Importance of User Representation in Propagation Network-based Fake News Detection: A Critical Review and Potential Improvements

Propagation network-based fake news detection methods have unique benefits over content based methods, like being language agnostic and less prone to adversarial attacks. Each node in the propagation network denotes a social user who is involved in spreading the news. Following a thorough review of existing works in propagation-based fake news detection research, we argue in this paper that associating rich user representation within propagation networks can improve the detection method’s accuracy and scalability. Experimental results of existing works provide sufficient evident to our argument. Motivated by another line of research, we introduce a representation learning algorithm that produce rich representations for social users who are involved in fake news dissemination. This work paves the path for a more powerful propagation network-based fake news detection, possibly opening a new research direction. There are two main outcomes of this work (1) Identifying the importance of social-context aware social user representation (2) providing a methodology to obtain social context-aware user representations and a way to incorporate them with propagation network-based fake news detection methods.