Generating Answer Variants Based on Tables of a Corpus

Patent metadata

Field	Value
Publication number	US-2017351677-A1
Application number	US-201615172216-A
Country	US
Kind code	A1
Filing date	Jun 3, 2016
Priority date	Jun 3, 2016
Publication date	Dec 7, 2017
Grant date	—

Abstract

Mechanisms are provided for implementing a candidate answer variant engine. The mechanisms receive an indication of a structured portion of content in a corpus, generate a plurality of groupings of elements of the structured portion of content, and generate, for each grouping of elements in the plurality of groupings of elements, and for each element in the grouping of elements, a corresponding grouping vector representation, corresponding to the element. The mechanisms, for each grouping vector representation of each grouping of elements in the plurality of groupings of elements perform a similarity measure calculation between the grouping vector representation and a vector representation of an input question, and select an element corresponding to the grouping vector representation for inclusion as a candidate answer variant based on results of the similarity measure calculation. The mechanisms perform question answering operations based on an analysis of one or more candidate answer variants.

First claim

What is claimed is:

1 . A method, in a data processing system comprising a processor and a memory, the memory comprising instructions executed by the processor to cause the processor to implement a candidate answer variant engine, the method comprising:
receiving, by the candidate answer variant engine, an indication of a structured portion of content in a corpus; generating, by the candidate answer variant engine, a plurality of groupings of elements of the structured portion of content; generating, by the candidate answer variant engine, for each grouping of elements in the plurality of groupings of elements, and for each element in the grouping of elements, a corresponding grouping vector representation, corresponding to the element; for each grouping vector representation of each grouping of elements in the plurality of groupings of elements:
performing, by the candidate answer variant engine, a similarity measure calculation between the grouping vector representation and a vector representation of an input question; and
selecting, by the candidate answer variant engine, an element corresponding to the grouping vector representation for inclusion as a candidate answer variant in a candidate answer variant data structure based on results of the similarity measure calculation; and
performing question answering operations for answering the input question based on an analysis of one or more candidate answer variants in the candidate answer variant data structure.

2 . The method of claim 1 , wherein the structured portion of content comprises a table data structure present in a portion of content of the corpus, and wherein each grouping of elements in the plurality of grouping of elements comprises a triad of at least one row header, at least one column header, and a cell content.

3 . The method of claim 1 , wherein the structured portion of content is a structured portion of content corresponding to a candidate answer found in the corpus by a primary search operation of a question answering (QA) system.

4 . The method of claim 1 , wherein performing the similarity measure calculation comprises calculating a cosine similarity between the grouping vector representation and the vector representation of the input question.

5 . The method of claim 4 , wherein the similarity measure calculation comprises calculating the cosine similarity between other elements of the grouping vector representation than an element of the grouping vector representation being considered for selection as a candidate answer variant from the grouping of elements.

6 . The method of claim 5 , wherein selecting the element corresponding to the grouping vector representation for inclusion as a candidate answer variant in the candidate answer variant data structure based on results of the similarity measure calculation comprises:
comparing a cosine similarity value calculated as the cosine similarity between the other elements of the grouping vector representation and the vector representation of the input question, with a threshold value; and selecting the element being considered for selection as a candidate answer variant to be a candidate answer variant in response to the cosine similarity value having a predefined relationship with the threshold value.

7 . The method of claim 1 , Wherein generating a plurality of groupings of elements of the structured portion of content comprises:
analyzing at least one of metadata or computer code associated with the structured portion of content to identify a structure of the structured portion of content; identifying elements of the structure based on the analysis; and generating groupings of elements based on the identified elements and the identified structured of the structured portion of content.

8 . The method of claim 1 , wherein the vector representation of the input question comprises a vector representation of one or more features of the input question identified by a natural language processing of the input question.

9 . The method of claim 1 , wherein the similarity measure calculation comprises at least one of a textual syntactic similarity measure calculation or textual semantic similarity measure calculation.

10 . The method of claim 1 , wherein performing question answering operations for answering the input question based on an analysis of one or more candidate answer variants in the candidate answer variant data structure comprises:
generating, for each candidate answer variant in the candidate answer variant data structure, a confidence score value indicating a confidence that a corresponding candidate answer variant is a correct answer for the input question; ranking the candidate answer variants relative to one another, and to other candidate answers found during a primary search operation of a question answering (QA) system, based on the generated confidence scores to generate a ranked listing data structure; selecting at least one final answer from the ranked listing data structure; and outputting the at least one final answer to a source of the input question.

11 . A computer program product comprising a non-transitory computer readable medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:
receive an indication of a structured portion of content in a corpus; generate a plurality of groupings of elements of the structured portion of content; generate, for each grouping of elements in the plurality of groupings of elements, and for each element in the grouping of elements, a corresponding grouping vector representation, corresponding to the element, of each of the other elements in the grouping of elements; for each grouping vector representation of each grouping of elements in the plurality of groupings of elements:
perform a similarity measure calculation between the grouping vector representation and a vector representation of an input question; and
select an element corresponding to the grouping vector representation for inclusion as a candidate answer variant in a candidate answer variant data structure based on results of the similarity measure calculation; and
perform question answering operations for answering the input question based on an analysis of one or more candidate answer variants in the candidate answer variant data structure.

12 . The computer program product of claim 11 , wherein the structured portion of content comprises a table data structure present in a portion of content of the corpus, and wherein each grouping of elements in the plurality of grouping of elements comprises a triad of at least one row header, at least one column header, and a cell content.

13 . The computer program product of claim 11 , wherein the structured portion of content is a structured portion of content corresponding to a candidate answer found in the corpus by a primary search operation.

14 . The computer program product of claim 11 , wherein performing the similarity measure calculation comprises calculating a cosine similarity between the grouping vector representation and the vector representation of the input question.

15 . The computer program product of claim 14 , wherein the similarity measure calculation comprises calculating the cosine similarity between other elements of the grouping vector representation than an element of the grouping vector representation being considered for selection as a candidate answer variant from the grouping of elements.

16 . The computer program product of claim 15

Assignees

• IBM

Inventors

• Chaubal Amrish V
• Eyigoz Kadriye E
• Santos Moraes Priscilla
• Sinha Ravi S

Classifications

• G06F16/285

  Clustering or classification · CPC title

• G06F16/248

  Presentation of query results · CPC title

• G06F16/345

  Summarisation for human users · CPC title

• G06F16/24578 Primary

  using ranking · CPC title

• G06F16/3344 Primary

  using natural language analysis · CPC title