QueStER: Query Specification for Generative keyword-based Retrieval

2025/11/07 by Satouf, Arthur, Zong, Yuxuan, Amadou-Boubacar, Habiboulaye +2
Computer Science · #68P20 #68T50 #Computation and Language (cs.CL) #FOS: Computer and information sciences #H.3 #Image Retrieval and Classification Techniques #Information Retrieval (cs.IR) #Information Retrieval and Search Behavior #Machine Learning (cs.LG) #Topic Modeling

paper · doi:10.48550/arxiv.2511.05301

openalex publication_date 2025/11/07 · openalex created_date 2025/11/11 · openalex updated_date 2026/07/28

Abstract

Generative retrieval (GR) differs from the traditional index-then-retrieve pipeline by storing relevance in model parameters and generating retrieval cues directly from the query, but it can be brittle out of domain and expensive to scale. We introduce QueStER (QUEry SpecificaTion for gEnerative Keyword-Based Retrieval), which bridges GR and query reformulation by learning to generate explicit keyword-based search specifications. Given a user query, a lightweight LLM produces a keyword query that is executed by a standard retriever (BM25), combining the generalization benefits of generative query rewriting with the efficiency and scalability of lexical indexing. We train the rewriting policy with reinforcement learning techniques. Across in- and out-of-domain evaluations, QueStER consistently improves over BM25 and is competitive with neural IR baselines, while maintaining strong efficiency.

Citations

• Decoding Dense Embeddings: Sparse Autoencoders for Interpreting and Discretizing Dense Retrieval
• Rational Retrieval Acts: Leveraging Pragmatic Reasoning to Improve Sparse Retrieval
• Open-Reasoner-Zero: An Open Source Approach to Scaling Up Reinforcement Learning on the Base Model
• Grounding by Trying: LLMs with Reinforcement Learning-Enhanced Retrieval
• DeeperImpact: Optimizing Sparse Learned Index Structures
• RaFe: Ranking Feedback Improves Query Rewriting for RAG
• Large Language Model based Long-tail Query Rewriting in Taobao Search
• GLEN: Generative Retrieval via Lexical Index Learning
• Auto Search Indexer for End-to-End Document Retrieval
• Enhancing Conversational Search: Large Language Model-Aided Informative Query Rewriting
• Semantic-Enhanced Differentiable Search Index Inspired by Learning Strategies
• Generative Retrieval via Term Set Generation
• How Does Generative Retrieval Scale to Millions of Passages?
• Generative and Pseudo-Relevant Feedback for Sparse, Dense and Learned Sparse Retrieval
• Learning to Tokenize for Generative Retrieval
• Precise Zero-Shot Dense Retrieval without Relevance Labels
• Ultron: An Ultimate Retriever on Corpus with a Model-based Indexer
• Autoregressive Search Engines: Generating Substrings as Document Identifiers
• ColBERTv2: Effective and Efficient Retrieval via Lightweight Late Interaction
• SPLADE v2: Sparse Lexical and Expansion Model for Information Retrieval
• BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models
• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
• Dense Passage Retrieval for Open-Domain Question Answering
• Passage Re-ranking with BERT
• Representation Learning with Contrastive Predictive Coding
• Attention Is All You Need
• Task-Oriented Query Reformulation with Reinforcement Learning
• ExpandR: Teaching Dense Retrievers Beyond Queries with LLM Guidance

Related

• Retrieval-in-the-Chain: Bootstrapping Large Language Models for Generative Retrieval
• Generative Query Reformulation for Effective Adhoc Search
• Scalable and Effective Generative Information Retrieval
• Beyond [CLS] through Ranking by Generation
• Generative Query Reformulation Using Ensemble Prompting, Document Fusion, and Relevance Feedback