Microblog Topic Mining Based On Relation Network

Recently, microblog has a rapid development. Users in microblog not only can make friends, chat with others and publish a tweet, but also can initiate a topic, comment other topics. It is very meaningful to analyze and mine microblog topic. Existing related works about mining microblog topics mostly are based on text analysis. The text in microblog is always short text, and it has the characteristics of sparse data and lots of noise. It brings difficulties for analysis of microblog topic. Microblog is a social media which is based on relationship. Users who have close relation always focus on similar topics. The results are not good enough when we mine microblog base on text analysis purely. So we start from the relation between users, and we propose a model of relation network to mine microblog hot topics. In this paper, we use this model of relation network to develop a microblog topic mining system based on relation network, and we achieve certain results.The article has proposed a method of presenting microblog topic based on relation network. Through analysis of microblog topic, we found that users have states in the process of diffusion of microblog topic between them. And the states of users will change in this process. Users’ operations in microblog cause change of users’ states. In microblog, users’ behaviors include posting a tweet, commenting a tweet, forwarding a tweet and @ other users. The users’ operations also generate relation between users. We propose method of presenting microblog topic based on relation network because of this. In the microblog topic relation network, the nodes represent users and the edges represent the relation between users in microblog topic. After constructing the topic relation network, we pre-analyze the constructing topic relation networks, including the betweenness centrality distribution, closeness centrality distribution, network clustering coefficient and so on.The article proposed a method based on microblog topic relation network. We mined the relation in real hot topics from the Sina Weibo. All four real hot topics in Sina Weibo are "Beijing college entrance examination reform", "inter-provincial medical reimbursement", "pensions" and "extending the payment period". In the experiment of mining microblog topic, we also use the current method based on text analysis. And we compared the two results. The result shows that the F value in the method of mining microblog topic based on microblog topic relation network is 10% larger than the F value of the current method based on text analysis.The article proposed a display policy of visualization which is based on Gaussian Random Numbers. We used the policy to show clustering results through the method of community detection. The proposed method of mining microblog topic based on relation network and the proposed police of visualization based on Gaussian Random Numbers have been successfully applied to the microblog topic mining system based on relation network.