基于上下文信息的新闻故事单元分割

News Story Segmentation Based on Contextual Information

为了有效检索和管理新闻视频资源,提出了一种利用多模态特征、基于上下文信息的新闻故事单元分割方法.首先利用字幕变化、音频类型和视频镜头类型信息将新闻镜头序列转换成为相应的关键词序列,使新闻故事单元分割转换成为文体序列分割的问题;然后应用一种有效的序列分割模型——条件随机场,来分割新闻故事单元.对多段视频的测试结果证明了该方法的有效性,取得了查全率为72.9%、查准率为88.4%的较好结果.

To effectively retrieve and manage news video archive, a news video story segmentation method based on con- textual information was proposed. Firstly, news video shot sequence was transformed into corresponding keyword sequences by using multimodality information, such as caption changing,audio type, and video shot category. And then news video was partitioned into stories with a powerful sequence segmentation model, conditional random fields. The algorithm was applied to multiple test videos, and the test results prove the effectiveness of the method, achieving a better performance with recall of 72.9% and precision of 88.4%.