High-Quality Datasets in AI for Humanities: Construction and Value Creation

Transforming dormant cultural resources into high-quality, computable, and interconnected datasets is a key task in advancing the integration of culture and technology. Drawing on a collection of representative humanities-AI datasets presented at the 1st CCF AI for Humanities Conference, this article examines the field from two perspectives: dataset construction and value creation. Humanities data differ from general-purpose data in semantic richness, contextual dependence, relational structure, cultural sensitivity, and evaluation criteria. Their quality depends fundamentally on expert-driven annotation, supported by the preservation of relational structure and by advanced computational techniques. These datasets support both cutting-edge humanities research and applications in cultural industries and broader society, transforming cultural resources into new quality cultural productive forces. The article further discusses shared challenges concerning annotation expertise and scalability, the applicability of general-purpose methods, the availability of data and computing resources, interdisciplinary collaboration, and the governance and long-term stewardship, providing practical insights for the construction and application of humanities-AI datasets.