微博中话题的传播分析及热点预测

Microblog is playing a more and more important role in spread of public opinion. The contents of many microblogs are around the same event, which constitute a topic. The hotspot prediction and propagating analysis of topics could help people keep up with the latest trend about events, which can be used in the mining of news and the regulation of the bad information. We divide the propagating process into three stages: incubation, outbreak and recession period, use the four features to describe the propagation, such as the proportion of VIP, distribution of the fans number, etc. We propose a classification method based on a latent source model, and use the timing signal to represent the trends of topics. The result shows that, we can detect trends before Sina does 83.5% of the time, with a mean value of 1.6 hour in advance. The effectiveness of the method is demonstrated.