Sentiment Analysis for Topics based on Interaction Chain Model

Abstract

With the rapid development of Internet and Web 2.0, online social networks, such as Facebook, Twitter, LinkedIn, have become valuable sources for public opinion mining and sentiment analysis. Millions of users are sharing their views and discussing current issues through social media every day. Microblog, as a convenient and easily access platform, also attracts more and more people to express their feelings about hot topics. However, as the maximum message length is only 140 characters in microblog, traditional sentiment analysis methods for topics cannot well performed due to the lack of information. In this paper, we propose a novel sentiment analysis methods based on interaction chain. Firstly, we organize messages as interaction-chains by taking advantages of the explicit interaction markers in microblog. Then, interaction-chains are clustered into different topics by comparing the similarity among them. After that, we perform sentiment analysis using semantic-based SBV polarity algorithm. We also proposed two heuristics according to the specificities of microblog. Experimental evaluations show that the proposed heuristic interaction-chain-based algorithm can extract discriminative and meaningful topics and could conduct sentiment analysis effectively.