REDUCING RECOMMENDATION INEQUALITY VIA TWO-SIDED MATCHING: A FIELD EXPERIMENT OF ONLINE DATING

Leading dating platforms usually recommend only a small fraction of users based on users' popularity and similarity, leading to recommendation inequality. We use a stylized matching model from economics to modify existing algorithms to reduce inequality. We evaluate the proposed method through a large-scale field experiment on a dating platform. Experiment results suggest that our recommender reduces inequality, improves predictive accuracy, and leads to substantially more matched couples than other competing algorithms.