Exploiting the Chronological Semantic Structure in a Large-scale Broadcast News Video Archive for its Efficient Exploration

Recent advance in digital storage technology has enabled us to archive more than 1,700 hours of video data from a daily Japanese news show in the last nine years. In this paper, to effectively make use of the video data in the archive, we first present a news video structuring method based on the chronological semantic relations between stories, namely the “topic thread structure”. Next, we introduce an interface based on the structure, which allows users to track topics along their development and also choose video segments to visually “tell their own stories” using them as source materials. Analyses on the topic thread structures obtained by applying the proposed method to actual news footages revealed interesting relations between topics in the real world, while analyses on their size quantified the efficiency of tracking the topics and finding video materials for post-editing.