Exécution de notre outil de synthèse sur un nouvel article

Résumé : Query rewriting, which aims to enhance query efficiency by modifying the structure of SQL queries without altering their results, stands as one of the pivotal research directions in the database domain. Traditional query rewriting approaches perform language-equivalent transformations based on manually predefined rewriting rules, thereby failing to fully harness the potential of Materialized Views (MVs) for query optimization. Furthermore, with the emergence of Large Language Models (LLMs), natural language interaction with LLMs can better unlock the potential of MVs in query rewriting. Nevertheless, the direct application of LLMs for query rewriting exhibits several limitations, including overreliance on predefined rules, inadequate accuracy in complex tasks, and insufficient utilization of MVs. To address these challenges, this paper proposes a prototype system called LLM-MVs, which leverages natural language processing with LLMs to generate and select MVs. To further enhance the reasoning capability of LLMs in recommending MVs, we introduce a MV management model that provides LLMs with more efficient MV options. Experimental findings demonstrate that our proposed approach substantially enhances query execution efficiency.