To improve the user experience as well as business outcomes,social platforms aim to predict user behavior. To this end,recurrent models are often used to predict a user’s next be-havior based on their most recent behavior. However, peoplehave habits and routines, making it plausible to predict theirbehavior from more than just their most recent activity. Ourwork focuses on the interplay between ephemeral and cyclical components of user behaviors. By utilizing user activitydata from social platform Snapchat, we uncover cyclic andephemeral usage patterns on a per user-level. Based on ourfindings, we imbued recurrent models with awareness: we aug-ment an RNN with a cyclic module to complement traditionalRNNs that model ephemeral behaviors and allow a flexibleweighting of the two for the prediction task. We conductedextensive experiments to evaluate our model’s performance onfour user behavior prediction tasks on the Snapchat platform.We achieve improved results on each task compared againstexisting methods, using this simple, but important insight inuser behavior: both cyclical and ephemeral components mat-ter. We show that in some situations and for some people,ephemeral components may be more helpful for predictingbehavior, while for others and in other situations, cyclicalcomponents may carry more weight.