Question Classification Based on Focus

Question classification plays an important role in Question Answer system. This paper proposes a method based on question focus, which combines advantages of both rule based methods and statistical methods. The question focus is the kernel of a question and represents the form and content of the doubt. Firstly, question focus definition is given according to linguistics and extraction method is described based on dependency analysis and semantic role labeling on which both depends statistical machine learning. And then, classifying questions with same focus to one category, finer question taxonomy without unreliable human effects is introduced with support of domain ontology. In order to evaluate contributions of question focus, a classifier using question focus is designed and implemented in a practical QA system in restricted domain of computer hardware. Experimental result shows efficiency of question focus and contributions to improve accuracy of question classification and the overall performance of QA.