Against the backdrop of data becoming a new type of production factor, libraries need to leverage their data resource advantages and advance the construction of high-quality datasets to support artificial intelligence applications. After clarifying the conceptual connotation and sorting out the construction requirements of library high-quality datasets, this study develops an evaluation indicator system and partitions construction stages. Using this framework, it examines domestic and international practices of library high-quality datasets construction and identifies shortcomings in standardized data organization, compliant data governance, and data application and optimization. It further proposes future development paths, including deepening data processing and organization for AI applications, improving full-process governance and tiered sharing mechanisms, and building a feedback-driven data iteration mechanism, with a view to providing references for libraries to leverage the advantages of their data resources in the AI era.