safe-graph/GNN-FakeNews

Frames fake news detection as a graph classification task by modeling news propagation networks as hierarchical trees, where root nodes represent articles and leaf nodes are users, with multiple node feature types (BERT embeddings, user profiles, content features). Implements multiple GNN architectures (GCN, GNN-CL, BiGCN) under the UPFD framework and integrates with both PyTorch-Geometric and DGL official libraries for standardized dataset access. The UPFD dataset comprises ~5,800 news propagation graphs from Twitter (Politifact and Gossipcop sources) with node-to-user ID mappings and temporal metadata for reproducible evaluation.

544 stars.