Experimenting with distant supervision for emotion classification

Abstract

We describe a set of experiments using automatically labelled data to train supervised classifiers for multi-class emotion detection in Twitter messages with no manual intervention. By cross-validating between models trained on different labellings for the same six basic emotion classes, and testing on manually labelled data, we conclude that the method is suitable for some emotions (happiness, sadness and anger) but less able to distinguish others; and that different labelling conventions are more suitable for some emotions than others.