Bootstrapped Learning of Emotion Hashtags #hashtags4you

We present a bootstrapping algorithm to automatically learn hashtags that convey emotion. Using the bootstrapping framework, we learn lists of emotion hashtags from unlabeled tweets. Our approach starts with a small number of seed hashtags for each emotion, which we use to automatically label tweets as initial training data. We then train emotion classifiers and use them to identify and score candidate emotion hashtags. We select the hashtags with the highest scores, use them to automatically harvest new tweets from Twitter, and repeat the bootstrapping process. We show that the learned hashtag lists help to improve emotion classification performance compared to an N-gram classifier, obtaining 8% microaverage and 9% macro-average improvements in F-measure.