Trending topic prediction on social network

The fast information sharing on social network services generates more than thousands of topics every day. It is extremely important for business organizations and administrative decision makers to learn the popularity of these topics as quickly as possible. In this paper, we propose a prediction mode based on SVM with features of three subsets: quantity specific features, quality and user specific features which supplement each other. Furthermore, we divide topic data into time slices which is used as a unit of feature construction. Our findings suggest that the capability of our prediction model outperforms previous methods and also reveals that subsets of features play different role in the prediction of trending topics.