ThesisResearchAI/ML 8-12 months
Graph Neural Networks for Social Network Analysis and Fake News Detection
Using Graph Neural Networks (GNNs) to detect fake news by modeling the propagation structure on social networks — capturing both article content and the way real vs fake news spreads through networks.
PythonPyTorch GeometricNetworkXHugging Face TransformersNeo4jFastAPI
Find GitHub repos & source code →How to build it — step by step
- 1Dataset Construction: Use FakeNewsNet / LIAR datasets; build propagation graphs: news article → retweet cascade → user network
- 2GNN Architecture: Implement GraphSAGE or GAT; node features = BERT embeddings of tweet/post text + user credibility scores
- 3Heterogeneous Graph: Model multiple node types (News, User, Tweet); use HAN (Heterogeneous Attention Network) for multi-type graphs
- 4Evaluation: Compare to text-only BERT baseline; ablation study on graph topology vs content contribution
Key features to implement
- ✓Graph-based fake news classification
- ✓Propagation pattern visualization
- ✓User credibility scoring from historical behavior
- ✓Real-time detection API
- ✓Explainability: which subgraph triggered the prediction
💡 Unique twist to stand out
Investigate temporal graph learning — how does the spread pattern at T=1h vs T=24h differ for fake vs real news? Use dynamic graph models to catch fake news in its early propagation phase.
🎓 What you'll learn
Graph neural networks, social network analysis, NLP + graph fusion models, misinformation detection, and research paper writing.