Rumors detection in chinese via crowd responses

In recent years, microblogging platforms have become good places to spread various spams, making the problem of gauging information credibility on social networks receive considerable attention especially under an emergency situation. Unlike previous studies on detecting rumors using tweets' inherent attributes generally, in this work, we shift the premise and focus on identifying event rumors on Weibo by extracting features from crowd responses that are texts of retweets (reposting tweets) and comments under a certain social event. Firstly the paper proposes a method of collecting theme data, including a sample set of tweets which have been confirmed to be false rumors based on information from the official rumor-busting service provided by Weibo. Secondly clustering analysis of tweets are made to examine the text features extracted from retweets and comments, and a classifier is trained based on observed feature distribution to automatically judge rumors from a mixed set of valid news and false information. The experiments show that the new features we propose are indeed effective in the classification, and especially some stop words and punctuations which are treated as noises in previous works can play an important role in rumor detection. To the best of our knowledge, this work is the first to detect rumors in Chinese via crowd responses under an emergency situation.