A Corpus-based Multi-label Emotion Classification using Maximum Entropy

Thanks to the Internet, blog posts online have emerged as a new grassroots medium, which create a huge resource of text-based emotion. Comparing to other ideal experimental settings, what we obtained from the World Wide Web evolve and respond more to real-world events. In this paper, our corpus consists of a collection of blog posts, which annotated as multi-label to make the classification of emotion more precise than the single-label set of basic emotions. Employing a maximum entropy classifier, our results show that the emotions can be clearly separated by the proposed method. Additionally, we show that the microaverage F1-scores of multi-label detection increase when the mount of available training data further increases.