Which side are you on? Insider-Outsider classification in conspiracy-theoretic social media

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 4975 - 4987 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Which side are you on? Insider-Outsider classification in conspiracy-theoretic social media Pavan Holur1, Tianyi Wang1, Shadi Shahsavari1, Timothy Tangherlini2, and Vwani Roychowdhury1 1 Department of Electrical and Computer Engineering, UCLA 2 Department of Scandinavian, UC Berkeley {pholur,tianyiw,shadihpp,vwani}@ucla.edu, tango@berkeley.edu Abstract Social media is a breeding ground for threat narratives and related conspiracy theories. In these, an outside group threatens the integrity of an inside group, leading to the emergence of sharply defined group identities: Insiders – agents with whom the authors identify and Out- siders – agents who threaten the insiders. In- ferring the members of these groups constitutes a challenging new NLP task: (i) Information is distributed over many poorly-constructed posts; (ii) Threats and threat agents are highly contextual, with the same post potentially hav- ing multiple agents assigned to membership in either group; (iii) An agent’s identity is often implicit and transitive; and (iv) Phrases used to imply Outsider status often do not follow com- mon negative sentiment patterns. To address these challenges, we define a novel Insider- Outsider classification task. Because we are not aware of any appropriate existing datasets or at- tendant models, we introduce a labeled dataset (CT5K) and design a model (NP2IO) to address this task. NP2IO leverages pretrained language modeling to classify Insiders and Outsiders. NP2IO is shown to be robust, generalizing to noun phrases not seen during training, and ex- ceeding the performance of non-trivial baseline models by 20%. 1 Background and Motivation Narrative models – often succinctly represented as a network of characters, their roles, their inter- actions (syuzhet) and associated time-sequencing information (fabula) – have been a subject of con- siderable interest in computational linguistics and narrative theory. Stories rest on the generative backbone of narrative frameworks (Bailey, 1999; Beatty, 2016). While the details might vary from one story to another, this variation can be com- pressed into a limited set of domain-dependent nar- rative roles and functions (Dundes, 1962). Social narratives that both directly and indirectly contribute to the construction of individual and group identities are an emergent phenomenon re- sulting from distributed social discourse. Currently, this phenomenon is most readily apparent on so- cial media platforms, with their large piazzas and niche enclaves. Here, multiple threat-centric nar- ratives emerge and, often, over time are linked to- gether into complex conspiracy theories (Tangher- lini et al., 2020). Conspiracy theories, and their constituent threat narratives (legend, rumor, per- sonal experience narrative) share a signature se- mantic structure: an implicitly accepted Insider group; a diverse group of threatening Outsiders; specific threats from the Outsider directed at the Insiders; details of how and why Outsiders are threatening; and a set of strategies proposed for the Insiders to counter these threats (Tangherlini, 2018). Indeed, the Insider/Outsider groups are fundamental in most studies of belief narrative, and have been exhaustively studied in social theory and more specifically, …

Content truncated to 500 words