Research on Focal Figures-Oriented Sentimental Question Answering

Abstract

Besides the related issues about sentiment, such as identifying opinion holders or sentiment polarity, the answers of the user's questions are more complicated and scattered in opinion or sentimental QA than in fact QA. In this paper, a lot of sentimental questions have been gathered from zhidao.baidu.com and five kinds of sentimental question types have been summarized according to the characteristics of the questions. Since sentimental questions are different from fact-based questions, the question classification method only based on the interrogative words is not fit for sentimental questions. So the method of the chunk-based CRF model combined with rules and semantic sentimental resources in sentimental question classification is used in this paper. In answer extraction step, answers are obtained by different methods based on various question types combined with the identification of the chunks and the sentimental polarities of the sentences.