Link Prediction in a Weighted Network Using Support Vector Machine

• Main Authors: Co, Jan Miles, Fernandez, Proceso L, Jr
• Format: text
• Published: Archīum Ateneo 2016
• Subjects: link prediction; vector auto regression; support vector machine; weighted networks; Computer Sciences; Databases and Information Systems
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/284; https://archium.ateneo.edu/cgi/viewcontent.cgi?article=1315&context=discs-faculty-pubs
• Institution: Ateneo De Manila University

Link prediction is a field under network analysis that deals with the existence or emergence of links. In this study, we investigate the effect of using weighted networks for two link prediction techniques, which are the Vector Auto Regression (VAR) technique and our proposed modified VAR that uses Support Vector Machine (SVM). Using a co-authorship network from DBLP as the dataset and the Area Under the Receiver Operating Curve (AUC-ROC) as the fitness metric, the results show that the performance of both VAR and SVM are surprisingly lower in the weighted network than in the unweighted network. In an attempt to improve the results in the weighted network, we incorporated features from the unweighted network into the features of the weighted network. This enhancement improved the performance of both VAR and SVM, but the results are still inferior to those in the unweighted networks. We identified that the true positive rate was generally lower in the weighted network, thus resulting to a lower AUC.