Research On Blog Friends Recommendation Mechatism In Blogsphere

With the development of Web2.0 social network community, Blogsphere has become a tremendous research information source with increasing speed of expansion. For this ever-changing source of information, finding bloggers with common interests for a user means searching the information source the user want. At the same time, friend recommendation is also a urgent problem for Blog websites to solve.This thesis reviews current blogger clustering methods and friend recommendation methods. On the basis of current Blog research, this thesis analyzes features of Blog page and proposes model and process of Blog friend recommendation. Before proposing a recommendation method in this thesis, a clear definition of friend and the reason of application of blogger clustering in friend recommendation are provided. Subsequently, this thesis presents text based Blog similarity computing method, and then provides a method for blogger set similarity and mergence. On the basis of computing method of Blog and Blog set, this thesis improves classical K-means algorithm and K-centroid algorithm and design two text based blogger clustering methods. In the Blog clustering algorithm based on the mergence of links and Blog texts, this thesis expounds the construction method of Blogsphere adjacency graph based on combination of link information and text information, and design the calculation method of link intensity between Blog sets in order to apply classical rock algorithm to cluster bloggers with adjacency graph. This thesis focuses on improving the efficiency of friend recommendation. Based on clustering results of the blogger set, friend recommendation can be made according to Blog article similarity computing method of this thesis in small blogger sets clustered by clustering methods. The blogger cluster method based on the combination of link information and text information considers not only direct links which reflect the related interest to some extent, but also the association of the potential interests among bloggers.In the experiment, this thesis represents two criterias for blogger clustering. The first criteria based on statistics of blogger similarities is fit for computer automatic evaluation. Another criteria with involvement of human beings evaluates Blog friend recommendation through calculating searching precision of friend recommendation. Through implementation and comparition of results of these three blogger clustering methods on different datasets based on the first criteria, the best method and parameters under different data sets are determined. Finally, a comparative experiment is executed between the best method and parameters of this thesis and other researcher's method.