DuVideoSenti多模态情感标签数据集

面向推荐场景推出多模态情感标签数据集DuVideoSenti，引入视频情感泛标签预测任务，构建了视频情感泛标签体系。该体系由人工定义的“文艺清新”、“时尚炫酷”、“舒适温馨”等11个情感泛标签组成，用以刻画用户浏览视频后的视觉和情感方面的感受。数据集为每个小视频标注了情感泛标签，并且提供了视频的标题、帧图特征用于模型的训练和预测。

We present the multimodal emotion tag dataset DuVideoSenti tailored for recommendation scenarios, which introduces the video emotion general tag prediction task and constructs a video emotion general tag taxonomy. This taxonomy consists of 11 manually-defined general emotion tags such as "literary and fresh", "fashionable and cool", and "cozy and warm", designed to depict users' visual and emotional perceptions after browsing videos. Each short video in this dataset is annotated with a general emotion tag, and the dataset also provides video titles and frame image features for model training and prediction.