To Build a National-Level NLP/New Media Laboratory

Challenge: NLP annotation is subjective and difficult, demanding annotators with strong Chinese language skills and a good grasp of political context. During the initial stages of the project, there's a lack of clear annotation requirements and a complete annotation rule system. Moreover, the articles are lengthy, with 22 major labels and a total of 70-80 labels and options. There are various article types, covering nine different scenarios, and multiple data formats, including text, images, and videos.

Solution: To address these challenges, we use the automated exam system. This system helps by setting the right standards for annotators based on specific needs, ensuring high-quality annotations. For improving our annotation rules, we have a dedicated PM who offers immediate feedback and multiple solutions during the annotation process. This drives the optimization of our annotation guidelines. Additionally, for complex annotation scenarios and requirements, the Stardust system comes with pre-annotations and automated QC, greatly boosting both efficiency and quality in annotation.

"With the power of computation, we seek insights from data and seamlessly integrate everything into media, embracing the era of AI." The integration of AI with the media is the future of the media industry.