Research For Social Network Based On Semantic Analysis And Graph Mining

With the development of the Internet, an increasing number of people and topics are active at the online social networks. The bursty topic detection, which supports the detection of the burst and special events, becomes one of the hottest research areas in social media mining. By now, there have been a lot of research efforts on bursty topic in the information flow, topic models and propagation models. But the following problems still exist:the lack of mathematical definition of bursty topic, and the insufficient of social attribute study in bursty topic. In this context, we proposed new models in bursty topic detection.First of all, we developed a JS(Jensen-Shannon) distance based model for bursty topic detection. And we proposed two algorithms to solve the model:the random walk algorithm and the time split algorithm. Experiments prove that the model can effectively solve the problem of large-scale social network data of bursty topic detection.Secondly, taking into account the impact of the author of documents, we developed a joint model of social influence and topic model to detect the burst topics. In contrast with traditional model experiments, the new model achieved better experimental results in the accuracy of bursty topic detection.