Analysing definition questions by two machine learning approaches

Abstract

In automatic question answering, the identification of the correct target term (i.e. the term to define) in a definition question is critical since if the target term is not correctly identified, then all subsequent modules have no chance of providing relevant nuggets. In this paper, we present a method to tag a question sentence experimenting with two learning approaches: QTag and Hidden Markov Model. We tested the methods in five collections of questions, PILOT, TREC 2003, TREC 2004, CLEF 2004 and CLEF 2005. We performed ten-fold cross validation for each collection and we also tested with all questions together. The best accuracy rates for each collection were obtained using QTag, but with all questions together the best accuracy rate is obtained using HMM.