recording medium according to claim 17 wherein in said procedure of generating relation analysis results, relation network information linking between said novel case candidates and extraction rule candidates satisfying said derivational relation and between said extraction rule candidates satisfying said overlapping relation is generated;

and

said relation network information includes a first set consisting of multiple extraction rule candidates satisfying said derivational relation and overlapping relation, and

in said procedure of selecting said novel case candidates, a second set is generated by excluding the extraction rule candidates of which the extraction results include any case information of which said correct/incorrect information is incorrect from multiple extraction rule candidates included in said first set, and said second set is used to calculate said priorities.

19. The recording medium according to claim 16 wherein said program further allows said computer to perform the following procedures:

inquiring of a user terminal about the correct/incorrect determination of said selected novel case candidates; and

receiving the determination results indicating said correct/incorrect determination from said user terminal and determining the correct/incorrect of said selected novel case candidates based on said determination results.