Movie101

Movie101是由中国人民大学信息学院创建的大型中文电影数据集，旨在帮助视觉障碍人士通过自动电影叙述系统更好地理解电影内容。数据集包含30,174个叙述片段，总计92小时，这些片段在演员无对话时提供角色意识强的叙述段落。数据集通过自动处理和人工校正从西瓜视频平台的无障碍频道收集的电影中提取叙述和演员台词，并爬取相关元信息。Movie101支持多种电影理解任务，如电影片段叙述和时间叙述定位，旨在解决自动电影叙述中的实际应用需求，并通过新提出的电影叙述评分（MNScore）进行评估，该评分与人类评估高度一致。Movie101 is a large-scale Chinese movie dataset created by the School of Information, Renmin University of China, aiming to assist visually impaired individuals in better comprehending movie content through automatic movie narration systems. The dataset comprises 30,174 narration segments, with a total duration of 92 hours, providing character-aware narration passages during intervals when actors have no dialogue. Narration content and actor lines are extracted from movies sourced from the accessibility channel of Xigua Video platform via automatic processing and manual correction, and relevant metadata is also crawled. Movie101 supports multiple movie understanding tasks, such as movie segment narration and temporal narration localization. It is designed to address practical application demands in automatic movie narration, and is evaluated using a newly proposed Movie Narration Score (MNScore), which exhibits strong consistency with human evaluations.