Simple linguistic processing effect on multi-label emotion classification

Emotion plays a significant role in human communications in our daily life. With progress in human-machine interface technology, recent research has placed more emphasis on the recognition of emotion reaction. Comparing to some other ideal experimental settings, blog posts online would be respond more to real-world events. And a huge resource of text-based emotion can be found from the World Wide Web nowadays. This paper reports a study to investigate the effectiveness of using SVM (Support Vector Machine) on linguistic features considering emotion keywords and negative words, and classify a collection of blog posts sentences tagged by one or more labels finally. Our results show that individual emotions can be clearly separated by the proposed approach. To the multi-label classification of emotion, it also obtained a higher accuracy rate than the baseline unigram approach using SVM.