Análise de sentimentos utilizando uma base de treinamento limitada

Abstract

Sentiment analysis is a classification algorithm that needs labeled data to train your neural network. This work raised the question whether it was possible to create a neural network from a small database, given the difficulty of manually labeling large masses of data. From the observation of the manually classified base, criteria were raised and automatic labeling was performed on more data in order to obtain a training base with the necessary amount of classified data. Finishing with the training of a model and gauging its metrics based on the test base, obtaining a satisfactory result of 77% of accuracy.