Reinforcement Learning-based Product Delivery Frequency Control

Frequency control is an important problem in modern recommender systems. It dictates the delivery frequency of recommendations to maintain product quality and efficiency. For example, delivering promotional notifications impacts daily metrics as well infrastructure resource consumption (e.g. CPU memory usage). There remain open questions on what objective we should optimize represent business values long term best, how balance between a dynamically fluctuating environment. We propose personalized methodology for problem, which combines long-term value optimization using reinforcement learning (RL) with robust volume technique termed "Effective Factor". demonstrate statistically significant improvement efficiency by our method several notification applications at scale billions users. To best knowledge, study represents first deep RL application such industrial scale.