Preserving User Privacy Through Ephemeral Sharing Design: A Large-Scale Randomized Field Experiment in the Online Dating Context

Abstract

Due to privacy concerns, users of online dating platforms often refrain from voluntary sharing of sensitive personal information in the initial interaction stage. However, the lack of information sharing impedes trust-building, further hurting their probability of securing matches and receiving follow-up conversations from prospective dating partners due to information asymmetry. We propose ephemeral sharing as a privacy-preserving mechanism to navigate the balance between users’ privacy concerns and voluntary information sharing in the initial interaction stage in online dating. Ephemeral sharing refers to the digital design that once the information being shared by a sender is disclosed to a receiver for a relatively short period, it will be invisible and non-retrievable to the receiver in the future. In partnership with an online dating platform, we report a large-scale randomized field experiment with over 70k users to understand how ephemeral sharing influences users’ information sharing, match outcomes, and follow-up conversational engagements. We find that the subjects in the ephemeral treatment group (versus those in the control group) send a significantly more significant number of personal photos along with their matching request and a more significant number of photos disclosing the human face. Meanwhile, the ephemeral sharing treatment also leads to a significant number of more matches. Further, through a set of mediation tests, we find that the observed effect on the number of matches is explained by the increases in the senders’ disclosure of personal photos. The sequential mediation tests further show that the increased sharing of personal photos also induces receivers’ follow-up conversational engagements, mitigating the communication cold-start issue. Lastly, we apply a recursive partitioning algorithm to explore the heterogeneous treatment effects, and the results illustrate that the platform can perform treatment optimization based on user’s popularity, education, and gender. Our study contributes to the literature and practice on the design of matching platforms and the privacy-preserving mechanisms.