Microblog sentiment analysis with weak dependency connections

With the rise of microblogging services like Twitter and Sina Weibo, users are able to post their real-time mood and opinions conveniently and swiftly. At the same time, the ubiquitous social media results in abundant social relations such as following and follower relations. Social relations create a new source for microblog sentiment analysis, which attracts a great amount of attention in recent years. There are two theories that support the use of social relations for sentiment analysis - sentiment consistency and emotional contagion. However, most existing microblog sentiment analysis methods only employ direct connections which cannot fully use the heterogeneous connections in social media. As online social networks consist of communities and nodes in the same community which form weak dependency connections usually share similarities, we investigate how to exploit weak dependency connections as an aspect of social contexts for microblog sentiment analysis in this paper. In particular, we employ community detection methods to capture weak dependency connections and propose a new model for microblog sentiment analysis which incorporates weak dependency connections, sentiment consistency, and emotional contagion together with text information. Experimental results on two real Twitter datasets demonstrate that our proposed model can outperform baseline methods consistently and significantly.