A Survey of Query Optimization in Large Language Models

ACL ARR 2025 May Submission1947 Authors

Abstract: \textit{Query Optimization} (QO) refers to techniques aimed at improving the operational efficiency and response quality of Large Language Models (LLMs) in processing complex queries, particularly within Retrieval-Augmented Generation (RAG) frameworks. RAG dynamically retrieves current external information to complement model knowledge as a cost-effective solution addressing LLMs' tendencies to generate factually inconsistent outputs. With recent advancements expanding RAG into multi-component systems, QO has become pivotal for optimizing the evidence retrieval phase - critically determining the system's ability to source accurate, multi-faceted supporting information for query resolution. Effective query optimization strategies directly enhance information retrieval performance (e.g., improving recall rates of evidentiary documents) while indirectly strengthening the model's semantic comprehension and final response generation. This paper systematically examines the developmental trajectory of QO techniques through a comprehensive analysis of seminal research. By establishing a structured categorization framework, we aim to synthesize existing QO methodologies in RAG implementations, clarify their technical underpinnings, and emphasize their transformative potential for expanding LLM capabilities across diverse applications.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: Query Optimization, Large Language Models

Contribution Types: Surveys

Languages Studied: English

Keywords: Query Optimization, Large Language Models

Submission Number: 1947