Unified Transformer-Graph Framework for Cross-Platform Misinformation Detection and Knowledge Propagation Analysis

The massive growth of online social media networks has contributed greatly to misinformation propagation. The currently used misinformation detection systems are usually single platform and are based on textual semantics or network propagation patterns, which do not adequately cover the interactions at heterogeneous social media ecosystems. In order to overcome these issues, this paper presents a framework of Unified Transformer-Graph Framework (UTGF) which combines a contextual semantic learning task with a Transformer encoder and a relational reasoning task with a Graph Neural Network (GNN). The model semi-simultaneously learns textual data, user metadata, and propagation graphs on several benchmark datasets such as Twitter15/16, Pheme, PolitiFact, and GossipCop, which is a collection of about 100,000 labeled posts. The end-to-end training of the model is done with AdamW optimizer and dropout regularization and early stopping in order to pursue strong generalization. The experimental findings show that UTGF performs better with accuracy of 94.5, F1-score of 0.945, MAE of 0.087 and RMSE of 0.112 and low inference latency of about 25 ms per sample.