Detect Rumors in Microblog Posts Using Propagation Structure via Kernel Learning

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 708–717 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1066 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 708–717 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1066 Detect Rumors in Microblog Posts Using Propagation Structure via Kernel Learning Jing Ma1, Wei Gao2, Kam-Fai Wong1,3 1The Chinese University of Hong Kong, Hong Kong SAR 2Qatar Computing Research Institute, Doha, Qatar 3MoE Key Laboratory of High Confidence Software Technologies, China 1{majing,kfwong}@se.cuhk.edu.hk, 2wgao@qf.org.qa Abstract How fake news goes viral via social me- dia? How does its propagation pattern dif- fer from real stories? In this paper, we attempt to address the problem of identi- fying rumors, i.e., fake information, out of microblog posts based on their prop- agation structure. We firstly model mi- croblog posts diffusion with propagation trees, which provide valuable clues on how an original message is transmitted and de- veloped over time. We then propose a kernel-based method called Propagation Tree Kernel, which captures high-order patterns differentiating different types of rumors by evaluating the similarities be- tween their propagation tree structures. Experimental results on two real-world datasets demonstrate that the proposed kernel-based approach can detect rumors more quickly and accurately than state-of- the-art rumor detection models. 1 Introduction On November 9th, 2016, Eric Tucker, a grass- roots user who had just about 40 followers on Twitter, tweeted his unverified observations about paid protesters being bused to attend anti-Trump demonstration in Austin, Texas. The tweet, which was proved false later, was shared over 16 thou- sand times on Twitter and 350 thousand times on Facebook within a couple of days, fueling a nation-wide conspiracy theory1. The diffusion of the story is illustrated as Figure 1 which gives the key spreading points of the story along the time line. We can see that after the initial post, the tweet 1https://www.nytimes.com/2016/11/20/ business/media/how-fake-news-spreads. html was shared or promoted by some influential online communities and users (including Trump himself), resulting in its wide spread. A widely accepted definition of rumor is “un- verified and instrumentally relevant information statements in circulation” (DiFonzo and Bordia, 2007). This unverified information may eventu- ally turn out to be true, or partly or entirely false. In today’s ever-connected world, rumors can arise and spread at lightening speed thanks to social me- dia platforms, which could not only be wrong, but be misleading and dangerous to the public society. Therefore, it is crucial to track and debunk such rumors in timely manner. Journalists and fact-checking websites such as snopes.com have made efforts to track and de- tect rumors. However, such endeavor is man- ual, thus prone to poor coverage and low speed. Feature-based methods (Castillo et al., 2011; Yang et al., 2012; Ma et al., 2015) achieved certain suc- cess by employing large feature sets crafted from message contents, user profiles and holistic statis- tics of diffusion patterns (e.g., number of retweets, …

Content truncated to 500 words