Fine-grained Video Attractiveness Dataset (FVAD)

本研究构建了首个细粒度视频吸引力数据集FVAD，该数据集由全球最受欢迎的视频网站收集而来，包含1019集电视剧，总时长780.6小时，涵盖多种类别和广泛的视频内容。数据集不仅包含大量视频，还记录了数亿用户在观看视频时的行为数据，如观看次数、快进、快退等，这些数据反映了视频吸引力和用户与视频的互动。FVAD的创建旨在解决细粒度视频吸引力预测这一挑战性问题，通过设计不同的序列模型，仅依赖视频内容进行吸引力预测。该数据集的应用领域包括在线营销和视频推荐系统，旨在通过预测视频片段的吸引力来优化广告投放和推荐策略。

This study constructs the first fine-grained video attractiveness dataset (FVAD), which is collected from the world's most popular video websites. The dataset includes 1019 episodes of TV dramas with a total duration of 780.6 hours, covering diverse categories and a wide range of video content. It not only contains a large volume of video materials, but also records hundreds of millions of user viewing behaviors such as play counts, fast-forwarding, fast-rewinding and others, which reflect video attractiveness and user-video interactions. The development of FVAD aims to address the challenging problem of fine-grained video attractiveness prediction, by enabling the design of various sequence models that perform attractiveness prediction solely based on video content. The application scenarios of this dataset cover online marketing and video recommendation systems, with the goal of optimizing advertising placement and recommendation strategies by predicting the attractiveness of video clips.