From Question Context to Answer Credibility: Modeling Semantic Structures for Question Answering Using Statistical Methods

摘要

Within a Question Answering (QA) framework, Question Context plays a vital role. We define a Question Context to be background knowledge that can be used to represent the user's information need more completely than the terms in the query alone. The Aspect-Based Relevance Language Model is an approach that uses statistical language modeling techniques to model a semantic Question Context. This paper proposes a novel measure called Answer Credibility, which we derive from this semantic Question Context using a metric called Perspective Similarity. Our approach is significant because it allows us to extend the usage of statistical language modeling techniques, which have been successfully applied to the first stage (the IR stage) of QA, into the second stage. Because we use the document corpus itself as a knowledge source, our techniques do not require external resources such as ontologies or thesauri. Answer Credibility is incorporated into the QA process in the Answer Selection phase; we interpolate the final QA answer score using Answer Credibility. Our results are promising and show significant improvements in Mean Reciprocal Rank (MRR) and accuracy for "who, " "what," and "where " type questions over the baseline approach.