Social networks are a means of wide dissemination of ideas and expression of opinions in various fields, the political issue is no exception, arousing much interest with passionate comments, proclamations, opinions, advertising of a p a rticula r ca nd id a te or political party. Twitter, as a widely used social network, allows the publication of short messages that can be obtained through some extraction techniques allowing then to be analyzed. Authorship Attribution presents methods that help to determine the author of a certa in text, as well as the stylistic characteristics of writing that allow to identify a feeling, affinity to a certain idea, etc. This article aims to investigate through experimentation, the possibility of classifying Ecuadorian Twitter users according to their political affinity through the analysis of short texts published in this network, using Machine Learning (ML) techniques for Authorship Attribution. For this purpose, the political parties with the highest vote in the first round of the 2021 presidential elections in Ecuador are taken as a reference. Cla ssification methods such as Support Vector Machine (SVM) and, from Naive Bayes, Bernoulli and Multinomial are evaluated, co m paring them with performance measures to establish which is the most suitable for the proposed task.

No abstract