G-Retriever: Retrieval-Augmented Generation for Textual Graph Understanding

Explore a comprehensive talk on G-Retriever, a novel approach for textual graph understanding using retrieval-augmented generation. Delve into the development of a flexible question-answering framework for real-world textual graphs, applicable to scene graph understanding, common sense reasoning, and knowledge graph reasoning. Learn about the Graph Question Answering (GraphQA) benchmark and how G-Retriever enhances graph understanding through soft prompting. Discover how this method outperforms baselines on various textual graph tasks, scales with larger graph sizes, and mitigates hallucination. Gain insights into the future directions of this research and participate in a Q&A session to deepen your understanding of this innovative approach to integrating large language models with graph neural networks.

- Intro + Background
- Review of Graphs and LLMs
- G-Retriever
- Conclusion
- Future Directions
- Q+A