Ordering and Ranking Products as an Online Newsvendor
In e-commerce, being displayed in a prominent space on the website is important to a product’s sales. Customers are more likely to click and buy a highly positioned item than products down the page. Data from both industry and academia has shown that the traffic difference between the first and tenth position can be ten-fold. In this work, we consider the problem wherein an online newsvendor wants to jointly order inventory and rank products to maximize its profits. We discover that ordering inventory and ranking products without accounting for each other can lead to worst-case profit losses of 50 percent or more. Therefore, we develop an optimal polynomial-time algorithm for the joint problem, which is based on sequentially solving a newsvendor problem and an assignment problem. In addition, we study the problem under two additional settings: one where rankings can be personalized over time, and one where demand needs to be learned over time. In both cases, our algorithm is proven to be asymptotically optimal, which indicates that it performs well when the user base grows. We observe the same strong performance in our computational experiments, even when we specify the demand distribution to be significantly different from the model that was studied.
Meeting ID: 974 0643 3088