Question answering framework

• US 9,213,771 B2
• Filed: 06/06/2013
• Issued: 12/15/2015
• Est. Priority Date: 06/04/2013
• Status: Active Grant

• Current Assignee: SAP SE
• Original Assignee: SAP SE
• Inventors: Chen, Yingyu, Dong, Bin, Shen, Yongyuan, Liu, Xiaowei, Wei, Zhenglong
• Primary Examiner(s): Arjomandi, Noosha
• Application Number:
  US13/911,088
• Publication Number:
  US 20140358890A1
• Time in Patent Office: 922 Days
• Field of Search: 707/710
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation

Abstract

Described herein is a technology to facilitate automated question answering. In one implementation, an input question is first received. Different search strategies may be used to search multiple types of data from multiple types of knowledge databases to generate one or more candidate answers to the input question. The one or more candidate answers are evaluated to generate a final answer to the input question.

Claims

1. A computer-implemented question-answering method, comprising:

receiving an input question;

determining multiple types of knowledge databases available for searching, wherein the multiple types of knowledge databases provide data in different formats and include a question-answer paired knowledge database, a plain text knowledge database, a resource description framework (RDF) knowledge database or a combination thereof;

when the question-answer paired knowledge database is available, searching question-answer paired data from the question-answer paired knowledge database to determine a first candidate answer to the input question;

when the plain text knowledge database is available, searching plain text data from the plain text knowledge database to determine a second candidate answer to the input question;

when the resource description framework (RDF) knowledge database is available, searching RDF data from the RDF knowledge database to determine a third candidate answer to the input question; and

evaluating the first, second or third candidate answer to generate a final answer to the input question.

2. The method of claim 1 further comprising parsing the input question and identifying one or more properties associated with the input question.

3. The method of claim 2 wherein identifying the one or more properties includes identifying a question class or a question sub-class.

4. The method of claim 2 wherein identifying the one or more properties includes identifying an expected answer type.

5. The method of claim 2 wherein identifying the one or more properties includes identifying a focus.

6. The method of claim 2 wherein identifying the one or more properties includes identifying a keyword.

7. The method of claim 1 wherein searching the question-answer paired data from the question-answer paired knowledge database comprises:

generating a query statement based on one or more properties associated with the input question, wherein the query statement is executable to perform a full text search of the question-answer paired data to generate search results.

8. The method of claim 7 wherein the full text search comprises a fuzzy search based on an n-gram algorithm.

9. The method of claim 7 further comprising ranking the search results to determine at least one of the one or more candidate answers.

10. The method of claim 1 wherein searching the plain text data from the one plain text knowledge database comprises:

generating a query statement based on a mining set of search terms associated with the input question, wherein the query statement is executable to perform a search of the plain text data to generate search results.

11. The method of claim 10 further comprising iteratively refining the search by broadening or narrowing search criteria until a number of the search results meets a pre-specified threshold.

12. The method of claim 10 further comprising constructing, using an answer template, at least one of the one or more candidate answers based on the search results.

13. The method of claim 12 further comprising reducing a level of detail of the candidate answer by using a machine learning framework.

14. The method of claim 1 wherein evaluating the one or more candidate answers to generate a final answer to the input question comprises ranking the one or more candidate answers based on one or more pre-defined heuristic rules.

15. A non-transitory computer-readable medium having stored thereon program code, the program code executable by a computer to:

receive an input question;

identify multiple types of knowledge databases available for searching, wherein the multiple types of knowledge databases provide data in different formats and include a question-answer paired knowledge database, a plain text knowledge database, a resource description framework (RDF) knowledge database or a combination thereof;

when the question-answer paired knowledge database is available, search question-answer paired data from the question-answer paired knowledge database to determine a first candidate answer to the input question;

when the plain text knowledge database is available, search plain text data from the plain text knowledge database to determine a second candidate answer to the input question;

when the resource description framework (RDF) knowledge database is available, search RDF data from the RDF knowledge database to determine a third candidate answer to the input question; and

evaluate the first, second or third candidate answer to generate a final answer to the input question.

16. A system comprising:

a non-transitory memory device for storing computer readable program code; and

a processor in communication with the memory device, the processor being operative with the computer readable program code to;

receive an input question;

identify multiple types of knowledge databases available for searching, wherein the multiple types of knowledge databases provide data in different formats and include a question-answer paired knowledge database, a plain text knowledge database, a resource description framework (RDF) knowledge database or a combination thereof;

when the question-answer paired knowledge database is available, search question-answer paired data from the question-answer paired knowledge database to determine a first candidate answer to the input question;

when the plain text knowledge database is available, search plain text data from the plain text knowledge database to determine a second candidate answer to the input question;

when the resource description framework (RDF) knowledge database is available, search RDF data from the RDF knowledge database to determine a third candidate answer to the input question; and

evaluate the first, second or third candidate answer to generate a final answer to the input question.