Trending AI Content on YouTube

We employed several versions of the trending videos dataset to address different aspects of our analysis. The Raw dataset comprises over 1.13 million records, encompassing all trending videos across various regions and categories. From this, we generated the DeduplicatedAcrossDaysOnly dataset, which preserves regional duplicates but removes instances of the same video appearing on multiple days, resulting in approximately 413,000 entries. This version supports analyses of audience exposure while maintaining information about cross-regional visibility.

For production-oriented analyses, we created the CompletelyUnique dataset, which excludes all duplicates by Video ID, resulting in approximately 138,000 videos, each linked to a single country of origin to ensure accurate regional attribution. Finally, we derived a Sampled dataset of 11,304 videos from the Completely Unique set to enable balanced comparisons between AI and non-AI videos, with stratified representation across both regions and categories. The Recommendations dataset contains videos that appear in YouTube’s recommendation panels, allowing for the investigation of exposure and algorithmic propagation patterns.

本研究采用多版本热门视频数据集，以覆盖分析任务的不同维度。原始数据集（Raw dataset）包含超113万条记录，涵盖全球各区域、各分类的全部热门视频。基于该数据集，我们构建了仅跨日去重数据集（DeduplicatedAcrossDaysOnly）：该数据集保留区域内重复项，但移除同一视频多次上榜多日的重复记录，最终得到约413,000条条目。此版本可支撑受众曝光量分析，同时保留跨区域可见性相关信息。

面向生产场景的分析任务，我们构建了完全唯一数据集（CompletelyUnique）：该数据集按视频ID（Video ID）移除所有重复项，最终得到约138,000条视频条目，每个视频仅关联单一原产国，以确保区域归因的准确性。最后，我们从完全唯一数据集中抽取得到采样数据集（Sampled），共包含11,304条视频，用于实现AI与非AI视频的均衡对比，且该采样集在区域与分类维度上均采用分层抽样策略。推荐数据集（Recommendations dataset）收录出现在YouTube推荐面板中的视频，可用于研究视频曝光与算法传播模式。