Abstract: Mobile personal assistants aim at addressing users’ information needs by anticipating their actions at different points in time. One such application which has been the focus of researchers recently, is regarding the anticipation of users’ app usage patterns on their smartphones. The rapid proliferation of smartphone applications, have changed these mobile devices from mere communication tools to means for accessing personalized content that fit various needs and tastes. In this paper, we propose a novel method that given a user’s previous smartphone activities and their contexts, predicts the user’s activity at different times and under certain contexts. Such prediction could be used to organize content on a mobile phone in a personalized fashion such that users would need less time to access their desired content. Our temporal model captures local patterns of actions of a user over consecutive time slices. Our experimental results using an app usage dataset demonstrate the efficacy of our proposed method outperforming two major state-of-the-art baselines, namely, the Singular Value Decomposition (SVD), and the Author Topic Model (ATM).
Loading