Research On Mining Technology For Heterogeneous Network Community

Social network analysis which is an important branch of data mining has developed rapidly in recent years. The research content mainly based on homogeneous networks is analyzed. With the continuous development of network and the increasing amount of data, a single type of object has been insufficient to cope with and solve problems in reality. In order to get useful information in the mixed network, heterogeneous network (multi-types network) community mining become an important research trend. Due to the complexity and diversity of heterogeneous networks, the theory and algorithm research is still not fully mature. So integrating the novel thought to improve the efficiency of algorithm is a challenging subject, which is the main work of this article.By studying the ranking and clustering which are the two important network analysis technologies, this paper presents a framework combining ranking and clustering to solve community mining on heterogeneous network.In heterogeneous networks, the framework achieves clustering of the target type of object. And based on clustering, the framework also achieves relative rankings of all types of objects. The framework calculates relative ranking of objects based on the initial K clusters, in order to consider the attribute objects’ ranking distribution as characteristics of clustering. Then set up the mixture model, and make the K-dimension vectors for each target object. Next, in this new feature space, the class centre vector method is used to adjust the clustering, so clustering quality is improved effectively. This process is iterated until the clustering results little changing or reaching a predetermined number of iterations. The clustering and ranking effect reinforce each other in the iterative process. Finally, the clustering result is more accurate, and the ranking result is more meaningful. In order to better understand the thought of framework, this paper puts forward the algorithm CluBRank (Cluster Based on Rank) based on two-type heterogeneous network to explain and prove it. CluBRank focuses on two kinds of ranking function, simple ranking and authoritative ranking.Experiments on real data sets and simulated data sets prove clustering results of CluBRank more accurate, by comparing the CluBRank with the traditional algorithms based on link. The CluBRank effectively avoids calculating the similarity between objects. So the framework is a more efficient solution to the heterogeneous network community mining. And the clustering result including ranking provides more information.