Twitter is a microblogging platform that allows users to share opinions with a restricted amount of characters. Given the social characteristic of Twitter, it is a potential source for sentiment analysis. For this reason, opinions of certain subjects such as people or brands can change in short periods of time. A traditional approach of a sentiment classifier implementation performs poorly since it depends on how it is trained. We propose a novel method for tackling this problem, with the implementation of a multi-agent system for classifying and corpus analysis mechanism for retraining the classifier. This mechanism consists of a critic agent which compares the trained corpus of the classifier agent with new collections of documents from future time steps, using primarily two methods: hypothesis test analysis and histogram differences. A Naïve-Bayes based classifier was implemented with this mechanism with multiple configurations. The results of experimental data show that the mechanism boosts its performance, when compared to a pure Naïve Bayes classifier.
CITATION STYLE
Rodríguez, S., Alfaro, R., Allende-Cid, H., & Cubillos, C. (2017). Automatic tweets classification under an intelligent agents framework. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10283 LNCS, pp. 295–311). Springer Verlag. https://doi.org/10.1007/978-3-319-58562-8_23
Mendeley helps you to discover research relevant for your work.