Automatic categorization of tweets on the political electoral theme using supervised classification algorithms

Abstract

The increase and use of social networks to share content and opinions with different characters allows to have a large volume of information. Twitter, is just one of the most used social networks and has been selected for this study; the users of this network they become not only passive actors of reception and consumption of information, they are also generators of contents. Tweets analysis requires a systematic process for collecting, processing and classification, which is why this article determines the best Classifier categories: positive, negative and neutral public opinion corresponding electoral political issues. For this, a total of 745 tweets collected in Spanish from the main accounts of media, political figures and political organizations of Ecuador. These tweets were preprocessed, transformed and the results indicated that the vector support machines (SMO) with a sensitivity error rate (RECALL) of 0.8% proved to be the best. Likewise, the algorithm Syntetic Minority Over-Sampling Technique was used (SMOTE) to balance classes and increase capacity predictive of the models excluding the decision trees for the categorization of this type of tweets.