As the Internet applications, such as social networks and micro-blogs, become popular, their scale of users has been increasing rapidly. Community detection in these large-scale networks could provide important insights into customer behavior for service recommendation and product marketing. The difference of these networks from traditional ones is that besides topology, they have frequent information interaction between nodes. Information flow makes these networks directed and dynamic. Traditional community detection approaches fall short in these networks because they do not consider these new characteristics. Inspired by the dynamics of infectious disease theory, this paper proposes a novel community detection approach based on information flow analysis. This approach effectively groups the nodes with frequent information interaction in the same community. Between communities, there would be little information flow. This paper experiments on real-world networks demonstrate that compared with previous community detection methods, the proposed approach is more effective at identifying the dynamics in the networks.