Seek improves document retrieval by revisiting and self-evaluating results
Seek: Self-Evaluative Exploration for Knowledge Retrieval
Information Retrieval
Summary
Finding all relevant documents in a large collection can be hard because many systems look at the data only once and miss important information. The authors created Seek, a method where an AI model repeatedly checks and reconsiders which documents might be useful, learning from its own previous guesses. This iterative approach helps it find more relevant documents and improves the ranking quality compared to older methods. The method works well on tests involving complex reasoning and large text collections without needing extra training.
What this means in practice
- •For search engine developers: Improve search result quality by iteratively retrieving and re-evaluating documents to recover missed relevant content without extra training.
- •For legal document teams: Enhance retrieval of relevant legal passages by using an iterative method that refines search results based on ongoing relevance assessments.
Authors
Amin Bigdeli, Radin Hamidi Rad, Negar Arabzadeh, Sajad Ebrahimi, Hai Son Le, Charles L. A. Clarke, Ebrahim Bagheri
Abstract
LLM-based retrievers and rerankers have advanced passage ranking, yet both paradigms interact with the corpus in a single pass and commit to the resulting candidate set, leaving relevant documents permanently unrecoverable once missed. We introduce Seek, Self-Evaluative Exploration for Knowledge Retrieval, a training-free framework that addresses this limitation through iterative corpus interaction at test time. At each round, an LLM generates pseudo-passages conditioned on accumulated relevance feedback, a retriever surfaces fresh candidates, and a dedicated assessor assigns graded relevance judgments that guide subsequent rounds. On TREC Deep Learning, Seek matches trained rerankers in ranking quality while consistently improving Recall@100 over single-pass BM25. On the reasoning-intensive BRIGHT benchmark, Seek with Qwen2.5-7B achieves an 82% relative gain over BM25, surpassing all trained baselines, and Seek with GPT-4.1 reaches 37.4 average nDCG@10, exceeding the strongest baseline by 37%.