Redes neurais profundas para auxílio à tomada de decisão no mercado de ações

Abstract

This work presents a Deep Learning application composed by Convolutional Neural Networks and Recurrent Neural Networks (Long Short-Term Memory) for semantic extraction from portuguese language news and technical indicators processing from Banco do Brasil stock (BBAS3). Furthermore, it applies entropy filters and ensemble methods to create an investment decision support system. Throughout this research, we discuss the advantages of using different techniques of Natural Language Processing and sliding window training. The results of experiments indicate a high potential for the joint application of these techniques, since they reach good assertiveness in stock movements predictions and achieve higher financial returns than market benchmarks during the period.