SPREADSHOT: Analysis of fake news spreading through topic modeling and bipartite weighted graphs

Abstract

Spreading of fake news is one of the primary drivers of misinformation in social networks. Graph-based approaches that analyze fake news dissemination are mostly dedicated to fake news detection and consider homogeneous tree-based networks obtained by following the diffusion of single messages through users, thus lacking the ability to identify implicit patterns among spreaders and topics. Alternatively, heterogeneous graphs have been proposed, although the detection remains their main goal and the use of graph centralities is rather limited. In this paper, bipartite weighted graphs are used to analyze fake news and spreaders by utilizing topic modeling and a combination of network centrality measures. The proposed architecture, called SPREADSHOT, leverages a topic modeling technique to identify key topics or subjects within a collection of fake news articles published by spreaders, thus generating a bipartite weighted graph. By projecting the graph model to the space of spreaders, one can identify the strengths of links between them in terms of fakeness correlation on common topics. Moreover, the closeness and betweennes centralities highlight spreaders who represent key enablers in the dissemination of fakeness on different topics. The projection of the bipartite graph to the space of topics allows one to identify topics which are more prone to misinformation. By collecting specific network measures, a synthetic fakeness networking index is defined which characterizes the behaviors and roles of spreaders and topics in the fakeness dissemination. The effectiveness of the proposed technique is demonstrated through tests on the LIAR dataset.