Meta Learning Based Rumor Detection with Awareness of Social Bot

Zhilong Lv, Zhen Huang, Menglong Lu, Yuxin Yang, Zhiliang Tian, Xin Niu, Dongsheng Li

Abstract: Rumors are widely spread on the Internet, and rumor detection is crucial for preserving the Internet environment. Sociological studies have shown that social bots play an important role in the rumor-spreading process. Although existing methods take into account the credibility of the user to distinguish between real users and bots, there are some “camouflage behaviors” from both the true user and the social bot, i.e., bots posting or replying to true posts and real users unintentionally posting or replying to rumors. Such a situation makes the model learn misleading knowledge from social bot detection that does not assist in rumor detection. In the paper, we introduce a model called MRS, which learns to make the model learn how to assist the rumor detection task based on the social bot detection task. Specifically, MRS combines two related but different tasks in a meta learning manner. The model is pseudo-updated in the inner loop, and then the updated model is applied to the rumor detection task in the outer loop, so the meta learning process allows the model to focus on the outer-loop rumor detection task. The experimental results show the superiority of MRS and that MRS can achieve 90% accuracy within two hours.