Query disambiguation in a question-answering environment

• US 10,169,490 B2
• Filed: 09/24/2015
• Issued: 01/01/2019
• Est. Priority Date: 03/02/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Allen, Corville O., Kohler, Claire L.
• Primary Examiner(s): Featherstone, Mark D
• Assistant Examiner(s): Doraiswamy, Ranjit P
• Application Number:
  US14/864,218
• Publication Number:
  US 20160259863A1
• Time in Patent Office: 1,195 Days
• Field of Search: None
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/90332 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars

Abstract

Query disambiguation can include identifying, a first set of word clusters in a first candidate answer generated in response to a first input query for a question-answering system. Each of the first set of word clusters can be classified according to a syntactic element of a first set of syntactic elements. Query disambiguation can include identifying, in the first input query, a second set of word clusters. Each of the second set of word clusters can be classified according to a syntactic element of a second set of syntactic elements. Query disambiguation can include determining that the first word cluster is classified according to a first syntactic element that is not present in the second set of syntactic elements and generating a second input query using the first input query and the first word cluster.

Claims

1. A computer-implemented method of query disambiguation in a question-answering environment, the method comprising:

identifying, using a first natural language processing technique, a first set of word clusters in a first candidate answer of a plurality of candidate answers generated in response to a first input query for a question-answering system, the first set of word clusters including a first word cluster, each of the first set of word clusters classified according to a semantic term of a first set of semantic terms and a syntactic element of a first set of syntactic elements, wherein one or more syntactic elements of the first set of syntactic elements are a first set of parts of speech and the identification of the first set of word clusters includes tagging the first set of semantic terms and the first set of syntactic elements;

identifying, using the first natural language processing technique, in the first input query, a second set of word clusters, each of the second set of word clusters classified according to a semantic term of a second set of semantic terms and a syntactic element of a second set of syntactic elements, wherein one or more syntactic elements of the second set of syntactic elements are a second set of parts of speech and the identification of the second set of word clusters includes tagging the second set of semantic terms and the second set of syntactic elements;

determining, by comparing the first and second sets of semantic terms and the first and second set of syntactic elements, respectively, that the first word cluster is classified according to a first semantic term and a first syntactic element that is not present in the second set of semantic terms and the second set of syntactic elements, respectively;

determining, using a second natural language processing technique, a set of domains from the first set of word clusters, each of the set of domains corresponding to a topic, and wherein at least one domain of the set of domains is an ontological domain;

determining a number of domains in the set of domains;

determining an amalgamated confidence value for the first candidate answer; and

wherein the determining that the first word cluster is classified according to a syntactic element not present in the second set of syntactic elements, and generating the clarification query in response to determining that the amalgamated confidence value achieves a first ambiguity criterion and the number of domains achieves a second ambiguity criterion, wherein the second ambiguity criterion is based on at least a ratio of the number of domains to the number of candidate answers in the plurality of candidate answers; and

generating, using a third natural language processing technique, a user-display clarification query using the first input query and the first word cluster; and

providing, using a display, the user input query to a user that provided the first input query.

2. The method of claim 1, wherein the first word cluster is classified according to a second syntactic element and a second word cluster of the second set of word clusters is classified according to a third syntactic element, and wherein the method further comprises:

determining that the first syntactic element and the second syntactic element achieve a similarity criterion; and

generating, by replacing the second word cluster with the first word cluster, a third input query, the third input query generated in response to determining that the first and second syntactic elements achieve the similarity criterion.

3. The method of claim 1, wherein the first natural language processing technique and second natural language processing technique include using a slot grammar parser.

4. The method of claim 1, wherein the clarification query includes a plurality of word clusters arranged in a first order, and wherein the method further comprises:

determining, based on the first order of the plurality of word clusters, a set of input query attributes for the clarification query, the set of input query attributes including at least a first threshold for comparing the parts of speech and a second threshold for comparing the order of syntactic elements among the plurality of word clusters;

determining, by comparing the set of input query attributes to a set of input query criteria, that the set of input query attributes does not satisfy the set of input query criteria; and

generating, in response to determining that the set of input query attributes does not satisfy the set of input query criteria, a third input query including the plurality of word clusters arranged in a second order.

5. The method of claim 4, wherein the clarification query including the plurality of word clusters arranged in the first order is structured to have a passive voice and the third input query including the plurality of word clusters arranged in the second order is structured to have an active voice.

6. The method of claim 1, further comprising:

displaying, using a graphical user interface, the clarification query;

identifying, using a visual emphasis in the graphical user interface, the first word cluster;

toggling, between the clarification query and the first input query, in response to receiving one or more user inputs via the graphical user interface, the one or more user inputs corresponding to the visual emphasis of the first word cluster.

7. The method of claim 1, further comprising:

receiving the first input query for the question-answering system;

analyzing, using a natural language processing technique, a corpus;

identifying, based on analyzing the corpus, a set of candidate answers for the first input query, the set of candidate answers including the first candidate answer.

8. The method of claim 1, wherein the each candidate answer of the plurality of candidate answers is color coded according to the corresponding amalgamated confidence value.

9. The method of claim 1, wherein syntactic elements further comprise word cluster size.

10. The method of claim 1, wherein the question-answering system employs a restricted syntax to receive the first input query.

11. The method of claim 1, wherein the question-answering system employs a RESTful interface to receive the first input query.

12. The method of claim 1, wherein the user-display clarification query includes candidate answers represented in audio and image formats.

13. The method of claim 1, further comprising, in response to receiving a user interaction with the clarification query, weighting the calculation of confidence values, based on the user interaction.

14. The method of claim 1, wherein the first input query includes audio data.

15. The method of claim 1, wherein the first input query includes image data.

16. The method of claim 1, wherein the first input query includes video data.