Handling concept drift in fake news detection for the Portuguese language

• Grant number: 24/17834-6
• Support Opportunities: Regular Research Grants
• Start date: July 01, 2025
• End date: June 30, 2028
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Renato Moraes Silva
• Grantee: Renato Moraes Silva
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated researchers: Anderson Raymundo Avila

Abstract

The spread of fake news is a growing problem that, through misinformation, can incite violence, manipulate political decisions, and harm the health and well-being of the population. Machine learning is a commonly studied solution to automatically filter fake news, but many studies rely on offline learning, which creates static models that don't adapt to changes in news patterns over time. This study will investigate how the evolution of news patterns, a phenomenon known as concept drift, can reduce the accuracy of offline models in classifying fake news. Some studies have already shown that the performance of offline models can be overly optimistic, making it better to use incremental learning methods that adapt to changing text patterns over time. However, most of these studies are focused on news in English. This project aims to analyze the impact of concept drift in Portuguese-language news during the Covid-19 pandemic and the Brazilian presidential elections, to see if there was a concept drift due to changes in the news focus. The task of fake news detection will be evaluated from two perspectives: binary classification and stance detection. (AU)