Classificação de Polaridade de Sentimentos de Mensagens Curtas Utilizando Word2Vec

Abstract

The Sentiment Analisys and the polarity classification of texts constitutes one of the main tools currently used by companies and organizations for the most varied purposes. Among the many techniques used for this purpose, the vector representation of words, together with the Machine Learning classifiers, is attracting the attention of the scientific community. This work makes an analysis of the use of word embeddings, built through Word2Vec, in the process of features extraction for polarity classification of short messages written in English. The texts used were extracted from Twitter, one of the largest social networks currently used. The results obtained show that, in spite of the possible need to use larger textual bases to obtain better vectors, Word2Vec is a promising tool for the features extraction of textual data, contributing to obtain good classification results.