Skip to:

Optimizing a Social Network for Desirable Outcomes

Optimizing a Social Network for Desirable Outcomes

Weak signals in online dating networks inform system design.

Social media has dramatically changed how people communicate, shop, do business, and more. The online systems and social networks we use to achieve such tasks affect our experience and outcomes. Small changes to system features can often have unexpected effects. Rigorous studies provide insight designers can use to build systems that satisfy users and improve businesses.

The Study

Like most everything else, dating has gone online. In 2012, more than half of single people started their search for a mate online.

women are 8 times less likely to send the first messageWorking in collaboration with one of the top dating sites in the country, University of Minnesota researchers co-designed a randomized experiment to test the effects of anonymity in the network. The experiment gathered micro-level data to illustrate 
how 100,000 users interacted with two million others on the site.

Many sites offer features like anonymous browsing as part of a premium for-purchase service. In the dating network, anonymous profile browsing removed the ability to leave “weak signals” of interest. As a result, desirable outcomes—namely connections to possible dates—went down, especially for women. In short, the dating site’s model conflicted with social norms and behaviors.

This experiment established a causal relationship between the anonymity feature and the outcome, based on actual human behavior in a live network. This is a good example of how the social graph can become a unique lab for study and learning.

Impact

Dating sites aside, this study offers an important lesson for any company doing business online: small changes in feature design may have big impact on outcome. With so many services, systems, and activities happening online, such differences can translate to real dollars and even the success or failure of some online ventures. Studies like this one demonstrate how the social graph can be used to test system features and measure outcomes, even when the subject is something as elusive as romance.


Bapna, R., Ramaprasad, J., Shmueli, G., and Umyarov, A. (2013), “One-Way Mirrors and Weak-Signaling in Online Dating: A Randomized Field Experiment”, National Bureau of Economic Research Summer Institute on the Economics of IT and Digitization.

Methods

  • Randomized control trial with 100,000 subjects

Tools

  • SAS, R for visualization, zero inflated Poisson model

Co-authors

J.Ramaprasad and G. Shmueli