Providing answers to questions using hypothesis pruning
A method, system and computer program product for generating answers to questions. In one embodiment, the method comprises receiving a query, conducting a search through one or more data sources to identify candidate answers to the query, and providing each of the candidate answers with a preliminary score. The method further comprises filtering out any of the candidate answers with a preliminary score that does not satisfy a defined condition. The candidate answers having preliminary scores that satisfy this condition form a subset of the candidate answers. Each of the candidate answers in this subset is processed to produce further scores. A ranking function is applied to these further scores to determine a ranking for each of the candidate answers in the subset; and after this ranking function is applied, one or more of the candidate answers are selected as one or more final answers to the query.
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 and operative for:
receiving an input query;
conducting a search through one or more data sources to identify a plurality of candidate answers to the input query, wherein each of the candidate answers is found in an associated supporting passage having identified content;
providing each of the candidate answers with a content independent preliminary score based on one or more defined criterion and independent of the content of the supporting passage in which said each candidate answer was found;
filtering out any of the candidate answers with a content independent preliminary score that does not satisfy a defined condition, wherein the candidate answers having content independent preliminary scores that satisfy said defined condition form a subset of the candidate answers;
for each of the candidate answers in said subset, processing said each of the candidate answers to produce a plurality of further scores for said each of the candidate answers, said further scores including a content dependent score based on the content of the supporting passage in which said each candidate answer was found;
applying a candidate ranking function to said plurality of further scores to determine a ranking for said each of the candidate answers in said subset; and
after the applying the candidate ranking function, selecting one or more of the candidate answers as one or more final answers to the input query; and wherein:
the computer device includes a pre-filter context independent candidate answer processing module, and a hypothesis filtering module;
the pre-filter context independent candidate answer processing module performs the providing each of the candidate answers with a content independent preliminary score; and
the hypothesis filtering module receives the content independent preliminary scores from the pre-filter context independent candidate answer processing module and filters the candidate answers based on said content independent preliminary scores.
2. The system according to claim 1 , wherein:
the processing said each of the candidate answers in the subset of candidate answers includes using a supporting passage retrieval process to identify passages in which said each of the candidate answers occur; and
the applying the candidate ranking function includes sending said subset of candidate answers from the supporting passage retrieval process to a candidate ranking function module; and
the computer device is further operative for sending any filtered out candidate answers to the candidate ranking function, including sending said any filtered our candidate answers to the candidate ranking module in a manner by-passing the supporting passage retrieval process; and
wherein the filtering out includes filtering out any of the candidate answers with a content independent preliminary score below a defined threshold, and said defined threshold is established by running a test on a given set of data to obtain a desired tradeoff between computational cost and a quality metric for the final answer.