Lexical convergence and collective identities on Facebook (1903.11452v3)

Abstract: Recent studies, targeting Facebook, showed the tendency of users to interact with information adhering to their preferred narrative and to ignore dissenting information. Primarily driven by confirmation bias, users tend to join polarized clusters where they cooperate to reinforce a like-minded system of beliefs, thus facilitating fake news and misinformation cascades. To gain a deeper understanding of these phenomena, in this work we analyze the lexicons used by the communities of users emerging on Facebook around verified and unverified contents. We show how the lexical approach provides important insights about the kind of information processed by the two communities of users and about their overall sentiment. Furthermore, by focusing on comment threads, we observe a strong positive correlation between the lexical convergence of co-commenters and their number of interactions, which in turns suggests that such a trend could be a proxy for the emergence of collective identities and polarization in opinion dynamics.