A/B testing, also known as bucket testing, split testing, or controlled experiment, is a standard way to evaluate user engagement or satisfaction from a new service, feature, or product. It is widely used in online websites, including social network sites such as Facebook, LinkedIn, and Twitter to make data-driven decisions. The goal of A/B testing is to estimate the treatment effect of a new change, which becomes intricate when users are interacting, i.e., the treatment effect of a user may spill over to other users via underlying social connections.When conducting these online controlled experiments, it is a common practice to make the Stable Unit Treatment Value Assumption (SUTVA) that each individual's response is affected by their own treatment only. Though this assumption simplifies the estimation of treatment effect, it does not hold when network interference is present, and may even lead to wrong conclusion.
In this paper, we study the problem of network A/B testing in real networks, which have substantially different characteristics from the simulated random networks studied in previous works. We first examine the existence of network effect in a recent online experiment conducted at LinkedIn; Secondly, we propose an efficient and effective estimator for Average Treatment Effect (ATE) considering the interference between users in real online experiments; Finally, we apply our method in both simulations and a real world online experiment. The simulation results show that our estimator achieves better performance with respect to both bias and variance reduction. The real world online experiment not only demonstrates that large-scale network A/B test is feasible but also further validates many of our observations in the simulation studies.
Download Full PDF Version (Non-Commercial Use)