Enhancing Personalized Recommendation Systems: A Hybrid Etc-Ucb Bandit Algorithm Approach

Dundee International Institute, Central South University, Changsha, 414000, China

This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

In the era of information overload, personalized recommendation systems have become indispensable tools for enhancing user experience by filtering relevant content. However, the exploration-exploitation tradeoff remains a critical challenge, where systems must balance between leveraging known user preferences (exploitation) and exploring potential new interests (exploration). This paper introduces a hybrid Equal-Time Exploration followed by an Upper Confidence Bound (ETC-UCB) bandit algorithm to address this challenge. The proposed algorithm divides the recommendation process into two phases: an initial Equal-Time Exploration (ETC) phase to ensure uniform exploration of all content categories and a subsequent Upper Confidence Bound (UCB) phase to exploit historical data for optimal arm selection. Empirical evaluations on the MovieLens dataset demonstrate that the ETC-UCB algorithm significantly reduces cumulative regret across varying interaction scales (n = 50, 500, 5000, 50000), with regret growth adhering to theoretical logarithmic bounds and stable standard deviations indicating robust performance. By integrating structured exploration with adaptive exploitation, this study provides a novel framework for real-time optimization in dynamic recommendation environments, bridging the gap between theoretical bandit models and practical system implementations.