Platform Design in Curated Dating Markets

Motivated by online dating apps, we study how to select subset of profiles to show to each user in each period in a two-sided matching platform. Users on each side observe the profiles set by the platform and decide which of them to like. A match occurs if and only if two users mutually like each other, potentially in different periods. The goal of the platform is to maximize the total expected number of matches. We study different platform designs, varying (i) how users interact with each other, i.e., whether one or both sides of the market can initiate an interaction, and (ii) the timing of matches, i.e., whether the platform allows non-sequential matches in addition to sequential ones. We focus on the case with two periods and study the performance of different approaches. First, we show that algorithms that exploit the submodularity of the problem and properties of its feasible region can achieve constant factor approximation guarantees that depend on the platform design, ranging from $1-1/e$ to $1/3$. Finally, we show that the Dating Heuristic (DH) (Rios et al., 2023), which is commonly used and achieves good performance in practice, provides an approximation guarantee of $1-1/e$ for all platform designs. We show theoretically and empirically that the performance of the DH is robust to the platform design. Our simulation results -- using real data from our industry partner -- also show that platforms using a one-directional design should initiate interactions with the side that leads to the smallest expected backlog per profile displayed, balancing size and selectivity. Moreover, we find that a one-directional design can lead to at least half of the matches obtained with a two-directional design. Finally, our results show that avoiding non-sequential matches has no sizable effect, regardless of the platform design.