A two-stage approach to retrieving answers for how-to questions

Abstract

This paper addresses the problem of automatically retrieving answers for how-to questions, focusing on those that inquire about the procedure for achieving a specific goal. For such questions, typical information retrieval methods, based on key word matching, are better suited to detecting the content of the goal (e.g., 'installing a Windows XP server') than the general nature of the desired information (i.e., procedural, a series of steps for achieving this goal). We suggest dividing the process of retrieving answers for such questions into two stages, with each stage focusing on modeling one aspect of a how-to question. We compare the two-stage approach with two alternative approaches: a baseline approach that only uses the content of the goal to retrieve relevant documents and another approach that explores the potential of automatic query expansion. The result of the experiment shows that the two-stage approach significantly outperforms the baseline but achieves similar result with the systems using automatic query expansion techniques. We analyze the reason and also present some future work.