EVIDENCE DIFFUSION AMONG CANDIDATE ANSWERS DURING QUESTION ANSWERING

• US 20130018652A1
• Filed: 09/14/2012
• Published: 01/17/2013
• Est. Priority Date: 09/28/2010
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Ferrucci, David A., Gondek, David C., Kalyanpur, Aditya A., Lally, Adam P.
• Granted Patent: US 8,738,365 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 16/24564 Applying rules; Deductive queries
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems

Abstract

Diffusing evidence among candidate answers during question answering may identify a relationship between a first candidate answer and a second candidate answer, wherein the candidate answers are generated by a question-answering computer process, the candidate answers have associated supporting evidence, and the candidate answers have associated confidence scores. All or some of the evidence may be transferred from the first candidate answer to the second candidate answer based on the identified relationship. A new confidence score may be computed for the second candidate answer based on the transferred evidence.

Claims

1. A computer readable storage medium storing a program of instructions executable by a machine to perform a method of diffusing evidence among candidate answers during question answering, comprising:

identifying, by a processor, a relationship between a first candidate answer and a second candidate answer, wherein the candidate answers are generated by a question-answering computer process, the candidate answers have associated supporting evidence, and the candidate answers have associated confidence scores;

transferring all or some of the evidence from the first candidate answer to the second candidate answer based on the identified relationship; and

computing a new confidence score for the second candidate answer based on the transferred evidence.

2. The computer readable storage medium of claim 1, wherein the transferring includes merging the transferred evidence with existing evidence information associated with the second candidate answer.

3. The computer readable storage medium of claim 1, wherein the transferring includes adding new evidence information denoting direction or strength, or both direction and strength, of the identified relationship.

4. The computer readable storage medium of claim 1, wherein the transferring includes transferring feature scores across candidates.

5. The computer readable storage medium of claim 1, wherein transferring further includes determining based on heuristics whether to transfer entire feature score as-is, or a fraction of the feature score based on type and strength of the identified one or more relationships.

6. The computer readable storage medium of claim 1, wherein the re-ranking step is performed by machine learning.

7. The computer readable storage medium of claim 1, wherein the identifying includes consulting a structured knowledge base and mapping strings appearing in the candidate answers.

8. The computer readable storage medium of claim 1, wherein the identifying step is based on lexical knowledge extracted from unstructured text.

9. A system for diffusing evidence among candidate answers during question answering, comprising:

a processor; and

an evidence diffusion module operable to identify a relationship between a first candidate answer and a second candidate answer, wherein the candidate answers are generated by a question-answering computer process, the candidate answers have associated supporting evidence, and the candidate answers have associated confidence scores, the evidence diffusion module further operable to transfer all or some of the evidence from the first candidate answer to the second candidate answer based on the identified relationship, the evidence diffusion module further operable to compute a new confidence score for the second candidate answer based on the transferred evidence.

10. The system of claim 9, wherein direction of transferring between one candidate answer and another candidate answer is controlled based on a question associated with the candidate answers.

11. The system of claim 9, wherein the evidence diffusion module transfers evidence by merging the transferred evidence with existing evidence information associated with the second candidate answer.

12. The system of claim 9, wherein the evidence diffusion module transfers evidence by adding a new evidence information denoting direction or strength, or both direction and strength, of the identified one or more relationships

13. The system of claim 9, wherein the evidence diffusion module transfers evidence by transferring feature scores across candidates.