Classifying emotions in Twitter messages using a deep neural network

Abstract

Many people use social media nowadays to express their emotions or opinions about something. This paper proposes the use of a deep learning network architecture for emotion classification in Twitter messages, using the six emotions model of Ekman: happiness, sadness, anger, fear, disgust and surprise. We collected the tweets from a labeled dataset that contains about 2.5 million tweets and used the Word2Vec predictive model to learn the relations of each word and transform them into numbers that the deep network receives as input. Our approach achieved a 63% accuracy with all the classes and 77% accuracy on a binary classification scheme.