Artificial intelligence and the digital economy are core characteristics of our era. This paper explores the connotative characteristics, logical structure and construction path of library data collections in the digital and intelligent era and proposes that libraries need to shift from document management to data services against the backdrop of the development of AI technology. Data collection is the knowledge data set combination for intelligent agents of different granularities formed by the digitization of literature, characterized by machine-oriented, interconnected, scenario-driven and multi-granular. The logical composition of the data collection includes document corpus data sets, library user datasets, library business activity data sets, and application datasets. The construction path of the data collection system should include system architecture and content architecture. Among them, data architecture, data collection evaluation, and data value co-creation constitute key issues that constrain the development of data collection resource construction’s satisfy the service requirements of smart libraries, methods for data collection planning and dataset construction are proposed, with an emphasis on leveraging AI technology for data architecture design and processing to fully unleash the data value of library resources.