Search, Inspect, Fetch: Exploiting Boolean Retrieval for Deep-Research Agents

Hugging Face Daily Papers Papers
deep-research boolean-retrieval retrieval-agents language-agents token-efficiency nlp

Summary

This paper introduces SIEVE, a search-inspect-fetch strategy that uses Boolean Query Language to make deep-research agents retrieve only relevant document sections, achieving higher accuracy with 20.7–50.6% fewer tokens across multiple benchmark datasets and agent backbones.

Existing deep-research agents use a search-visit workflow that retrieves and reads whole pages, without considering the addressable structure that web sources expose through titles, headings, sections, and metadata. This prevents agents from directly constraining retrieval to document fields and often carries irrelevant page content into their context. We introduce SIEVE, a search-inspect-fetch interface driven by fielded Boolean retrieval (BQL). SIEVE filters candidates over document fields, ranks the admitted set, presents structure-rich result cards for inspection, and fetches only selected sections. Across three QA collections, SIEVE achieves higher accuracy than the most accurate conventional Search-Visit configuration on each collection while using 20.7-50.6% fewer tokens. Further analyses show that BQL filtering improves all tested rankers and that the accuracy-context advantage persists across retriever choices and agent backbones. Code and data are available at https://github.com/ielab/skim-search-agent.
Original Article
View Cached Full Text

Cached at: 08/06/26, 01:48 AM

Paper page - Search, Inspect, Fetch: Exploiting Boolean Retrieval for Deep-Research Agents

Source: https://huggingface.co/papers/2608.02751 Deep-research agents usually retrieve and read whole webpages, wasting context on irrelevant sections. We introduce SIEVE, a search-inspect-fetch strategy built on a Boolean Query Language (BQL): the agent searches over document fields (title, section, body, date, etc.) to select eligible pages, ranks the eligible set with any standard ranker (BM25 / dense / BM25+Dense), inspects structure-rich result cards with query-focused snippets, and fetches only the specific section it needs instead of the whole page.

To test this fairly, we built paired flat/structured versions of HotpotQA, MuSiQue, and BrowseComp-Plus so the same content can be accessed with or without addressable structure. SIEVE beats the strongest conventional Search-Visit baseline on accuracy across all three while using 20.7–50.6% fewer tokens, and the gains hold across rankers and three different agent backbones (Tongyi-DeepResearch, Qwen-AgentWorld, OpenResearcher).

Code:https://github.com/ielab/skim-search-agent

Similar Articles

Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus Interaction

Hugging Face Daily Papers

The paper introduces Direct Corpus Interaction (DCI), a novel approach allowing AI agents to query raw text directly using standard terminal tools instead of traditional embedding-based retrieval. By bypassing fixed similarity interfaces and offline indexing, DCI significantly outperforms conventional sparse, dense, and reranking baselines across multiple IR and agentic search benchmarks.