2026-11-18
National Data Administration to Accelerate High-Quality Dataset Development
Source:People’s Daily
Reporters learned from the National Data Administration on November 15 that a conference on pilot initiatives in high-quality dataset development and data annotation was recently held in Xiamen, Fujian Province. At the conference, the National Data Administration outlined plans to accelerate the development of high-quality datasets and promote the growth of the data annotation industry, with greater emphasis on supplying high-quality data to support the advancement of artificial intelligence (AI).
The conference noted that AI model developers’ demand for datasets has shifted from simply having access to data to having the right data in sufficient quantities and of high quality. It emphasized the need to closely align efforts with the evolving landscape of AI development, adopt a demand-driven, scenario-first approach, and clearly identify AI-related data requirements. The aim is to turn the advantages of real-world application scenarios into data advantages, which can then be leveraged to enhance model capabilities and industrial competitiveness.
The meeting also stressed the importance of further distilling and systematizing the approaches to developing high-quality datasets. This framework can be summarized as “one guiding principle, four key components and one closed loop”: scenario-driven development serves as the guiding principle, while the four key components are defining high-quality dataset requirements, data governance, specialized data annotation and quality assessment. These efforts continuously build a closed-loop application cycle around a “data flywheel.”
Greater emphasis should be placed on the role of data annotation in improving dataset quality. Data annotation is evolving from general-purpose annotation to knowledge-intensive annotation involving in-depth expert participation, and from simple labeling to process annotation. Efforts should continue to promote the transition from traditional human-led annotation to a multi-tiered annotation model based on human-machine collaboration and in-depth expert involvement.
Source: People’s Daily

