Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces CoSimRec, an offline agent-based evaluation framework designed to assess the impact of coordinated accounts on recommender systems through a closed-loop feedback process. By employing the Algorithmic Penetration Rate (APR) metric family, the study quantifies how coordinated interactions influence content exposure and engagement across various recommendation algorithms. The findings reveal that while random controls show no significant penetration, popularity-based and feedback-sensitive recommenders achieve notable APR-Lift, indicating a substantial amplification of coordinated content in user feeds.
Coordinated activity can amplify content exposure in recommender systems by up to 45%, revealing vulnerabilities in popular ranking strategies.
Recommender systems increasingly shape which content reaches users, making it important to understand whether coordinated activity is amplified beyond the accounts that initiate it. Existing robustness evaluations largely focus on static target-rank changes and do not capture how coordinated interactions, recommendation, and user response evolve within a feedback loop. To address this gap, we propose CoSimRec, an offline agent-based evaluation framework that models coordinated accounts, dynamic ranking, non-bot responses, and ranking interventions in a shared closed-loop process. CoSimRec introduces the Algorithmic Penetration Rate (APR) metric family to measure target content's share of non-bot exposure and engagement, lift against matched no-attack baselines, and exposure gained per coordinated interaction. We evaluate CoSimRec on MIND, MovieLens, and LastFM using random, popularity-based, feedback-sensitive, MF, and BPR-MF recommenders, with ten-seed inference for the primary APR analysis and population-scale experiments of up to 1000 users. Random controls show no statistically supported positive penetration, whereas popularity-based and feedback-sensitive ranking produce significant positive APR-Lift in all six master-worker dataset--recommender settings, reaching 0.4505 on LastFM; synchronization-aware ranking reduces APR in every corresponding defense setting.