In user-oriented article collection system, users construct their collection sets by adding articles which they are interested in, studying why users desire specific articles and find the specific articles is particularly an interesting issue in social science. In this paper, we present the prediction users' preference on articles by considering content information, i.e. semantic information and tags. In this study, we propose a new model which jointly performs deep representation learning for the content information and collaborative filtering for the collection matrix. Extensive experiments on the real-world datasets show that it can significantly advance the state-of-the-art.