A Framework to Answer Questions of Opinion Type

Abstract

In this paper, we propose a framework to answer questions of opinion type. The data source is the web pages returned from the search engine. By using Bayes Classifier, the main texts on the pages are classified into three categories at sentence level: positive review, negative review and neutral review. K-means method is used to cluster the sentences of positive review and negative review respectively. The final answers are extracted from the sentence groups after clustering and presented in the form of quaternion. We design a system to test this framework. The experimental results show that it is effective.