Event detection and evolution based on entity separation

By computing the relevance of follow-up stories, the traditional topic tracking approaches could track the stories. However, subtopics derived from one topic and the evolution process of these subtopics could not be identified with traditional approaches. A method is proposed to detect event and subtopics and get the evolution process of event from media data. Firstly creating entity vectors with entities in the media data and computing similarity between two entity vectors to separate single event. Then creating full vectors with all keywords in the dataset of an event and computing similarity between two full vectors to get several subtopics of an event and the evolution process of the event. Experiments show that this method can get events and subtopics of them effectively.