Link Prediction based On Deep Neural Network using SubGraph

Abstract

Link prediction has become quandary of predicting future link or surmises the missing link in an observed  network.Expedite link prediction method to reduce the experimental and computational efforts.Dynamic properties of  network  makes efficiency and accuracy tedious task to achieve.Real world applications demands hybrid approach for link prediction that uses node information as well as link structure and works well for any type of network.Graph and network  relation is also elaborated in this paper .

Proposed algorithm for link prediction method named as LPDNNUG(Link Prediction By Deep Neural Network In Universal Graph) and UGLP(Universal Graph Link Prediction) )works with sub graphs node information and link structure.Sub graphs are obtained with the optimised diameter size of graph  as K .It uses tailored deep feed forward Artificial Neural Network for automatic link prediction.

Here an attempt is made to prove deep neural network can work wonders in this problem domain as well.Automatic link prediction is achieved based on the various features extracted from sub graphs.Properties of ’prediction friendly’ and ‘prediction unfriendly‘ network is highlighted here.Proposed algorithms perform well across variety of graph topologies and outperform in terms of AUC(Area Under Curve) of many conventional link prediction methods as stated in implementation and result section .

Keywords: Link prediction, ANN, Deep learning, Supervised Feed Forward Neural network, Universal graph ,Data mining .