The contextual multi-armed bandit (MAB) is a widely used framework for problems requiring sequential decision-making under uncertainty, such as Recommendation Systems. In applications involving a large number of users, the performance of contextual MAB can be significantly improved by