Subtopic Based Topic Evolution Analysis

Traditional topic tracking approaches can obtain the relevant stories. However, the relationship between stories occurred during the topic developing process can not be exhibited clearly. By analyzing the evolution of the topic, the concept subtopic is put forward and the focus of topic evolution analysis is subtopic instead. Four levels topic model is constructed. The subtopic detection algorithm, time slices partition algorithm and topic evolution analysis algorithm are designed. These algorithms make use of the temporal characteristic of topic, partition the news stories into time slices and compute the similarity of these units. As a result, the relationships between the various subtopics in the process of the topic evolution are achieved. Experiments show our algorithms are effective.