Real-Time News Cer tification System on Sina Weibo

Xing Zhou, Juan Cao, Zhiwei Jin, Fei Xie, Yu Su, Dafeng Chu, Xuehui Cao, Junqiang Zhang

Abstract: In this paper, we propose a novel framework for real-time news certification. Traditional methods detect rumors on message-level and analyze the credibility of one tweet. However, in most occasions, we only remember the keywords of an event and it's hard for us to completely describe an event in a tweet. Based on the keywords of an event, we gather related microblogs through a distributed data acquisition system which solves the real-time processing needs. Then, we build an ensemble model that combine user-based, propagation-based and content-based model. The experiments show that our system can give a response at 35 seconds on average per query which is critical for real-time system. Most importantly, our ensemble model boost the performance. We also offer some important information such as key users, key microblogs and timeline of events for further investigation of an event.Our system is already deployed in the Xihua News Agency for half a year. To the best of our knowledge, this is the first real-time news certification system for verifying social media contents.