Question Classification Using a Knowledge-Based Semantic Kernel

Abstract

Question classification is a significant component of any question answering system and plays a vital role in the overall accuracy of the QA system, and the key to the accuracy of a question classifier depends on the set of features extracted. Generally, for all text classification problems, the data is represented in a vector space model with bag-of-words (BOW) approach. But despite the simplicity of BOW approach, it suffers some serious drawbacks like it cannot handle synonymy or polysemy and does not take into account the semantic relatedness between the words. In this paper, we propose knowledge-based semantic kernel that uses WordNet as its knowledge-base and a semantic relatedness measure SR. We experimented with five machine learning algorithms viz. Nearest Neighbors (NN), Naive Bayes (NB), Decision Tree (DT), and Support Vector Machines (SVM) to compare the results. For SVM we experimented with linear kernel and the proposed semantic kernel represented by SVMSR.