The Research On Topic Evolution For News Based LDA Model

The goal of news topic evolution is to extract semantic information from news and express the topic. Then, organize the topic change over time. After that, we can organize the news by topic evolution and provide a skeleton of news.Now most method focus on topic trend over time with topic model. Although the methods can got a curve about topic trends, most of them cann't provide the evolution about content of topic. And another problem is that not every one topic generated by topic model can be explained.This paper proposes a model to discover the news topic evolution. First,split the corpus into small one by their timestamp. Then, extract topic of each small corpus, and mark the nonsense topic, relate one topic with another topic adjacent to his timestamp by semantic relation. Experiment is conducted on our corpus of Chinese breaking news and NPC and PCC news. The results indicate that our approach can effectively discover the evolution of news topic.