In this paper, we propose a novel news video mining method based on statistical analysis and visualization. We divide the process of news video mining into three steps: preprocess, news video data mining, and pattern visualization. In the first step, we concentrate on content-based segmentation, clustering and events detection to acquire the metadata. In the second step, we perform news video data mining by some statistical methods. Considering news videos' features, in the analysis process we mainly concentrate on two factors: time and space. And in the third step, we try to visualize the mined patterns. We design two visualization methods: time-tendency graph and time-space distribution graph. Time-tendency graph is to reflect the tendencies of events, while time-space distribution graph is to reflect the relationships of time and space among various events. In this paper, we integrate news video analysis techniques with data mining techniques of statistical analysis and visualization to discover some implicit important information from large amount of news videos. Our experiments prove that this method is helpful for decision-making to some extent.
展开▼
