Research Overview
My research lies at the intersection of social recommendation systems, reinforcement learning, and causal inference. Throughout my Ph.D., I have been motivated by a central question: how does influence propagate through social networks, and how can adaptive recommender systems learn from and govern it? My work develops reinforcement learning-based social recommendation frameworks that integrate peer influence effects within connected systems.
Previous Research
My early research on cluster-based bandits [1] introduced a causal estimation framework that leverages adaptive intervention across network clusters to estimate total treatment effects under the presence of peer influence effects, bridging causal inference and online learning. Building on this foundation, my NetCB framework [2] extended traditional contextual multi-armed bandits to explicitly leverage peer influence effects in reward maximization, showing how accounting influence heterogeneity improves efficiency in social recommendation. I further advanced this idea in the SpillCB framework [3], which learns peer influence probabilities through intervention to optimize stimulated word-of-mouth diffusion. My subsequent work, the InfluenceCB framework [4] formalized the fundamental trade-off between maximizing rewards and influence-probability estimation error, revealing the Pareto frontier of achievable learning outcomes.
References
[1] A. Faruk, J. Sulskis, E. Zheleva. Estimating Causal Effects in Networks with Cluster-based Bandits. AAAI Workshop on Artificial Intelligence for Behavioral Change (AI4BC), 2022.
[2] A. Faruk, E. Zheleva. Leveraging Heterogeneous Spillover in Maximizing Contextual Bandit Rewards. 33rd International World Wide Web Conference (WWW/WebConf), 2025.
[3] A. Faruk, E. Zheleva. Contextual Bandits for Maximizing Stimulated Word-of-Mouth Rewards. AAAI Workshop on Bridging the Gap between AI Planning and Reinforcement Learning (PRL), 2025.
[4] A. Faruk, M. Shahverdikondori, E. Zheleva. Learning Peer Influence Probabilities with Linear Contextual Bandits. arXiv preprint, arXiv:2510.19119, 2025.
Current Research
My recent work integrates social learning dynamics into recommendation systems to mitigate herding and improve fairness. I study how to balance personalization vs social influence, individual satisfaction vs collective outcomes, and efficiency vs fairness in recommendation systems. I also work on:
Social learning dynamics in recommender systems
Causal bandits for networks with spillover effects
Position bias-aware reranking in personalized recommendations
Selection bias correction in graph collaborative filtering
My future work aims to understand cooperation and conflict in networked decision systems. I plan to:
Model users as strategic agents in contextual bandits
Study how equilibrium behavior emerges under social influence
Design interventions that promote cooperation and reduce polarization
This work bridges machine learning, game theory, and computational social science to understand how collective behavior evolves in algorithmic systems.