Reproduce “CuriousLLM: Elevating Multi-Document Question Answering with LLM-Enhanced Knowledge Graph Reasoning” 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
Zukang Yang, Zixuan Zhu, Xuan Zhu
CuriousLLM, developed by researchers at the University of California, Berkeley, advances multi-document question answering by employing an LLM-enhanced agent that generates context-driven follow-up questions to guide knowledge graph traversal. This framework achieves superior accuracy and significantly improves efficiency, reducing training costs by 85% and enabling faster information retrieval through an early termination mechanism.
Key insights
- Guiding knowledge graph traversal through dynamically generated follow-up questions, rather than predicting missing evidence, leads to more accurate and less error-prone information retrieval.
- Implementing an early termination signal, based on the LLM's assessment of query completeness, dramatically improves retrieval efficiency and reduces latency without compromising answer accuracy.
- Specialized training on a synthetic dataset designed for follow-up question generation enables a smaller, efficient LLM to achieve superior performance in complex multi-document reasoning tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.