Recommendation algorithm is one of the major approaches to solve the information overload problem. The core task is to model and predict users’ preference. The algorithms based on the latent factor model have been made a great success recently. However, data sparseness could lead to the incompleteness of the factors in this completely data-driven modeling. To address this issue, this paper leverages certain knowledge of the influencing factors on user preferences to optimize the structure of latent factor model. This paper proposes a unified model with both explicit factors and latent factors. User demographic features and item content features are used as the clues reflecting users’ preferences. These features are introduced to the framework of latent factor model in the form of explicit factors. Experiments on MovieLens dataset suggest that the proposed method is feasible and effective