Abstract: Collaborative filtering is oneofthe most successful recommendation technologies, and the quality of collaborative filtering is determinedby the accuracy of the nearest neighbors. Data sparsity problem and similarity metricsseriously affect the choice of the nearest neighbors. Different from traditional recommendation tasks, in this paper, we propose an improved meta path-based collaborative filtering algorithm for weighted heterogeneous information networks. Firstly, we calculate the similarity among users based on different meta path by utilizing the rich semantic information and attribute information in weighted heterogeneous networks. Then we apply the similarity to user-based collaborative filtering algorithm and get multiple predicted rating scores based on different similarity. Finally we calculate the final predicted scores by combining various meta path information using supervised machine learning algorithms. The method is evaluated with the extended MovieLens dataset and experimental results show that our approach outperforms several traditional algorithms and make the result of recommendation more accurate in terms of accuracy.