Evaluation Method Study of Blog's Subject Influence and User's Subject Influence

Abstract

The large group of users and massive blogs information in microblog bring serious problem of information overload. It constitutes a great challenge in finding the most valuable and influential blogs or users for a given subject. To address this, we first constructed an improved text clustering algorithm called DSCURE based on CURE (clustering using representatives) to generate clusters on different subjects. The improvements include representative points' selection which considers both density and scatterness of the points and text distance calculation which combines Vector Space Model and Latent Dirichlet Allocation model. The experimental results show that this algorithm outperforms other algorithms and reaches the stable state earlier. Second, we proposed a blog's subject influence evaluation model which mainly considers the subject relevance, the content quality and the timeliness of a blog. Experimental results demonstrate that this model is effective and reasonable in identifying influential blogs. Based on this model, we further put forward a user's subject influence evaluation model - QualityRank which considers the characteristics of the user's personal attributes, the user's blog features and the network structure. Experimental results show that QualityRank outperforms other models we referred to.