Modeling Behaviour to Predict User State: Self-Reports as Ground Truth

Julian Frommel,Regan L Mandryk

DOI: https://doi.org/10.48550/arXiv.2007.14461

2020-07-29

Abstract:Methods that detect user states such as emotions are useful for interactive systems. In this position paper, we argue for model-based approaches that are trained on user behaviour and self-reported user state as ground truths. In an application context, they record behaviour, extract relevant features, and use the models to predict user states. We describe how this approach can be implemented and discuss its benefits in comparison to solely self-reports in an application and to models of behaviour without the selfreport ground truths. Finally, we discuss shortcomings of this approach by considering its drawbacks and limitations.

Human-Computer Interaction,Machine Learning