Cyberbullying is a critical socio-technical problem that seriously limits the use of online interaction spaces by different individuals. Emerging literature identifies cyberbullying as a continuous temporal phenomena rather than one-off incidents. However, as of yet, little computational work has been done to model the temporal dynamics of cyberbullying in online sessions. In this work, we model the temporal dynamics of commenting behavior as point processes and validate it over a crowd-labeled cyberbullying data-set of Instagram media sessions. We define several temporal features to model the distinguishing characteristics between cyberbullying and regular media sessions. We find that our approach is successfully able to identify significant differences between cyberbullying and regular media sessions, and provide a performance increase in cyberbullying detection. This paves the way for more nuanced work on the use of temporal modeling to detect and mitigate the occurrence of cyberbullying.