Boolean queries are all you need?

2026-07-13

AI Summary

This work investigates whether efficient document retrieval in Retrieval-Augmented Generation (RAG) can be achieved using only Boolean queries. To this end, the authors propose a Boolean query formulation grounded in a restricted subset of regular languages, integrated with a large language model agent and an unsupervised substring density scoring mechanism that ranks documents solely based on the count and length of matching substrings—without requiring supervised learning, global statistics, or term weighting. Evaluated on the deduplicated MS MARCO V2.1 corpus and 86 topics from the TREC 2024 RAG Track—with a budget of 100 model calls per topic—the method achieves an NDCG@10 of 0.6863, outperforming various dense, sparse, and learned sparse retrievers. These results demonstrate the surprising efficacy of simple pattern matching in intelligent retrieval settings.

Abstract

We equipped an LLM-based search agent with access to a Boolean retrieval engine to search the MS MARCO V2.1 deduped segment collection used by the TREC 2024 RAG track. Over a standard track subset of 86 topics, and operating under a budget of 100 model calls/topic, the agent achieved an NDCG@10 of 0.6863, which would place it above many dense, sparse, and learned-sparse first-stage retrievers. Ranking is based solely on the density of corpus substrings matching a query, with no requirement for supervised learning, global statistics, or term weights. Formally, the query language expresses a strict subset of the regular languages, with a document's score based on the number and length of matches it contains. Although the results are more exploratory than definitive, because they are based on a single test collection that was publicly available during model training, they suggest that simple pattern matching may be sufficient for agentic search.

Problem

• Boolean queries
• retrieval
• pattern matching
• RAG
• LLM-based search

Innovation

• Boolean retrieval
• LLM-based search agent
• pattern matching
• unsupervised ranking
• regular languages