Automatic emotion classification for interpersonal communication

We introduce a new emotion classification task based on Leary's Rose, a framework for interpersonal communication. We present a small dataset of 740 Dutch sentences, outline the annotation process and evaluate annotator agreement. We then evaluate the performance of several automatic classification systems when classifying individual sentences according to the four quadrants and the eight octants of Leary's Rose. SVM-based classifiers achieve average F-scores of up to 51% for 4-way classification and 31% for 8-way classification, which is well above chance level. We conclude that emotion classification according to the Interpersonal Circumplex is a challenging task for both humans and machine learners. We expect classification performance to increase as context information becomes available in future versions of our dataset.

Publication year:2011