Portal de Periódicos da CAPES

Neuroevolution for Sentiment Analysis in Tweets Written in Mexican Spanish

2021; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-030-77004-4_10

1611-3349

José-Clemente Hernández-Hernández, Efrén Mezura‐Montes, Guillermo-de-Jesús Hoyos-Rivera, Omar Rodríguez-López,

Advanced Text Analysis Techniques

In this article we propose a special kind of Neuroevolution, called NeuroEvolution of Augmenting Topologies (NEAT), which is based on a genetic algorithm, that is then used to generate an artificial neural network to analyze tweets written in Mexican Spanish, and then labeling them as positive, negative and neutral. Classification performance of neural networks generated through neuroevolution is compared to other Machine Learning approaches, such as Support Vector Machines, Naïve-Bayes, and a handcrafted neural network. This is made through a 10-fold cross validation. As it will be explained in this paper, statistical results suggest that the Neuroevolution neural network generated is simpler than the generated by hand, and gets a competitive performance with respect to the other compared methods.