Criação de um modelo de classificação de tweets depressivos utilizando máquina de vetores de suporte

Abstract

The increase of social network usage around the world has encouraged people to express and share their feelings and opinions on a particular topic or subject. From the patterns of these data, one can create prediction models using Natural Language Processing techniques in conjunction with Machine Learning algorithms. This work proposes the creation of a classification model of depressive tweets using the method of pattern recognition SVM (Support Vector Machine), having as data source two public domain datasets taken from the social network Twitter. The development environment used was the Jupyter Notebook, where, with the help of several auxiliary libraries, pre-processing techniques were implemented and applied on the tweets to eliminate undesirable information. The training of the constructed model obtaining a rate of success of 99.7%with the data of training after the model is optimized using the Search Grid technique. Using the created model, it was possible to label new tweets as depressive or non-depressive.