---

🧠 BiGCN-CLIP Fusion Net (BCFN) for Rumor Detection

This repository contains a minimal implementation of the BCFN model proposed in our paper:

"Multimodal Fusion for Rumor Sleuthing: A Comprehensive Approach"
Mohammad-Reza Farahi, Fateme Jafarinejad
Published in Expert Systems with Applications, 2025

📄 Paper Link
📧 Correspondence: jafarinejad@shahroodut.ac.ir

---

🧰 About the Project

The BCFN (BiGCN-CLIP Fusion Net) model fuses graph structure and text semantics for robust rumor detection on social media. It integrates:

• 🔁 Bi-Directional GCNs (Bi-GCN) for modeling rumor propagation networks
• ✍️ CLIP (text encoder) for extracting rich semantic features from textual posts
• 🎯 Cross-modal multi-head attention for fusing modalities
• 🧮 A lightweight MLP classifier for final prediction

---

📊 Datasets

We evaluate BCFN on three benchmark rumor detection datasets:

• Twitter15
• Twitter16
• Weibo

⚠️ Due to licensing issues, raw datasets are not included. Please follow the original papers to obtain them.

---

📦 Requirements

This code was tested using:

• Python ≥ 3.9
• PyTorch ≥ 1.12
• PyTorch Geometric (PyG)
• Transformers (for CLIP)
• scikit-learn
• tqdm

---

🧪 Results

Our model achieves state-of-the-art performance on all datasets. See the paper for full tables and ablation studies.

---

📌 Citation

If you use this code or paper in your work, please cite:

@article{farahi2025bcfn,
  title={Multimodal fusion for rumor sleuthing: A comprehensive approach},
  author={Farahi, Mohammad-Reza and Jafarinejad, Fateme},
  journal={Expert Systems with Applications},
  volume={288},
  year={2025},
  doi={10.1016/j.eswa.2025.128327}
}

---

📚 Acknowledgements

CLIP by OpenAI

Bi-GCN implementation inspired by [Bian et al., 2020]

Datasets from [Ma et al., 2016, 2017]

---

🛠 Status

This is a quick and dirty release of the source code. Bug reports, feature requests, or contributions are welcome!

---

☕ Contact

For questions, feel free to reach out via email: 📧 rqlzienc@gmail.com