Identificação de Polaridade de Sentimento no Twitter Aplicada à Indústria Calçadista

Abstract

There are several works reported in the scientific literature on sentiment classification, with the extraction of messages from the Twitter platform. However, no work was found specifically focused on the Portuguese language for the footwear area. The article shows how it is possible to recognize consumer opinion (positive or negative) from tweets regarding the footwear industry, using machine learning. A footwear company from southern Brazil was used for evaluation. We collected texts from Twitter, which had the preprocessing process with the cleaning of irrelevant terms, the extraction of characteristics to obtain measurements and the differentiation of polarity. And finally, the identification of which class or example under analysis belongs to the use of classifiers for polarity recognition. The classifiers used were Support Vector Machines (SVM), Multi-Layer Perceptron (MLP), Random Forest, Nearest Neighbors (KNN) and Linear Discriminant Analysis (LDA). The results showed that the best classifier for this type of problem was the MLP. The results with the MLP classifier obtained specificity of 78.5%, sensitivity of 95.6% and an accuracy of 86.0%.