Reproduce “Revisiting Feedback Models for HyDE” 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
Nour Jedidi, Jimmy Lin
Researchers at the University of Waterloo integrated traditional information retrieval feedback models, such as Rocchio and RM3, into the HyDE framework for sparse retrieval. This method enhanced query expansion by processing LLM-generated hypothetical documents, yielding average Recall@20 improvements of up to 4.2% overall and 6% in low-resource settings compared to string concatenation techniques.
Key insights
- Traditional feedback models, previously overlooked by many recent LLM-based methods, are highly pertinent and substantially improve HyDE's effectiveness when integrated with sparse retrievers.
- LLM-generated hypothetical documents provide a richer and more effective source for feedback compared to traditional pseudo-relevant documents retrieved by BM25.
- Systematic term selection and weighting mechanisms from traditional feedback models are crucial for optimizing LLM-based query expansion, consistently outperforming naive string concatenation approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.