Zhiming Peng

Fan Wang, Xiaomin Fang, Lihang Liu et al.

In recommender systems such as news feed stream, it is essential to optimize the long-term utilities in the continuous user-system interaction processes. Previous works have proved the capability of reinforcement learning in this problem. However, there are many practical challenges to implement deep reinforcement learning in online systems, including low sample efficiency, uncontrollable risks, and excessive variances. To address these issues, we propose a novel reinforcement learning method, namely model-based counterfactual advantage learning (MBCAL). The proposed method takes advantage of the characteristics of recommender systems and draws ideas from the model-based reinforcement learning method for higher sample efficiency. It has two components: an environment model that predicts the instant user behavior one-by-one in an auto-regressive form, and a future advantage model that predicts the future utility. To alleviate the impact of excessive variance when learning the future advantage model, we employ counterfactual comparisons derived from the environment model. In consequence, the proposed method possesses high sample efficiency and significantly lower variance; Also, it is able to use existing user logs to avoid the risks of starting from scratch. In contrast to its capability, its implementation cost is relatively low, which fits well with practical systems. Theoretical analysis and elaborate experiments are presented. Results show that the proposed method transcends the other supervised learning and RL-based methods in both sample efficiency and asymptotic performances.

Fan Wang, Xiaomin Fang, Lihang Liu et al.

In the combinatorial recommender systems, multiple items are fed to the user at one time in the result page, where the correlations among the items have impact on the user behavior. In this work, we model the combinatorial recommendation as the problem of generating a sequence(ordered list) of items from a candidate set, with the target of maximizing the expected overall utility(e.g. total clicks) of the sequence. Toward solving this problem, we propose the Evaluation-Generation framework. On the one hand of this framework, an evaluation model is trained to evaluate the expected overall utility, by fully considering the user, item information and the correlations among the co-exposed items. On the other hand, generation policies based on heuristic searching or reinforcement learning are devised to generate potential high-quality sequences, from which the evaluation model select one to expose. We propose effective model architectures and learning metrics under this framework. We also offer series of offline tests to thoroughly investigate the performance of the proposed framework, as supplements to the online experiments. Our results show obvious increase in performance compared with the previous solutions.