Incremental learning for mitigating concept drift in fake news detection in Portuguese

• Grant number: 25/13608-4
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: August 01, 2025
• End date: July 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Renato Moraes Silva
• Grantee: Lucca Baptista Silva Ferraz
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 24/17834-6 - Handling concept drift in fake news detection for the Portuguese language, AP.R

Abstract

The spread of fake news presents a growing problem that can, through misinformation, incite violence, manipulate political decisions, and harm the health and integrity of the population. Machine learning is a widely studied solution for automatically filtering fake news; however, many of these studies work with static models that do not account for the changing nature of news and assume that their characteristics remain unchanged over time. This approach is called offline learning. In this study, it will be shown how a change in concepts over a period of time can deteriorate the accuracy of offline models, an effect known as concept drift, which influences the classification of fake news. Some studies in the literature have already shown that the performance of offline models can be overly optimistic, making it preferable to use incremental learning methods to adapt to changes in textual patterns over time. However, most studies are conducted with news in the English language. This work aims to analyze the impact on Portuguese-language news datasets from the Covid-19 period and the Brazilian presidential elections, to verify whether there was concept drift due to shifts in the focus of the news.