With the growing demands for accessing mobile applications, the cellular network is currently overloaded. Recent work has proposed to exploit opportunistic networks to offload cellular traffic for mobile content dissemination services. The basic idea is to distribute the content object to only part of subscribers (called initial sources) via the cellular network, and allow initial sources to propagate the object through opportunistic communications. The preliminary study focused on selecting a given number of initial sources only based on the probability of encounters between users. However, without consideration of social relationships between users, the selected sources might not be able to propagate the object across different social communities opportunistically. In addition, there exists a dilemma of selecting a suitable number of sources to take the trade-off between offloading cellular traffic and reducing the latency. Hence, in this paper, we propose community-based opportunistic dissemination, which automatically selects a sufficient number of initial sources to propagate the object across disjointed communities in parallel. The trace-based evaluation shows that, compared to encounter-based dissemination, our community-based scheme improves the amount of offloaded cellular traffic up to 29%. In addition, users experience a significantly shorter latency.