Community detection is a popular and difficult problem in the field of social network analysis. Most of the current researches mainly focus on optimizing the modularity index, evaluating the similarity of nodes, and designing different models to fit particular networks. These approaches usually suffer from following problems:(1) just a few of them can deal with directed networks as well as undirected networks; and (2) real-world networks being more complex than synthetic networks, many community detection strategies cannot perform well in real-world networks. To solve these problems, this paper presents an algorithm for community detection in complex networks based on random walk method. Different from existing methods based on random walk method, the asymmetric transition probability is designed for the nodes according to network topology and other information. The event propagation law is also applied to the evaluation of nodes importance. The algorithm CDATP performs well on both real-world networks and synthetic networks.