China's National Data Bureau said on Nov. 10 an on-site meeting in Xiamen called
for accelerating the build-out of high-quality datasets and the data-annotation
industry. It said model firms' demand has shifted from simply having data to
having the right, sufficient, high-quality data, and urged a demand-led,
scenario-first approach to convert real-world scenario advantages into data
advantages, then into model capability and industrial competitiveness. The
bureau outlined a methodology — one main line, four stages, one closed loop —
with a scenario-driven mainline and four stages: demand definition, data
governance, professional annotation and quality evaluation, forming a continuous
'data flywheel'. It also stressed that professional data annotation is key to
improving dataset quality.
