Reproduce “Progressive Query Expansion for Retrieval Over Cost-constrained Data Sources” 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
Muhammad Shihab Rashid, Jannat Ara Meem, Yue Dong, Vagelis Hristidis
ProQE is an iterative query expansion algorithm that integrates pseudo-relevance feedback with large language models to enhance retrieval accuracy while minimizing the number of documents retrieved from cost-constrained data sources. It achieves an average gain of 37% in MRR and R@1 compared to baseline retrievers, proving effective across both sparse and dense retrieval systems.
Key insights
- In certain real-world applications, such as legal or medical databases, the dominant financial constraint for information retrieval is the per-document retrieval cost, rather than the cost of Large Language Model (LLM) API calls.
- A hybrid approach that combines the iterative, document-grounded feedback loop of pseudo-relevance feedback (PRF) with the advanced semantic understanding and generation capabilities of LLMs can effectively mitigate the noise of classic PRF and the hallucination issues of purely generative LLM methods.
- Optimal retrieval performance can be achieved with a relatively small number of progressive iterations (around 5), suggesting a robust and cost-efficient design for systems that incrementally access documents.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.