This document discusses LinkedIn's efforts to personalize users' news feeds by ranking the large number of heterogeneous activity updates. Personalization models were developed to predict a user's click-through rate based on affinities between the user and activity types as well as between the user and actors. Testing showed that personalization achieved higher click-through rates than non-personalized approaches. The models were deployed at large scale to rank billions of possible user-activity pairs daily based on past interactions. Further work aims to personalize at finer levels such as by activity topic or individual user.