Influence is a complex and subtle force that governs the dynamics of social networks as well as the behaviors of involved users. Understanding influence can benefit various applications such as viral marketing, recommendation, and information retrieval. However, most existing works on social influence analysis have focused on verifying the existence of social influence. Few works systematically investigate how to mine the strength of direct and indirect influence between nodes in heterogeneous networks. To address the problem, we propose a generative graphical model which utilizes the heterogeneous link information and the textual content associated with each node in the network to mine topic-level direct influence. Based on the learned direct influence, a topic-level influence propagation and aggregation algorithm is proposed to derive the indirect influence between nodes. We further study how the discovered topic-level influence can help the prediction of user behaviors. We validate the approach on three different genres of data sets: Twitter, Digg, and citation networks. Qualitatively, our approach can discover interesting influence patterns in heterogeneous networks. Quantitatively, the learned topic-level influence can greatly improve the accuracy of user behavior prediction.