Learning the Query Generation Patterns

Abstract

With the current method of query formation, a Question Answering system retrieves a set of documents that are similar to a question, while what is mostly required is a set where an answer occurs frequently. This paper addresses this problem by presenting the Query Generation Pattern method. The aim of the method is to automatically learn an optimal combination, modifications of question words and possible extension of a query with non-question words, which for a given question category and syntax, form reliable queries that retrieve an answer-rich set of documents. © Springer-Verlag Berlin Heidelberg 2005.