Finding “Interesting” Correlations in Multi-Faceted Personal Informatics Systems

Personal Informatics (PI) systems are capable of uncovering interesting insights about their users by identifying statistical correlations in multi-faceted data. However, they often produce an overwhelming quantity of information. We explore the feasibility of automatically filtering correlational information, based on its interest to users. We analyze users’ subjective ratings of correlations within their data to gain a deeper understanding of the factors that contribute to users’ interest. We then use this understanding to identify candidate measures for information filtering, which can be applied without input from the user. Finally, we test the predictive power of these measures. Our main findings reveal that users in our study valued the Surprisingness and Utility of correlational information above other factors.