A New Role for Relevance: Guiding Corpus Interaction in Agentic Search

2026-07-27

AI Summary

This work addresses the limitations of existing retrieval agents, which rely solely on relevance filtering and struggle to accurately locate, combine, and verify evidence in complex question answering. Direct corpus interaction approaches, lacking relevance guidance, suffer from delayed clue discovery and slow convergence. To overcome these issues, the paper proposes a relevance-guided intelligent search mechanism that leverages multi-granularity relevance—both document-level and passage-level—as an execution prior. This mechanism dynamically orchestrates the entire corpus interaction process by optimizing document traversal order, initializing with query-relevant passages, and re-ranking matching results. Implemented within a RipGrep-based framework integrated with large language models, the approach significantly outperforms both traditional and direct-interaction baselines on complex browsing-style QA and high-reasoning retrieval tasks, achieving a notable trade-off between accuracy and efficiency.

Abstract

Relevance is a query-dependent estimate of whether a document or excerpt contains useful evidence. Existing retrieval agents use relevance to select top-$k$ content, but document relevance alone cannot localize, compose, or verify the evidence required by complex questions. Direct Corpus Interaction (DCI) enables such fine-grained operations through grep-style exploration, but its relevance-agnostic search can expose useful clues late and delay convergence. Recent advances use relevance to narrow the corpus into a working space for interaction. Once interaction begins, however, relevance still does not directly guide which documents grep searches first or distinguish informative excerpts from a broad set of matches to let LLMs see them first. We introduce the Relevance-Aware RipGrep Search Agent (RARG), which turns relevance into an execution prior for corpus interaction. RARG provides coarse-to-fine relevance guidance: it orders documents for sequential 'ripgrep' traversal to expose globally relevant clues earlier, initializes promising entry points with query-relevant paragraphs, and reranks grep matches to surface informative excerpts that document-level ranking may otherwise obscure. Across challenging browse question answering and reasoning-intensive retrieval, RARG improves the accuracy--efficiency frontier over retrieval-based and direct-interaction agents. These results demonstrate that relevance-aware interaction enables faster and more reliable search convergence.

Problem

• relevance
• corpus interaction
• retrieval agent
• evidence localization
• search convergence

Innovation

• Relevance-aware search
• Direct Corpus Interaction
• RipGrep
• Retrieval agent
• Execution prior