Instance Genetic Selection for Fuzzy Rule-based Systems Optimization to Opinion Classification

Abstract

Opinion Mining aims to identify people’s feelings about some item of interest based on content available on the Web, without the user having to find and read all the news about it. As opinions are related to feelings that are often described by imprecise terms, Fuzzy Logic appears as an alternative to treat the information subjectivity. However, the commonly used textual databases are extensive and make it difficult to achieve accuracy with a low computational cost of the Fuzzy System, since the algorithms used to generate the rules base usually result in bases with many rules. To deal with this problem, databases can be reduced in order to save only relevant data to extracting information. So, methods to reduce databases have beenproposed,withemphasisonInstanceSelection,whichselects a subset of instances that can be used to generate classification models with the same precision as the models generated from the original set. Thus, the aim of the present study is develop a mechanism to optimize the rules base of a Fuzzy System, using the instances selection, in order to generated a reduced rules baseconsideringsatisfactoryperformanceintheclassification.As the problem mentioned is a multiobjective problem, which seeks to increase precision and reduce the number of rules, we have chosen to apply a Multiobjective Genetic Algorithm, since this approach has been shown to be promising in the literature. The results demonstrate that the Multiobjective Genetic Algorithms canbeappliedintheinstancesselectionforopinionsclassification problems, presenting a reduction in the number of instances and the execution time, without significant changes in precision.