With the development of information technology, Internet service platforms become more and more important. Recommendation systems, among many others, play a vital role in generating revenue and competition for these platforms. This thesis investigates the revenue management problems in two recommendation systems.

In the first part, we develop a variant of the multinomial logit model with impatient customers who make choice decisions in multiple stages. In each stage, if the maximum utility of viewed products is larger than a minimum acceptable one, the customer purchases right away. Otherwise, the customer stays till the last stage within her patience level and leaves without purchasing. We propose efficient algorithms for pure assortment/pricing, joint assortment and pricing problems....[ Read more ]

With the development of information technology, Internet service platforms become more and more important. Recommendation systems, among many others, play a vital role in generating revenue and competition for these platforms. This thesis investigates the revenue management problems in two recommendation systems.

In the first part, we develop a variant of the multinomial logit model with impatient customers who make choice decisions in multiple stages. In each stage, if the maximum utility of viewed products is larger than a minimum acceptable one, the customer purchases right away. Otherwise, the customer stays till the last stage within her patience level and leaves without purchasing. We propose efficient algorithms for pure assortment/pricing, joint assortment and pricing problems. In the second part, we propose one innovative model for the live-streaming e-commerce. In reality, both Amazon and Alibaba have embraced live-streaming as a novel marketing tool. We use a mechanism design framework to demonstrate how the live-streaming platform can leverage information provision to improve its revenue compared to traditional position auctions. We show that the revenue loss from neglecting information provision can be arbitrarily large. The optimal mechanism problem is NP-hard and we develop several heuristics with constant performance bounds. We attribute the effectiveness of information provision to its discriminatory role in serving heterogeneous advertisers.