Emotion Classification of Twitter Data Using an Approach Based on Ranking

Abstract

In this work, a model for textual emotion classification based on Ranking technique is presented. The Ranking technique uses the frequencies of words in order to assign a relevance for each in a tweets (Spanish) after calculating the total relevance of the tweet for each classes. The classes are associated to four emotions: happiness, sadness, anger and fear and the highest relevance indicates to which class the tweet belongs. The training and test corpora are created by manually selected key words as references for a crawling tool, both contain manually tagged tweets extracted from Twitter; the training corpus was validated by K-Fold Cross Validation having a 90% of acceptance. The results are compared with Naïve Bayes and Bigrams Probabilities models using precision, recall and F-measure.