A Comparative Study for Sentiment Analysis on Election Brazilian News

摘要

A midia brasileira tern sido acusada ao longo dos anos de favorecer algumas entidades politicas e suas campanhas. Avallar a verdade dessa afirmacao nao e urna tarefa simples dado o grau de subjetividade de quem avalia. Em periodos de eleicao, essa controversia se torna mais acentuada visto a influencia que a midia exerce na opiniao publica. Analise de Senti-mento pode ser urna ferramenta util na avaliacao de noticias politicas. Este trabalho propoe um estudo comparativo de desempenho de tres algoritmos de aprendizagem (Naive Boyes, SVM e MaxEnt) e de tres metodos de selecao de atributos (Qui Quadrado, CPD e CPPD) para classificacao textos relacionados as eleicoes de 2014 para presidente e governador de Sao Paulo. Brazilian news media have been accused to be biased over the years, supporting some political parties and its agendas. To judge this statement as truth or lie is a hard task due its subjectivity. In election periods, this controversy become stronger given the influence of the media in the public opinion. Sentiment Analysis could be a useful tool for evaluate political news. Here is proposed a comparative study test between three learning algorithms (Naive Boyes, SVM and MaxEnt) and three feature selection methods (Chi-Square, CPD and CPPD) for classifying texts related to president and governor of Sao Paulo 2014 elections in Brazil.