arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2302.01680cs.LGcs.IR
收藏

用于短视频推荐的两阶段约束演员-评论家算法

Two-Stage Constrained Actor-Critic for Short Video Recommendation

Kuaishou Technology(快手科技);Hong Kong University of Science and Technology(香港科技大学)

Qingpeng Cai, Zhenghai Xue, Chi Zhang, Wanqi Xue, Shuchang Liu, Ruohan Zhan, Xueliang Wang, Tianyou Zuo, Wentao Xie, Dong Zheng, Peng Jiang, Kun Gai

AI总结:

本文将短视频推荐形式化为约束马尔可夫决策过程,提出两阶段约束演员-评论家算法,在优化累计观看时长主目标的同时平衡关注和分享等辅助目标,离线与在线实验均显著优于基线并已上线部署。

摘要:

短视频在社交媒体上的广泛普及,为优化视频分享平台上的推荐系统带来了新的机遇与挑战。用户顺序地与系统交互,并提供复杂且多方面的反馈,包括观看时长以及与多个视频的各种类型互动。一方面,平台旨在长期优化用户的累计观看时长(主要目标),这可以通过强化学习(Reinforcement Learning)有效优化。另一方面,平台还需要满足容纳多种用户互动反馈(辅助目标)的约束,例如关注、分享等。在本文中,我们将短视频推荐问题形式化为约束马尔可夫决策过程(Constrained Markov Decision Process, CMDP)。我们发现传统的约束强化学习算法在这种设置下无法很好地工作。我们提出了一种新颖的两阶段约束演员-评论家(actor-critic)方法:在第一阶段,我们学习独立的策略以优化每个辅助信号。在第二阶段,我们学习一个策略来优化主要信号,并使其与第一阶段学习的策略保持接近,这有效地保证了该主策略在辅助目标上的性能。通过广泛的离线评估,我们证明了我们的方法在优化主要目标以及平衡其他目标方面优于其他替代方案的有效性。我们进一步展示了我们的方法在短视频推荐在线实验中的优势,在观看时长和互动方面均显著优于其他基线。我们的方法已全面部署在生产系统中,以优化平台上的用户体验。

英文摘要:

The wide popularity of short videos on social media poses new opportunities and challenges to optimize recommender systems on the video-sharing platforms. Users sequentially interact with the system and provide complex and multi-faceted responses, including watch time and various types of interactions with multiple videos. One the one hand, the platforms aims at optimizing the users' cumulative watch time (main goal) in long term, which can be effectively optimized by Reinforcement Learning. On the other hand, the platforms also needs to satisfy the constraint of accommodating the responses of multiple user interactions (auxiliary goals) such like, follow, share etc. In this paper, we formulate the problem of short video recommendation as a Constrained Markov Decision Process (CMDP). We find that traditional constrained reinforcement learning algorithms can not work well in this setting. We propose a novel two-stage constrained actor-critic method: At stage one, we learn individual policies to optimize each auxiliary signal. At stage two, we learn a policy to (i) optimize the main signal and (ii) stay close to policies learned at the first stage, which effectively guarantees the performance of this main policy on the auxiliaries. Through extensive offline evaluations, we demonstrate effectiveness of our method over alternatives in both optimizing the main goal as well as balancing the others. We further show the advantage of our method in live experiments of short video recommendations, where it significantly outperforms other baselines in terms of both watch time and interactions. Our approach has been fully launched in the production system to optimize user experiences on the platform.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑