[The on-site meeting for the pilot construction of high-quality data sets and data annotations was held in Xiamen City] On November 14, according to the National Data Administration, on the afternoon of November 10, the on-site meeting for the pilot construction of high-quality data sets and data annotations was held in Xiamen City. The meeting emphasized that the National Data Administration will thoroughly implement the decisions and arrangements of the Party Central Committee and the State Council on the development of artificial intelligence, and accelerate the construction of high-quality data sets and the development of the data annotation industry. The first is to pay more attention to adapting to the development of artificial intelligence and providing high-quality data. Model companies’ demand for data sets has shifted from “whether there is data” to “whether there is correct, sufficient and high-quality data”. It is necessary to closely focus on the development trend of artificial intelligence, adhere to demand traction and scenario priority, clarify the data requirements for artificial intelligence, turn real scenario advantages into data advantages, and then transform them into model capabilities and industrial competitiveness. The second is to pay more attention to the summary of high-quality data set construction methods. The construction method of high-quality data sets can be summarized as "one main line, four links, and one closed loop", that is, the scene-driven construction is the main line, focusing on the four links of high-quality data set requirement definition, data governance, professional annotation, and quality assessment, and continuously forming a "data flywheel" application closed loop. The third is to pay more attention to the supporting role of data annotation in improving the quality of data sets. Data annotation is upgrading from general annotation to knowledge-intensive with deep participation of experts, and from simple label to process annotation. It is necessary to continue to promote the gradual transformation of data annotation from the traditional "human-centered" model to a multi-level annotation model with human-machine collaboration and in-depth participation of experts.