Recommender selects and presents top-K items to the user at each online request, and a recommendation session consists of several sequential requests. Formulating a recommendation session as a Markov decision process and solving it by reinforcement learning (RL) framework has attracted increasing attention from both academic and industry communities. In this paper, we propose a RL-based industrial short-video recommender ranking framework, which models and maximizes user watch-time in an environment of user multi-aspect preferences by a collaborative multi-agent formulization. Moreover, our proposed framework adopts a model-based learning approach to alleviate the sample selection bias which is a crucial but intractable problem in industrial recommender system. Extensive offline evaluations and live experiments confirm the effectiveness of our proposed method over alternatives. Our proposed approach has been deployed in our real large-scale short-video sharing platform, successfully serving over hundreds of millions users.

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.