Application of Linkclus in blogosphere

Abstract

This paper addresses clustering of blog users and posts in blogosphere. First, we model blogosphere as a bipartite graph where blog users and posts correspond to nodes of two types and actions on posts performed by blog users corresponds to links. Next, for clustering in blogosphere, we employ LinkClus, a link-based algorithm that finds clusters of nodes in a network effectively and efficiently. For more accurate clustering, we propose two refinements: (1) change of granularity from blog users to folders, and (2) removal of blog users and posts being highly likely to incur noises. Finally, we verify the effectiveness of the proposed approach by showing how the posts and blog users in the same cluster are similar to one another in terms of their contents.