Internet trolling, a form of antisocial online behavior, is a serious problem on contemporary social media, especially due to our information environment having become so closely intertwined with hybrid media. As skillful trolls can lure publics into polarized discussions and manipulate public opinion and even voting behavior (Bennett and Livingston 2018; Akhtar and Morrison 2019), there is need for novel methods that could uncover intentionally manipulative behaviors like trolling in online interaction.
In this talk, I present some key studies from two Academy of Finland projects, concentrating on how harmful online interaction can be analyzed from a conversation-level point-of-view. First, we utilized qualitative conversation analysis (Schegloff 1968, 2007; Schegloff and Sacks, 1973) to identify systematic interaction strategies used in trolling versus non-trolling behaviors. We found that at the level of turns in interaction trolling is identifiable by asymmetric social actions (mismatching, ignoring, challenging) that deflect any need for accountability or avoid explaining contra-normative social behavior – behavior which exploits others’ need for common grounding.
In our ongoing work, we explore the possibility of utilizing our qualitative insights to operationalize a computational model for analyzing interaction. I will present and evaluate our annotation scheme and the machine learning models we have thus far used for the classification task, and critically discuss their potential as well as the related challenges.
Aalto HELDIG DH pizza seminar on Friday 28 October 2022 at 12.00.