Improving Opinion Classifiers by Combining Different Methods and Resources

Lucas V. AvançoHenrico B. BrumMaria G. V. Nunes

This work describes the design of systems aimed at classifying thepolarity of opinions at document level, especially for reviews of products inBrazilian Portuguese on the web. For the binary polarity classification (positiveor negative), some Lexicon-based and Machine Learning approaches, as wellas a combination of them in a hybrid manner and as an ensemble of classifiers,have been implemented and evaluated. The F1-measures values have reached0.84 (Lexicon-based approach) and 0.95 (Machine Learning approach), whichare in accordance to the best ones from similar systems for other languages.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: