With the prosperity of social media, online rumors become a severe social problem, which often lead to serious consequences, e.g., social panic and even chaos. Therefore, how to automatically identify rumors in social media has attracted much research attention. Most existing studies address this problem by extracting features from the contents of rumors and their reposts as well as the users involved. For these features, especially diffusion features, these works ignore systematic analysis and the exploration of difference between rumors and non-rumors, which exert targeted effect on rumor identification. In this paper, we systematically investigate this problem from a diffusion perspective using Sina Weibo data. We first extract a group of new features from the diffusion processes of messages and then make a few important observations on them. Based on these features, we develop classifiers to discriminate rumors and non-rumors. Experimental comparisons with the state-of-the-arts methods demonstrate the effectiveness of these features.
展开▼
