Joint Multi-Dimensional Model for Global and Time-Series Annotations

Crowdsourcing is a popular approach to collect annotations for unlabeled data instances. It involves collecting large number of from several, often naive untrained annotators each instance which are then combined estimate the ground truth. Further, constructs such as affect multi-dimensional with rating multiple dimensions, valence and arousal, instance. Most annotation fusion schemes however ignore this aspect model dimension separately. In article we address by proposing generative fusion, models dimensions jointly leading more accurate truth estimates. The propose applicable both global time series problems treats latent variable distorted annotators. parameters estimated using Expectation-Maximization algorithm evaluate its performance synthetic real emotion corpora well on an artificial task human annotations.