Communication in social networks tends to exhibit complex dynamics both in terms of the users involved and the contents exchanged. For example, email exchanges or activities on social media may exhibit reinforcing dynamics, where earlier events trigger follow-up activity through multiple structured latent factors. Such dynamics have been previously represented using models of reinforcement and reciprocation, a canonical example being the Hawkes process (HP). However, previous HP models do not fully capture the rich dynamics of real-world activity. For example, reciprocation may be impacted by the significance and receptivity of the content being communicated, and modeling the content accurately at the individual level may require identification and exploitation of the latent hierarchical structure present among users. Additionally, real-world activity may be driven by multiple latent triggering factors shared by past and future events, with the latent features themselves exhibiting temporal dependency structures. These important characteristics have been largely ignored in previous work. In this dissertation, we address these limitations via three novel Bayesian nonparametric Hawkes process models, where the synergy between Bayesian nonparametric models and Hawkes processes captures the structural and the temporal dynamics of communication in a unified framework. Empirical results demonstrate that our models outperform competing state-of-the-art methods, by more accurately capturing the rich dynamics of the interactions and influences among users and events, and by improving predictions about future event times, user clusters, and latent features in various types of communication activities.