Reproduce “EWEK-QA: Enhanced Web and Efficient Knowledge Graph Retrieval for Citation-based Question Answering Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mohammad Dehghan, Mohammad Ali Alomrani, Sunyam Bagga, David Alfonso-Hermelo, Khalil Bibi, Abbas Ghaddar, Yingxue Zhang, Xiaoguang Li and 7 more
EWEK-QA introduces a system that enhances citation-based question answering by efficiently integrating both unstructured web content and structured knowledge graphs. It achieves superior accuracy and efficiency, largely through an adaptive web retriever and an LLM-call-free knowledge graph retrieval mechanism from Huawei Noah's Ark Lab and the University of Waterloo.
Key insights
- Combining both web and knowledge graph modalities is crucial for achieving high accuracy in citation-based QA across diverse question types, outperforming single-modality approaches.
- Efficient knowledge graph integration is possible by replacing LLM-driven graph traversal and pruning with embedding-based similarity, drastically reducing computational overhead.
- Adaptive passage extraction from the web, beyond simple heuristics, leads to higher quality and more complete evidence for LLMs, improving answer generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.