Recommender Systems for Online Dating

Users of large online dating sites are confronted with vast numbers of candidates to browse through and communicate with. To help them in their endeavor and to cope with information overload, recommender systems can be utilized. This thesis introduces reciprocal recommender systems that are aimed towards the domain of online dating. An overview of previously developed methods is presented, and five methods are described in detail, one of which is a novel method developed in this thesis. The five methods are evaluated and compared on a historical data set collected from an online dating website operating in Finland. Additionally, factors influencing the design of online dating recommenders are described, and support for these characteristics are derived from our historical data set and previous research on other data sets. The empirical comparison of the five methods on different recommendation quality criteria shows that no method is overwhelmingly better than the others and that a trade-off need be taken when choosing one for a live system. However, making that trade-off decision is something that warrants future research, as it is not clear how different criteria affect user experience and likelihood of finding a partner in a live online dating context