Modeling how information propagates in social networks driven by peer influence, is a fundamental research question towards understanding the structure and dynamics of these complex networks, as well as developing viral marketing applications. Existing literature studies influence at the level of individuals, mostly ignoring the existence of a community structure in which multiple nodes may exhibit a common influence pattern. In this paper we introduce CSI, a model for analyzing information propagation and social influence at the granularity of communities. CSI builds over a novel propagation model that generalizes the classic Independent Cascade model to deal with groups of nodes' (instead of a single node's) influence. Given a social network and a database of past information propagation, we propose a hierarchical approach to detect a set of communities and their reciprocal influence strength. CSI provides a higher level and more intuitive description of the influence dynamics, thus representing a powerful tool to summarize and investigate patterns of influence in large social networks. The evaluation on various datasets suggests the effectiveness of the proposed approach in modeling information propagation at the level of communities. It further enables us to detect interesting patterns of influence, such as the communities that play a key role in the overall diffusion process, or that are likely to start information cascades.