How Early Can We Detect? Detecting Misinformation on Social Media Using User Profiling and Network Characteristics

The rise of social media has amplified the need for automated detection of misinformation. Current methods face limitations in early detection because crucial information that they rely on is unavailable during the initial phases of information dissemination. This paper presents an innovative model for the early detection of misinformation on social media through the classification of information propagation paths and using linguistic patterns. We have developed and incorporated a causal user attribute inference model to label users as potential misinformation propagators or believers. Our model is designed for early detection of false information and includes two auxiliary tasks: predicting the extent of misinformation dissemination and clustering similar nodes (or users) based on their attributes. We demonstrate that our proposed model can identify fake news on real-world datasets with 86.5% accuracy within 30 min of its initial distribution and before it reaches 50 retweets, outperforming existing state-of-the-art benchmarks.