The research on topic detection based on multi-models and multi-characteristics

In this paper, a new approach was proposed for the topic detection, which combined the multi-models and multi-characteristics, entity information similarities were researched as features for support vector machine model (SVM) by us, for example, the content similarity, time similarity and location similarity methods can be proposed respectively, the Bayesian model also can be discussed to obtain the atomic characteristics in this paper. Except this features, the expert knowledge base has been studied to solve the difficult classification problem. The experimental results show that the approach combined the statistical model with expert rule base is effective.