Study Link Prediction Algorithms on Collaboration Data

Abstract

The last decade has seen emergence of widespread social networks. In accordance with the emergence of such social networks has attracted increasing attention from researchers of network analytics field. One of the highlighted aspects of network analysis is Link Prediction. Link prediction is a significant issue in social networks. We study Link prediction problem on collaboration network that wherein nodes represented authors and link represented co authorship Authors are supposed to be linked if they have written paper one or further together. Local proximity indices are link prediction algorithms depend on the supposition that if two nodes/vertices are structurally similar, there should be an edge between them. Link prediction algorithms can be generally divided into three types namely: maximum likelihood based technique, similarity based technique and probability based technique. We have chosen seven various similarity based techniques measures namely: Leicht-Holme-Newman Index, Jaccard Index, Preferential Attachment, Common Neighbor, Hub Promoted Index, Hub Depressed Index and Salton Index. Experiments on collaboration network dataset that has been gathered from Stanford Network Analysis Project and determine the Area Under the receiver operating Characteristic curve that shows our method can reach better results.