Providing answers to questions using multiple models to score candidate answers

Patent metadata

Field	Value
Publication number	US-10823265-B2
Application number	US-201815982855-A
Country	US
Kind code	B2
Filing date	May 17, 2018
Priority date	Sep 28, 2010
Publication date	Nov 3, 2020
Grant date	Nov 3, 2020

Abstract

A method, system and computer program product for generating answers to questions. In one embodiment, the method comprises receiving an input query; conducting a search to identify candidate answers to the input query, and producing a plurality of scores for each of the candidate answers. For each of the candidate answers, one, of a plurality of candidate ranking functions, is selected. This selected ranking function is applied to the each of the candidate answers to determine a ranking for the candidate answer based on the scores for that candidate answer. One or more of the candidate answers is selected, based on the rankings for the candidate answers, as one or more answers to the input query. In an embodiment, the ranking function selection is performed using information about the question. In an embodiment, the ranking function selection is performed using information about each answer.

First claim

The invention claimed is:

1. A system for generating answers to questions, comprising:
a computer device comprising:
at least one distinct software module, each distinct software module being embodied on a tangible computer-readable medium;
a memory; and
at least one processor coupled to the memory device and operative for:
receiving an input query;
obtaining, from passages in a data source, a plurality of candidate answers to the input query, each of the candidate answers being detected in one or more of the passages;
performing answer processing to produce first scores for each of the candidate answers;
performing context dependent answer processing to produce candidate answer scores for the candidate answers, the candidate answer score for each of the candidate answers being dependent on content of the passages having the candidate answer, including
for each of the candidate answers, identifying one or more passages in the data source having the each candidate answer, computing specified information about said each candidate answer based on the content of the one or more passages having said each candidate answer, sending the each candidate answer to a model selection module, the model selection module using said computed specified information about the each candidate answer to select one of a plurality of scoring models, sending the each candidate answer to the selected one of the scoring models, and using the selected one of the scoring models for weighting the first score for the each candidate answer to determine the answer score for the each candidate answer; and

generating at least one answer to the input query based on the answer scores for the candidate answers.

2. The system according to claim 1 , wherein the computing specified information about each candidate answer includes comparing the query and the one or more passages having the each candidate answer.

3. The system according to claim 2 , wherein the using the selected one of the scoring models for weighting the first score for the each candidate answer to determine the answer score for the each candidate answer includes basing the answer score on text alignment between the input query and the one or more passages in the data source having the each candidate answer.

4. The system according to claim 2 , wherein the using the selected one of the scoring models for weighting the first score for the each candidate answer to determine the answer score for the each candidate answer includes basing the answer score on grammatical relations between the input query and the one or more passages in the data source having the each candidate answer.

5. The system according to claim 2 , wherein the using the selected one of the scoring models for weighting the first score for the each candidate answer to determine the answer score for the each candidate answer includes basing the answer score on lexical relations between the input query and the one or more passages in the data source having the each candidate answer.

6. The system according to claim 2 , wherein the using the selected one of the scoring models for weighting the first score for the each candidate answer to determine the answer score for the each candidate answer includes basing the answer score on semantic relations between the input query and the one or more passages in the data source having the each candidate answer.

7. The system according to claim 1 , wherein the computing specified information about the each candidate answer includes computing the specified information about said each candidate answer before sending the candidate answers to the model selection module.

8. A method for generating answers to questions, comprising:
receiving an input query;
obtaining, from passages in a data source, a plurality of candidate answers to the input query, each of the candidate answers being detected in one or more of the passages;
performing answer processing to produce first scores for each of the candidate answers;
performing context dependent answer processing to produce candidate answer scores for the candidate answers, the candidate answer score for each of the candidate answers being dependent on content of the passages having the candidate answer, including
for each of the candidate answers, identifying one or more passages in the data source having the each candidate answer, computing specified information about said each candidate answer based on the content of the one or more supporting passages having said each candidate answer, sending the each candidate answer to a model selection module, the model selection module using said computed specified information about the each candidate answer to select one of a plurality of scoring models, sending the each candidate answer to the selected one of the scoring models, and using the selected one of the scoring models for weighting the first score for the each candidate answer to determine the answer score for the each candidate answer; and

generating at least one answer to the input query based on the answer scores for the candidate answers.

9. The method according to claim 8 , wherein the computing specified information about said each candidate answer based on the content of the one or more supporting passages having said each candidate answer includes comparing the input query and said one or more supporting passages.

10. The method according to claim 9 , wherein the computed specified information is based on text alignment of the input query and the one or more supporting passages.

11. The method according to claim 9 , wherein the computed specified information is based on grammatical relations of the input query and the one or more supporting passages.

12. The method according to claim 9 , wherein the computed specified information is based on lexical relations of the input query and the one or more supporting passages.

13. The method according to claim 9 , wherein the computed specified information is based on semantic relations of the input query and the one or more supporting passages.

14. The method according to claim 8 , wherein the computing specified information about the each candidate answer includes computing the specified information about said each candidate answer before sending the candidate answers to the model selection module.

15. An article of manufacture, comprising:
at least one tangible computer readable device having computer readable program code logic tangibly embodied therein to generate answers to questions, the computer readable program code logic, when executing, performing the following:
receiving an input query;
obtaining, from passages in a data source, a plurality of candidate answers to the input query, each of the candidate answers being detected in one or more of the passages;
performing answer processing to produce first scores for each of the candidate answers;
performing context dependent answer processing to produce candidate answer scores for the candidate answers, the candidate answer score for each of the candidate answers being dependent on content of the passages having the candidate answer, including
for each of the candidate answers, identifying one or more passages in the data source having the each candidate answer, computing specified information about said each candidate answer based on the content of the one or more supporting passages having said each candidate answer, sending the each candidate answer to a model selection module, the model selection module using said computed specified information about the each candidate answer to select one of a plurality of scoring models,

Assignees

• IBM

Inventors

• Brown Eric W
• Ferrucci David
• Murdock Iv James W

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G06F16/245

  Query processing · CPC title

• F16H37/02 Primary

  comprising essentially only toothed or friction gearings · CPC title

• G06F16/334

  Query execution (filtering based on additional data G06F16/335) · CPC title

• F16H2200/2005

  with one sets of orbital gears · CPC title