Ensembles of Methods for Tweet Topic Classification

This paper describes the system we developed for IberEval 2017 on Classification Of Spanish Election Tweets (COSET) task. Our approach is based on a weighted average ensemble of five classifiers: 1) a classifier based on logistic regression; 2) a support vector machine classifier; 3) a Naive Bayes classifier for multinomial models; 4) a Guassian Naive Bayes classifier; and 5) a classifier implementing the k-nearest neighbors vote. Each such classifier was choice taking into account its contributes to the success of the system. The aim is to design a approach by using a voting method, where individual classifiers can have weaknesses. The performance of the ensemble is compared to the individual classifiers, and the experimental results show that the ensemble has better results.