Improve Link Prediction Accuracy with Node Attribute Similarities

Abstract

Link prediction is one of the significant research problems in social networks analysis. Most previous works in this area neglect attribute similarity of the node pair which can easily obtain from real world dataset. Traditional supervised learning methods study the link prediction problem as a binary classification problem, where features are extracted from topology of the network. In this paper, we propose a similarity index called Attribute Proximity. The set of features are similarity index we proposed and four others well-known neighbourhood based features. We then apply a supervised learning based temporal link prediction framework on DBLP dataset and examine whether attribute similarity feature can improve the performance of the link prediction. In our experiments, the AUC performance is better when attribute similarity feature is considered.