Using a Genetic Algorithm Approach to Study the Impact of Imbalanced Corpora in Sentiment Analysis

The SVM classifier has been used in many methods to identify emotions in text due to their good generalization capability and robustness with high dimensionality data. However, most textual corpora usually subject to such methods are naturally imbalanced. As a consequence, the SVM, sensitive to imbalance data, assigns to most texts the majority class. In this article, we present a Genetic Algorithm based approach that aims to reduce the imbalance of the data in the context of emotions identification. This approach allowed us to study the impact of its application in a method of emotion identification in texts written in the Brazilian Portuguese. Experimentations showed us that balancing the corpus could be an alternative when using the SVM classifier for emotions identification, especially in a multiclass configuration.