Accurate models of smartphone call and text logs are needed to quantify social behavior. Current models for social communication tend to focus on volume related measures only. It is of great interest to capture the burstyness, when events occur rapidly followed by long intervals of no activity, as well as the circadian pattern, taking into account when events happen during the course of day, which are typically seen in human social communication dynamics. We propose a SeaHawk Process, an extension of the self-exciting Hawkes process, to allow for circadian patterns. We compare this approach with competing methods through simulation and in a digital phenotyping study of schizophrenia.