We examine how user activity graphs on third-party Facebook applications evolve over time. Using user activity data spanning a year on three Facebook applications, we show how graph-theoretic properties and their temporal variations are influenced by the maturity, popularity, and user engagement level of the applications. We also examine the heavy sub-graphs consisting of strong links, i.e., user-pairs with large number of interactions. We find that the general evolution trend of the graph properties remain unchanged although the absolute values differ. By analyzing individual usage patterns over time, we identify super heavy users in our gaming application (i.e., 5% of top users account for 80% of interactions), but not in social utility applications. We also see 'addictive' behavior on our gaming application, where heavy users remain dominant for months. Our findings are useful for application developers and OSN providers in devising provisioning and advertising strategies. |