Bayesian Deep Learning Based Exploration-Exploitation for Personalized Recommendations
Xin Wang, Serdar Kadıoğlu · 2019
Personalized Recommendation Systems require an effective method to balance exploration and exploitation. To learn effective strategies, user and item attributes are critical data sources to capture contextual information. In this paper, we first present an approach based on Bayesian Deep Learning to learn a compact representation of user and item attributes to guide exploitation. A key novelty of the approach lies in its ability to also capture the uncertainty associated with the model output to guide exploration. We then show how to further boost exploration by incorporating model uncertainty with that of data uncertainty. Experimental results demonstrate the benefits of our approach in terms of accuracy in recommendations as well as its performance in an online setting.