新华江苏政情数据集
收藏资源简介:
数据集为文本、图像、视频三模态融合,计划汇聚包含200万条政务新闻文本、80万张现场高清图、2万小时4K/8K视频。文本已做实体、事件、政策主题三级标签;图像完成人脸、场景、文字检测;视频已按镜头切分、关键帧抽取、字幕对齐。整体数据规模约230TB,可按“政策—人物—时间—地点”四维交叉检索。
This is a multimodal dataset integrating text, image and video modalities. It aims to collect 2 million government affairs news texts, 800,000 high-definition on-site images, and 20,000 hours of 4K/8K video content. The text data has been annotated with three-level tags including entities, events and policy themes; the image data has completed face, scene and text detection; the video data has undergone shot segmentation, key frame extraction and subtitle alignment. The total data volume of the dataset is approximately 230 TB, and it supports cross-retrieval across four dimensions: "policy", "person", "time" and "location".



