Classification of feelings expressed in texts on social networks through natural language processing techniques

Abstract

To understand how the application of the main NLP techniques improves the performance of a model, a database of more than 280 thousand records was compiled, since there is no good basis for training and testing natural language processing in Portuguese. In Brazil, most of the material is in English or has been automatically translated. Therefore, the model was trained without applying any technique and then each technique was applied, recording the results of each technique, in order to compare all the techniques at the end and understand how the performance gain was at each stage. After performing all the techniques, it became clear that NLP is essential for working with any type of text in data science, because in our model we were able to increase the accuracy by 19.47%.