Language Modeling for Document Selection in Question Answering

Usually, in the Question Answering domain, for a question in natural language, precise answers to the question are extracted from documents according only to the context of the question. In this work, we complemented this approach by adding a filtering process on top of the document retrieval. This way, the system reevaluates the documents it has originally selected during the information retrieval step before the answer extraction and scoring. Such re-evaluation aims at filtering out documents considered unusable for the search. Based on statistical language modeling, the filtering process firstly determines the intrinsic relevancy of a document and then decides whether this document is a priori relevant for finding answers. Evaluation on factoid questions and a collection of 500k web documents has shown our approach properly supports the Question Answering task.