Discovering Types of Smartphone Usage Sessions from User-App Interactions

Abstract

Understanding how and why people use their smartphones has enabled use cases ranging from correlating behaviour with psychological states through to on-device tasks such as app recommendations. However, being able to effectively and pervasively capture usage behaviour is challenging due to the wide range of functions, apps and interactions that are possible. In this paper, we examine how embedding physical user-app activity (e.g., taps and scrolls) can provide a rich basis for summarising device usage. Using a large dataset of 82,758,449 interaction events from 86 users over an 8-week period we combine feature embedding and unsupervised learning to extract prominent interactions within clusters of smartphone usage sessions. We find that high-level features such as session length, unlock state, and app switches are not representative of these clusters and can give a false sense of similarity or dissimilarity between sessions. The results motivate further exploration of the utility of using user-app interaction behaviour as the basis for the aforementioned use cases.

Smartphone App Usage Analysis: Datasets, Methods, and Applications

“…App usage is still strongly linked to the usage context, i.e., prior and posterior used apps. The fact that several apps need to work together to complete a single task causes such correlations [94], [95]. Rahmati et al [96] first demonstrated the app usage dependency of one-nearest prior used apps and found that such a dependency remains relatively constant for one to three months.…”