EMOTEX: Detecting Emotions in Twitter Messages

Abstract

Social media and microblog tools are increasingly used by individuals to express their feelings and opinions in the form of short text messages. Detecting emotions in text has a wide range of applications including identifying anxiety or depression of individuals and measuring well-being or public mood of a community. In this paper, we propose a new approach for automatically classifying text messages of individuals to infer their emotional states. To model emotional states, we utilize the well-established Circumplex model that characterizes aective experience along two dimensions: valence and arousal. We select Twitter messages as input data set, as they provide a very large, diverse and freely avail- able ensemble of emotions. Using hash-tags as labels, our methodology trains supervised classiers to detect multiple classes of emotion on potentially huge data sets with no manual eort. We investigate the utility of several features for emotion detection, including unigrams, emoticons, negations and punctuations. To tackle the problem of sparse and high dimensional feature vectors of messages, we utilize a lexicon of emotions. We have compared the accuracy of several machine learning algorithms, including SVM, KNN, Decision Tree, and Naive Bayes for classifying Twitter messages. Our technique has an accuracy of over 90%, while demonstrating robustness across learning algorithms.