Abstract:
|
In retailing and online advertising, the seller faces the problem of showing the customers a subset of products from many candidate items. Choice models are utilized to model the customers' behaviors, namely, the probabilities a customer selecting a product from the offered assortment. For better revenues, the seller needs to learn the choice model from the historical data and make the optimal assortment decisions based on the estimation. In the dynamic version of this assortment problem, the seller sequentially makes assortment decisions, receives choices from the customer, and learns the customer preferences for better decision making in the future. One key goal in dynamic assortment is to maximize the total revenues collected in a finite selling horizon, which requires a subtle balance between exploration, referring to learning customer preferences, and exploitation of current knowledge for better intermediate revenues. In this paper, we will present a new approach for dynamic assortment. Theoretical and numerical studies demonstrate the effectiveness of the proposed method.
|