A comparative study of algorithm-mediated behavior and toxic discourse on social media platforms

In the evolving digital communication landscape, social media platforms are pivotal in shaping public opinion and societal narratives. These platforms, characterized by their democratized access to information and the facility for real-time engagement, have the potential to enrich public discourse significantly. However, they may also foster environments that restrict users' exposure to diversified content, contributing to the formation of echo chambers (i.e., groups of like-minded individuals where homogenous ideologies are reinforced), exacerbating user polarization, and promoting antisocial behaviors with severe implications for the broader democratic process. This dissertation explores the complex interplay between the dissemination of misinformation and its impact on online discourse, with a methodological innovation at its core. Recognizing that human behavior, as reflected in social media data, is inherently mediated by platform-specific algorithms, this research proposes a novel analytical framework. Adopting a comparative approach across various platforms seeks to unmask the underlying patterns of information propagation and user engagement, transcending the limitations imposed by algorithmic mediation. We first conduct a detailed examination of misinformation and conspiracy theory diffusion across social media landscapes, offering a quantitative assessment of the effectiveness of existing moderation policies. Then, we explore new expressive forms emerging within online dialogues, mainly through the lens of meme diffusion, to understand the relationship between viral content and the generation of controversial user reactions. Expanding on these insights, we conduct a comprehensive cross-platform analysis of toxic conversational dynamics, assessing how polarization contributes to the proliferation of hate speech online. Lastly, we discuss potential advancements in moderation tool designs, aiming to mitigate such digital hostility proactively. The empirical findings underscore the effectiveness of moderation tools in counteracting the spread of conspiracy theories. They reveal a tendency for viral topics to spark controversial and heated discussions, with toxicity levels intensifying progressively. Importantly, this research demonstrates the predictive ability of classifiers, trained on different stages of conversations, to identify the presence of toxic comments with high accuracy, even within a constrained feature set. This dissertation underscores the necessity of comparative, cross-platform analysis to understand the digital communication ecosystem. It explores how platform-induced behaviors may influence public discourse, providing insights that bridge human behavioral studies and digital platform policies.