Papers

Note: Author ordering on all papers is alphabetical, as is convention in OR/MS.

Working Papers

• Churning while Learning: Maximizing User Engagement in a Recommendation System w/ Raghav Singal
  • Abstract

    Massive online media platforms like Spotify, Youtube Music, Tencent Music, NetEase Cloud Music, and others largely generate revenue through targeted content delivery, where the right choice of media for each user can induce them to further interact with the platform, view ads, sign up for premium memberships and more. In order to maximize this revenue, these platforms need to strike a delicate balance between showing content that is effectively tailored to their user bases, and experimenting with new content that may have been recently uploaded and where neither the ideal demographic nor the overall efficacy of the content is well understood. This balance, especially in the more specific context of deciding how to serve advertisements, has largely been modeled as an exploration versus exploitation trade-off, where the platform balances learning about new ad click-through-rates (CTRs) via \textit{experimentation} against maximizing its revenue by showing ads from well understood campaigns. In this paper we examine the role of such experimentation on platform user retention. Specifically, we are interested in which types of users should be shown newly uploaded content. To answer this question we set up and solve a natural two period model where user retention from period to period depends explicitly on user type and whether or not they had previously engaged with the platform. Our model is motivated by our analysis of data from NetEase Cloud Music \cite{zhang2020netease}, the second largest online media platform in China. We find strong numerical evidence that current practices to do not take into account the effect of content experimentation on user retention. Specifically, we find that new users to the platform are often recommended new, poorly understood creatives and that new users who do not engage with these creatives are substantially less likely to return than ones who do.
• Feature-Based Market Segmentation and Pricing
  • Abstract

    With the rapid development of data-driven analytics, many firms have begun experimenting with personalized pricing strategies, i.e. strategies that predict a customer's valuation then offer them a tailored price. Ideally, a firm would perfectly predict each customer's valuation and price their goods accordingly. Unfortunately, in practice these valuations must be predicted by the firm using noisy regression models, and the number of prices the firm can offer is constrained by operational considerations. In this work, we give a general framework for analyzing and optimizing semi-personalized pricing strategies where the seller uses features about their customers to jointly segment and price their market. Specifically, we show how a seller can leverage a noisy valuation model to construct segmentation and pricing decisions with provable bounds on the lost revenue. We then give a series of the results that explain how a seller can improve their strategies by decomposing their lost profits as stemming from either prediction error or limited price flexibility. Along the way we prove a number of structural properties about monopoly pricing when valuations are the output of a regression model that may be of independent interest.

Accepted Papers

• Loot Box Pricing and Design with Ningyuan Chen, Adam N. Elmachtoub, and Xiao Lei. [Direct Link] [Slides]
  • Management Science (Dec. 2020)
  • Accepted in The 21st ACM Conference on Economics and Computation (EC), 2020.
  • Invited to present at the Federal Trade Commission (FTC) Workshop on Consumer Issues Related to Loot Boxes, 2019 (one of four research papers selected).
  • 1st Place, IBM Best Student Paper Award in Service Science, 2019 (to Xiao Lei).
• The Value of Personalized Pricing with Adam N. Elmachtoub and Vishal Gupta. [Direct Link] [Slides]
  • Management Science (April 2021)
  • Accepted in The 15th Conference on Web and Internet Economics (WINE), 2019.
  • Finalist, INFORMS Service Science Cluster Best Paper Award, 2018.
• The Power of Opaque Products in Pricing with Adam N. Elmachtoub. [Direct Link] [Slides]
  • Management Science (Jan. 2021)
  • Accepted in The 13th Conference on Web and Internet Economics (WINE), 2017.