Beyond modeling private actions: predicting social shares

We study the problem of predicting sharing behavior from e-commerce sites to friends on social networks via share widgets. The contextual variation in an action that is private (like rating a movie on Netflix), to one shared with friends online (like sharing an item on Facebook), to one that is completely public (like commenting on a Youtube video) introduces behavioral differences that pose interesting challenges. In this paper, we show that users' interests manifest in actions that spill across different types of channels such as sharing, browsing, and purchasing. This motivates leveraging all such signals available from the e-commerce platform. We show that carefully incorporating signals from these interactions significantly improves share prediction accuracy.