Improving Short-Content Misinformation Detection Using Multiple Aspect Trajectories Classification Techniques

• Juan Pablo Chavarro Sanchez UFSC
• Tarlis Tortelli Portela IFPR
• Jônata Tyska Carvalho UFSC

Abstract

The proliferation of fake news and misinformation on social media poses a significant threat to social integrity. While there are approaches such as automatic content analysis and artificial intelligence detection to mitigate this problem, these methods face challenges in classifying misinformation in short content, like that posted on microblogging platforms or instant messaging services. In this study, we present an innovative approach to misinformation detection that combines content-based detection with multi-aspect trajectory classification. This approach models information propagation by considering each shared message as a moving object within the network, enriching the propagation trajectory with different semantic aspects. We implemented this approach on a misinformation dataset collected from WhatsApp and compared it with traditional content-based detection methods, and we combined these two detection orientations into a hybrid approach. Our findings indicate that the proposed hybrid approach achieves an F1-score of 0.89, surpassing baseline models by 8% in misinformation detection, even in short content. These results suggest that combining content-based detection with multi-aspect trajectory classification is a novel and promising strategy for addressing misinformation on social media.