Sentiment Diffusion in Large Scale Social Networks

Popularity of online social networks provides the chance to make sentiment analysis on every user instead of every document or sentence. And relations between users on social media sites often indicate correlation (negation) between users' opinions. In this work, we study how user's opinion spread in social networks. We employ the data from Tencent.com, the largest social network of China to empirically study the problem. Our work focuses on six different topics including policy, products, brand, sports, movie and politician. We study the distributions of peoples' opinions on different topics and how users' opinions are influenced by those he is following. We propose a graphical model to capture the essence of social network as well as an algorithm to perform semi-supervised learning. The learning algorithm can be used to accurately predict users' sentiment in the social network.