•The proposed model can effectively transfer knowledge from auxiliary domains, and evaluate the importance of auxiliary domains.•The proposed model can well alleviate the sparsity problem.•The proposed model can well solve the cold-start problem.•The proposed model can address the scenario of multiple auxiliary domains easily. Cross-domain collaborative filtering, which transfers rating knowledge across multiple domains, has become a new way to effectively alleviate the sparsity problem in recommender systems. Different auxiliary domains are generally different in the importance to the target domain, which is hard to evaluate using previous approaches. Besides, most recommender systems only take advantage of information from user- or item-side auxiliary domains. To overcome these drawbacks, we propose a cross-domain collaborative filtering algorithm with expanding user and item features via the latent factor space of auxiliary domains in this paper. In the proposed algorithm, the recommendation problem is first formulated as a classification problem in the target domain, which takes user and item location as the feature vector, their rating as the label. Then, Funk-SVD decomposition is employed to extract extra user and item features from user- and item-side auxiliary domains, respectively, with the purpose of expanding the two-dimensional location feature vector. Finally, a classifier is trained using the C4.5 decision tree algorithm for predicting missing ratings. The proposed algorithm can make full use of user- and item-side information. We conduct extensive experiments and compare the proposed algorithm with various state-of-the-art single- and cross-domain collaborative filtering algorithms. The experimental results show that the proposed algorithm has advantages in terms of four different evaluation metrics.