Enhancing web search through question classifier

Abstract

Question answering field has evolved alongside the Natural Language processing. There are several small-scale applications that use the linguistic, semantic, and syntactic interpretations of text and consume it in further processing. In case of a web search, knowing what the query is intended for can save hours of CPU processing and decrease response time tremendously. We have taken a small step in this direction by treating the query as a question and classifying it with the best suited classification algorithm. In this paper, we have tried to find out that when a perfect-informer of the question (knowing what is asked) is provided as an input for classification algorithms like SVN, Naïve Base, and decision tree, we want to observe their accuracy on the same data set of questions. In our experiment, we have used the concept of CRF to find question features that are relevant. CRF is a probabilistic model that treats features as observation sequence and emits all sequence labels with probability values.