Improving emotion recognition from text with fractionation training

Abstract

Previous approaches of emotion recognition from text were mostly implemented under keyword-based or learning-based frameworks. However, keyword-based systems are unable to recognize emotion from text with no emotional keywords, and constructing an emotion lexicon is a tough work because of ambiguity in defining all emotional keywords. Completely prior-knowledge-free supervised machine learning methods for emotion recognition also do not perform as well as on some traditional tasks. In this paper, a fractionation training approach is proposed, utilizing the emotion lexicon extracted from an annotated blog emotion corpus to train SVM classifiers. Experimental results show the effectiveness of the proposed approach, and the use of some other experimental design also improves the classification accuracy.