Sensory Perception & Interaction Research Group

Multipass labelling is one way to add richness and reliability to self-reports of emotions. However, consideration of alternative data sources and perspectives of the same event stream exposes conflicts which must be resolved at analysis time in a way that provides an estimate of confidence and validity. As part of a larger project that is exploring the feasibility of labelling emotion transitions rather than emotion state, this paper aims to initiate a discussion of practical issues and open questions triggered as we proceed to classify and analyze an unusual dataset with triangulated labels: (1) resolving classification details like label resolution differences (continuous vs discrete) and data instance granularity (size of data windows); and (2) considering what it means to have confidence in the consistency or accuracy between different types (passes) of data labelling