Using Abstraction Level in Question Answering System

Abstract

Traditional question answering systems extract answers in terms of the relevance between answer and question. However, when there are multiple relevant answers to a question, people usually use other dimensions besides relevance to select answers. Abstraction level is a frequently used dimension to distinguish general answer and specific answer. Previous question answering systems seldom consider the dimension. This paper proposes a way to calculate the abstraction level of answer. Experiments show that abstraction level can improve question answering in certain situations.