Link Mining in Online Social Networks with Directed Negative Relationships

Abstract

One of the most important work to analyse online social networks is link mining. A new type of social networks with positive and negative relationships are burgeoning. We present a link mining method based on random walk theory to mine the unknown relationships in directed social networks which have negative relationships. Firstly, we define an extended Laplacian matrix based on this type of social networks. Then, we prove the matrix can be used to compute the similarities of the node pairs. Finally, we propose a link mining method based on collaboration recommendation method. We apply our method in two real social networks. Experimental results show that our method do better in terms of sign accuracy and AUC for mining unknown links in the two real datasets.