Predicting the Emotional Intensity of Tweets

Abstract

Automated interpretation of human emotion has become increasingly important as human computer interactions become ubiquitous. Affective computing is a field of computer science concerned with recognizing, analyzing and interpreting human emotions in a range of media, including audio, video, and text. Social media, in particular, are rich in expressions of peoples’ moods, opinions, and sentiments. This thesis focuses on predicting the emotional intensity expressed on the social network Twitter. In this study, we use lexical features, sentiment and emotion lexicons to extract features from tweets, messages of 280 characters or less shared on Twitter. We also use a form of transfer learning – word and sentence embeddings extracted from neural networks trained on large corpora. The estimation of emotional intensity is a regression task and we use, linear and tree-based models for this task. We compare the results of these individual models as well as making a final ensemble model that predicts the emotional intensity of tweets by combining the output of the individual models. We also use lexical features and word embeddings to train a recently introduced model designed to handle data with sparse or rare features. This model combines LASSO regularization with grouped features. Finally, an error analysis is conducted and areas that need to be improved are emphasized.