Reproduce “CiteGuard: Faithful Citation Attribution for LLMs via Retrieval-Augmented Validation” 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
Yee Man Choi, Xuehang Guo, Yi R. Fung, Qingyun Wang
CiteGuard, an agent framework, substantially improves faithful citation attribution for Large Language Models by employing retrieval-augmented validation. This approach enhances accuracy to 68.1% with DeepSeek-R1, closely matching human performance, and demonstrates the ability to identify high-quality alternative references across diverse scientific domains.
Key insights
- The newly introduced retrieval actions, particularly `ask_for_more_context` and `search_text_snippet`, are crucial for performance gains, demonstrating that accessing full-text evidence and broader context significantly improves attribution accuracy.
- CiteGuard can effectively identify high-quality alternative citations, going beyond the original ground truth by performing both backward (foundational work) and lateral (related peer work) reasoning.
- There is a practical trade-off between retrieval depth (e.g., full paper text via 'read' action) and computational cost, with snippet retrieval offering a more cost-effective balance for significant accuracy improvements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.