Neighborhood Graph Embedding for Nodes Clustering of Social Network

Graph embedding is an important dimension reduction method for high-dimensional data. In this paper, a neighborhood graph embedding algorithm is proposed and it is applied in data clustering. Different from the traditional graph embedding algorithms, a dependence degree of node is defined and it represents the dependence of two nodes; the adjacency matrix of graph is determined by dependence degree. Then a new graph embedding is proposed. After transformation matrix is solved, the weight of each attribute can also be determined from transformation matrix. Finally, the data is partitioned into clusters by clustering algorithm with weighting distance. The proposed algorithm and comparison algorithms are executed on the real social network data sets. The experimental results show that the proposed algorithm out performances the comparison algorithms and it proves that the proposed algorithm is effective for data clustering in social network.

I. Introduction

In information society, social network is an important resource and the relationships of peoples in different places are contained in the network [1]–[3]. Social network contains abundant information, we can recommend merchandises to users and find friends with common interests by analyzing network which means huge commercial value [4]–[6]. However, because of high dimension and high complexity, it makes that it is not easy to mine knowledge and patterns from social network, therefore how to efficiently analyze social network has attracted much attention from researchers and many related works have been published [7]–[9].