Investigating sentiment impact on information propagation and its evolution in microblog

The emergence of microblog brings a fundamental change to traditional information diffusion patterns. While there are a significant amount of studies on modelling the information propagation in microblog networks, little prior work has considered sentiments of messages as factors or information entities for propagation studies. In this paper, we present our work on characterizing the impact of sentiments on information propagation as well as their evolution in microblog networks, which is based on a Chinese microblog 'Weibo'. We start with proposing a sentiment new word detection method considering properties of Chinese post, after which regression techniques are adopted to examine the impact of sentiments on the propagation of posts and reposts. Furthermore, we model sentiment evolution during the discussion of social events as sentiment repost trees and put forward a mining algorithm to discover frequent sentiment flows among different types of users. Experimental results illustrate that sentiments in both posts and reposts have positive impacts on information propagation, and more interestingly, the effects of positive and negative emotion vary in posts and reposts. Meanwhile, the discovered frequent sentiment flows demonstrate that people prefer to express their sentiments and debate intensely in negative events.