基于故事的新闻视频事件专题分析方法

News Video Event Topic Analysis Based on Stories

报道同一事件的新闻故事之间存在大量的冗余和重复,这对新闻事件的快速浏览和理解造成了巨大的困难,必须对故事之间的关系进行有效的分析及组织。文章提出一种基于故事的新闻视频事件专题分析方法。融合文本和视觉特征将报道同一事件的新闻故事聚类在一起,通过分析事件内部故事之间的相似性,获得故事之间的相互依赖关系,生成表现事件发展的专题结构。实验结果表明,这种事件专题分析方法能够直观有效地表现新闻事件的发展趋势和相关信息,对于新闻视频的浏览和摘要等实际应用具有重要意义。

News videos contain huge amount of daily information,in which there is a great deal of redundancy and repetition content.Thus it is necessary to analyze news stories＇ relationships and generate news topic effectively.This research proposes an approach for generating news video event topic based on stories.K-means cluster algorithm was used to group topic-evolving stories integrating textual and visual features.Based on the similarity and dependency between stories,an event topic was constructed automatically for the news video topic organization and threading.Finally,the generated event topic structure was visualized in event-time space.Experimental results show that the topic structure generated by the proposed approach can facilitate the fast navigation and understanding of the news topic.