Avaliacao De Modelos Baseados Em Aprendizagem De Maquina Supervisionado Para Tarefas De Reconhecimento De Emocoes Em Textos Curtos.

Abstract

The technological advances of the 4.0 revolution made it possible to create devices and software that aim to improve and/or facilitate the lives of individuals, especially in terms of human-computer interaction, a concept that is immersed in everyday life and has a strong link with the quality of life. Therefore, this work aims to present and evaluate the performance of models based on supervised Machine Learning for tasks of recognition of emotions joy, sadness and surprise in short texts in the Portuguese language collected in comments of empathic games. The implementations were carried out with the aid of the Scikit-Learn library, using the Support Vector Machine (SVM), Naive Bayes (NB), K-Nearest Neighbors (KNN), and Ensemble methods, which were trained using a predefined database classified. The development of the application is divided into 6 stages of evaluation, where in each stage it was sought to apply a different technique of data pre-processing, thus making it possible to ascertain the impact of each technique on the performance of the models. Results show that the SVM model reached 71% accuracy in scenarios with a large presence of preprocessing functions, while the NB and Votting models act respectively with 77% accuracy in scenarios with little preprocessing robustness and 77% accuracy in both scenarios.