Efficient Dynamic Assortment Optimization with Low-Rank Dual Contextual Information

The paper proposes a new low-rank dynamic assortment model that effectively leverages dual contextual information (user and item features) to optimize personalized product recommendations in online retail. The authors develop an efficient algorithm, ELSA-UCB, that estimates the intrinsic low-dimensional subspaces and utilizes the upper confidence bound approach to address the exploration-exploitation trade-off.