Abstract:Reinforcement learning (RL) has been widely applied in recommendation systems due to its potential in optimizing the long-term engagement of users. From the perspective of RL, recommendation can be formulated as a Markov decision process (MDP), where recommendation system (agent) can interact with users (environment) and acquire feedback (reward signals).However, it is impractical to conduct online interactions with the concern on user experience and implementation complexity, and we can only train RL recommenders with offline datasets containing limited reward signals and state transitions. Therefore, the data sparsity issue of reward signals and state transitions is very severe, while it has long been overlooked by existing RL <a class="link-external link-http" href="http://recommenders.Worse" rel="external noopener nofollow">this http URL</a> still, RL methods learn through the trial-and-error mode, but negative feedback cannot be obtained in implicit feedback recommendation tasks, which aggravates the overestimation problem of offline RL recommender. To address these challenges, we propose a novel RL recommender named model-enhanced contrastive reinforcement learning (MCRL). On the one hand, we learn a value function to estimate the long-term engagement of users, together with a conservative value learning mechanism to alleviate the overestimation <a class="link-external link-http" href="http://problem.On" rel="external noopener nofollow">this http URL</a> the other hand, we construct some positive and negative state-action pairs to model the reward function and state transition function with contrastive learning to exploit the internal structure information of MDP. Experiments demonstrate that the proposed method significantly outperforms existing offline RL and self-supervised RL methods with different representative backbone networks on two real-world datasets.

Rethinking Offline Reinforcement Learning for Sequential Recommendation from A Pair-Wise Q-Learning Perspective

Beyond Reward: Offline Preference-guided Policy Optimization

Offline Deep Reinforcement Learning Two-stage Optimization Framework Applied to Recommendation Systems

Rethinking Reinforcement Learning for Recommendation: A Prompt Perspective

Pseudo Dyna-Q

Value Penalized Q-Learning for Recommender Systems

Sequential Search with Off-Policy Reinforcement Learning

Model-enhanced Contrastive Reinforcement Learning for Sequential Recommendation

Offline Adaptive Policy Leaning in Real-World Sequential Recommendation Systems

AdaRec: Adaptive Sequential Recommendation for Reinforcing Long-term User Engagement

Robust Reinforcement Learning Objectives for Sequential Recommender Systems

Self-Supervised Reinforcement Learning for Recommender Systems

Doubly Constrained Offline Reinforcement Learning for Learning Path Recommendation

ROLeR: Effective Reward Shaping in Offline Reinforcement Learning for Recommender Systems

Deep Reinforcement Learning for List-wise Recommendations

Alleviating Matthew Effect of Offline Reinforcement Learning in Interactive Recommendation

On the Opportunities and Challenges of Offline Reinforcement Learning for Recommender Systems

KERL: A Knowledge-Guided Reinforcement Learning Model for Sequential Recommendation

M 3 Rec: A Context-aware Offline Meta-level Model-based Reinforcement Learning Approach for Cold-Start Recommendation

Offline Reinforcement Learning for Human-Guided Human-Machine Interaction with Private Information