Ahana Biswas

I study how content gains visibility and how audience reactions—likes, comments, shares—reinforce or reshape communication dynamics. This includes work on cross-partisan interactions, toxic amplification, and elite–audience behavior across TikTok, Twitter/X, and Facebook.

• Political Elites in the Attention Economy (ICWSM 2025): A network- and causally informed analysis examining how visibility signals and audience responses shape political elites’ communication incentives.
• Cross-Partisan Interaction Dynamics (ICWSM 2026 under review): A longitudinal study of how engagement with cross-partisan posts affects future interactions, using matching-based causal inference and visibility-feedback modeling.

I develop and apply causal frameworks to uncover mechanisms driving online engagement and information flow, linking micro-level behavior with macro-level patterns of discourse.

• Toxic Politics & TikTok Engagement (HKS Misinformation Review, 2025): A multimodal causal analysis investigating how toxicity interacts with algorithmic amplification and user engagement during the 2024 U.S. election.
• Political Narratives in Crisis Events (SBP-BRiMS 2023): Narrative and temporal modeling of discourse evolution during the Russia–Ukraine conflict, examining how geopolitical events structure online conversations.

I examine how emerging AI systems—particularly large language models—reshape online communication, coordination, and public expression. My work explores how AI agents participate in, mediate, or influence discourse, and how human–AI interaction alters conversational norms and visibility dynamics.

• LLM-Mediated Dialogue and Conflict Resolution: Controlled experiments where LLMs intervene in polarized discussions—either as mediators or co-participants—to test how AI-generated suggestions influence tone, cooperation, and conversational outcomes.
• Agent-Based Simulations with LLM-Generated Communication Styles: Using LLMs to generate diverse rhetorical and affective communication patterns and embedding them in simulation environments to model how AI-augmented messages diffuse through networks and shape engagement and narrative evolution.