Design of recommender systems with heterogeneous agents

Abstract

This project aims to study the design of a recommendation system in an environment with heterogeneous agents. Information design is the study of how the information structure available to agents affects the final allocation. There is a recent literature on the optimal information structure in a strategic environment, but little is known about this design when the Principal agent is restricted to some recommendation system. One of our goals is to understand the trade-off of a recommendation system in a heterogeneous agent environment with asymmetric preferences. On the one hand, this preference asymmetry may lead to more experimentation by agents, as agents with extreme preferences are more likely to have incentives to experiment even when public information is not abundantly available. On the other hand, experimenting with a fraction of the population is less informative, as it may only indicate the fact that preferences are different and not that the product in question is actually good. Studying recommendation systems will help us gain a better understanding of how online platforms work, but there are also a variety of other applications that we discuss at the end of this project. (AU)