ADORE: Iterative Query Expansion with Retrieval-Grounded Relevance Feedback

2026-06-11

AI Summary

Current query expansion methods driven by large language models often lack validation against the target corpus, leading to retrieval drift or loss of critical information. This work proposes ADORE, a novel framework that introduces, for the first time, a retrieval-driven relevance feedback mechanism to establish a closed-loop iterative process between generation and retrieval. In each iteration, the framework generates pseudo-paragraphs using a large language model, retrieves corresponding corpus responses via dense and sparse retrievers, and dynamically refines the expanded queries through a relevance evaluator. This approach significantly enhances the precision and robustness of query expansion, consistently outperforming state-of-the-art methods across the TREC Deep Learning, BEIR, and BRIGHT benchmarks, with notable gains of 3.6% and 9.2% in nDCG@10 on BEIR and BRIGHT, respectively.

Abstract

LLM-based query expansion improves retrieval by enriching the original query with additional context. Yet most methods remain generation-driven, producing plausible pseudo-documents or expansions without checking how the target corpus responds. This can introduce retrieval drift, amplify misleading vocabulary, or miss terms that distinguish relevant from non-relevant documents. We argue that effective expansion requires retrieval-grounded feedback, not just single-pass generation or unverified iteration. We introduce ADORE (ADapt, Observe, Relevance Evaluate), an iterative framework that turns retrieval outcomes into feedback for the next expansion. At each round, an LLM generates pseudo-passages, a retriever exposes the corpus response, and a relevance assessor evaluates retrieved documents against the original query. These judgments identify what to reinforce, what remains undercovered, and what to suppress. Across TREC Deep Learning, BEIR, and BRIGHT, ADORE consistently outperforms strong query expansion baselines with notable improvements across nearly all evaluation settings, improving average nDCG@10 by 24.5% over BM25 and 3.6% over the strongest prior query expansion method on BEIR, and by 122.9% over BM25 and 9.2% over the best query expansion baseline on BRIGHT. Our code and data are publicly available.

Problem

• query expansion
• retrieval drift
• relevance feedback
• pseudo-documents
• corpus response

Innovation

• iterative query expansion
• retrieval-grounded feedback
• relevance evaluation
• pseudo-passage generation
• LLM-based retrieval