Towards Hybrid Profiling: combining digital phenotyping with validated survey questions to balance data entry effort with predictive power

Tailoring apps based on user traits has attracted tremendous interest in developing mHealth apps, and understanding a users' personality is a key challenge in that context. This challenge is typically addressed via classic surveys, which pose a regrettably high burden on app users. This study aims to reduce the response burden of personality tests by introducing a model for predicting the user personality based on digital footprints of app usage. At the same time, skipping surveys completely turns out to undermine prediction accuracy. Therefore, this paper conceptualizes a hybrid framework that utilizes user event data in combination with surveys that have fewer questions than conventionally. The proposed method demonstrates a promising trade-off between the simplicity of using user event data and the accuracy of the validated survey methods: when applying the hybrid method to a retrospective case study, the accuracy is higher than when using the event data exclusively. Also, the number of survey questions needed is significantly lower. Since this is a novel method, we expect that results will strengthen as larger data sets available over time. To facilitate that process, we also present a mathematical model for rationalizing the design process of hybrid profiling systems. This may boost adoption by developers who aim to implement the method in their specific app.