In topic tracking, a topic is usually described byseveral stories. How to represent a topic is always an issueand a difficult problem in the research on topic tracking. Toemphasis the topic in stories, we provide an improved topicbasedtf*idf weighting method to measure the topicalimportance of the features in the representation model. Toovercome the topic drift problem and filter the noise existedin the tracked topic description, a dynamic topic model isproposed based on the static model. It extends the initialtopic model with the information from the incoming relatedstories and filters the noise using the latest unrelated story.The topic tracking systems are implemented on the TDT4Chinese corpus. The experimental results indicate that boththe new weighting method and the dynamic model canimprove the tracking performance.
展开▼