Large-scale Web-based services present opportunities for improving UI policies based on observed user interactions. We address challenges of learning such policies through model-free offline Reinforcement Learning (RL) with off-policy training. Deployed in a production system for user authentication in a major social network, it significantly improves long-term objectives. We articulate practical challenges, compare several ML techniques, provide insights on training and evaluation of RL models, and discuss generalizations.
@article{arxiv.2102.05612,
title = {Personalization for Web-based Services using Offline Reinforcement Learning},
author = {Pavlos Athanasios Apostolopoulos and Zehui Wang and Hanson Wang and Chad Zhou and Kittipat Virochsiri and Norm Zhou and Igor L. Markov},
journal= {arXiv preprint arXiv:2102.05612},
year = {2022}
}