Detect Rumors On Microblogging Websites

Nowadays,with the rise of social media,more and more people tend to post and get information via social websites.However,much misinformation or rumor are widely spread over the Internet and make an impact on the public and society in different degree.Therefore,Identifying rumors on social media websites automatically has become an important research issue.Most of the related works regard rumor detection as a classification process.Based on extracting a variety of features including message content,user files and diffusion properties,they can train supervised algorithms and obtain "rumor" classifiers.Nevertheless,a majority of related works only focus on the statistics of features on a fixed time,but ignore the variation of these features during the message spreading over time.The variation of these features can provide valuable information for rumor detection.This paper proposed a novel dynamic series-time structure to model these time sensitive feature,which can capture the variation properties.The main contribution of this paper is as followed:1)This paper analyzes the different pattern of feature vs time curve during the process of microblog propagation.2)We use two different datasets crawled from Sina Weibo and Twitter to confirm our proposed model outperforms state-of-the-art methods.3)In this paper,we examine two basic settings:(a)given the complete lifecycle of an event about some specific topic,we decide it is a rumor or not;(b)given the event data at the early stage of propagation,we apply our model for early rumor detection.Experimental results under the two settings demonstrate that our DSTS-based model achieves promising improvements over the state-of-the-art approaches on both datasets.4)We also apply our proposed framework for rumor detection on a sentiment analysis task on Twitter.And the experimental results show the effectiveness of our method.