Predicting Mood from Digital Footprints

Abstract

The increasing amount of time people spend with digital devices is driving researchers to investigate the impacts of digital technology usage on wellbeing. The debate on whether screen time affects wellbeing has not been settled yet, but the answer is likely to be complex. It is not how much time people spend on devices, but what they do that affects their wellbeing. Understanding the relationship between technology and wellbeing is important in order to raise awareness and to adjust interactions with digital technologies. Large technological companies, such as Google Digital Wellbeing, Apple Screen Time and Microsoft MyAnalytics, already collect digital behaviour data (digital footprints), often to investigate their impact on wellbeing. If such data can be used to predict mood, a continuous mood detection model could be built to predict mood unobtrusively. Continuous mood monitoring is significant because prolonged negative mood and lack of positive moods are common predictors of mental health issues, and the positive balance is one of the definitions of wellbeing (i.e., hedonic). The contributions of this thesis to this issue are (1) a tool (the MindGauge mobile application) to collect self-reports through experience sampling and to obtain digital technology usage data via integration with the RescueTime application; (2) a novel pre-processing and machine learning feature extraction method for mood prediction that includes feature generation from digital duration information and data mining techniques, such as clustering and frequent pattern mining; (3) a comparison of mood prediction models that exploit one or more of the extracted features to show how these features influence mood prediction performances. Overall, this study presents new results to better understand the relationship between digital activity and wellbeing. The methods proposed in this thesis provides important experimental bases to improve how scientists analyse and study digital technology usage.