This paper describes the solution of Shanda Innovations team to Task 1 ofKDD-Cup 2012. A novel approach called Multifaceted Factorization Models isproposed to incorporate a great variety of features in social networks. Socialrelationships and actions between users are integrated as implicit feedbacks toimprove the recommendation accuracy. Keywords, tags, profiles, time and someother features are also utilized for modeling user interests. In addition, userbehaviors are modeled from the durations of recommendation records. Acontext-aware ensemble framework is then applied to combine multiple predictorsand produce final recommendation results. The proposed approach obtained0.43959 (public score) / 0.41874 (private score) on the testing dataset, whichachieved the 2nd place in the KDD-Cup competition.