Fake News Detection and Fact Checking in X posts from Ecuador Chequea and Ecuador Verifica using Spanish Language Models

Resumen

Currently, verifying news content before its dissemination poses a significant challenge due to the rapidity with which it spreads and the ease of replication. These factors contribute to the proliferation of fake news. Collaborative initiatives like Duke Reporters' Lab and the International Fact-Checking Network (IFCN) have been established to enhance the accuracy of fact-checking to combat various forms of disinformation. The accredited fact-checking platforms in Ecuador are Ecuador Chequea and Ecuador Verifica. This paper details the outcomes from five transformer-based models, namely BETO, MarIA, RoBERTuito, BERTuit, and BERTin, for classifying fake news in Spanish. The rating system of Ecuador Chequea and Ecuador Verifica validated the news gathered from these platforms' accounts on the social network X (Formally known as Twitter), including X posts generated between January 2020 and March 2024. The findings validate that in terms of accuracy, recall, precision, and F1 score, the MarIA language model outperforms other Spanish-based models such as BERTin, RoBERTuito, BETO, and BERTuit.