Discovery of Relevance between Location and Topics of Micro-blogs

Historical travel’s records are useful for discovering preferences of a person in order to predict his travel behavior, do social recommendation and understand his identity. This paper presents a comprehensive statistical framework for mining the relevance between preferences on location and content of geo-tagged micro-blog in order to deeply understand travel motivation. Our method models the generation of travel history of a person as a process of sampling from some topics represented his location interests, furthermore, we consider the generation of micro-blogs as a similar process sampling from topics of text, the topics of micro-blog is related to the topics of location interests. To handle large-scale data set from the social network and estimate parameters of the model, we implement a training program based on a Gibbs sampling. We also performed experiments to test our approach from the data in Beijing actually gathered from Sina micro-blog. The results show that our model works well. In addition, we utilize preferences on location to find the communities of persons with the same interests; meanwhile the result of topics discovered provides some useful information to help us understanding functions of regions in the city.