Topic Tracking Based on Event Network

Topic detection and tracking (TDT) is a hot issue in text processing, which allows people to efficiently access interesting information they need from large amounts of narrative reports, and obtain the cause of a certain event and its subsequent events. In this paper, event network instead of vector space model (VSM) is used to represent the contents of texts. Firstly, event networks are constructed by using chapter structure and event relationships. Secondly, WGN algorithm is introduced to community discovery and network reduction. Finally, a topic model based on event-weight vector is built. Event vector similarity calculation is utilized to determine whether the new report belongs to known topics, and thus achieve the goal of the topic tracking task.