Summary: | In the sentiment classification process, the quality of the polarity varies depending on the characteristics or attributes possessed by the classifier and those of the tweet being analyzed; therefore, a sentiment classifier achieves its highest quality in scenarios in which its characteristics are similar to the characteristics of the tweet. This article presents SentiFuzzy, an algorithm that, based on the characterization of attributes of five sentiment classifiers recognized in the literature, implemented a series of inference rules and fuzzy sets, which allowed to define mathematical weights for each classifier; thus, to know which classifier should be selected according to the nature of the analyzed tweet. Additionally, these weights were optimized by the Hill-Climbing optimization algorithm, which yielded, in some scenarios, a higher polarity accuracy than that reported in the state of the art and, in other cases, a competitive polarity accuracy compared to the polarity reported by the compared classifiers.
|