Popularity Bias in Online Dating Platforms: Theory and Empirical Evidence.

This article investigates popularity bias in recommendation algorithms on online dating platforms and its impact on users' chances of finding compatible partners and the platform’s revenue generation. Using data from a major South Korean dating platform, the study models the platform’s recommendations and user interactions as a three-stage matching game involving recommendations, message sending, and message acceptance. The analysis reveals that platforms maximizing revenue tend to recommend popular users more frequently, creating bias against less popular users, but this bias can also increase both revenue and successful matches when popular users are not overly selective. A predictive model based on machine learning validates these findings through simulations, showing that unbiased recommendations yield fewer matches and lower revenue, especially when users face low congestion costs from incoming messages. An extension to a two-period model suggests that early-stage platforms may reduce bias to build reputation and grow their user base, whereas mature platforms prioritize revenue, increasing popularity bias.