Automatic Detection of Fake News in Spanish: Ecuadorian Political Satire

Abstract

The circulation of fake news on internet, especially those of political satire through social media, has affected the majority of the Ecuadorian population. This work presents a methodology based on statistical learning that accurately and automatically detects fake news in Spanish using machine learning and natural language processing techniques. The document begins by presenting basic concepts related to fake news and works related to their automatic detection. The second section explains the news corpus creation process, text processing, numerical representation with TF-IDF and training of supervised classification algorithms with two different data sets. Results obtained from the training are analyzed in the third section, being the models with support vector machines the ones that offer the best predictions, improving approximately 15%, 6% and 3% to the performance of the models with naive bayes, random forests and boosting trees respectively. Finally, conclusions of the research and future work is presented in the fourth section.