DeepEye: Link prediction in dynamic networks based on non-negative matrix factorization

Nahla Mohamed Ahmed,Yun Li,Ling Chen,Bin Li,Wei Liu,Yulong Wang

doi:10.26599/bdma.2017.9020002

Nahla Mohamed Ahmed, Yun Li + Show 4 more

Open Access

https://doi.org/10.26599/bdma.2017.9020002

Copy DOI

Abstract

A Non-negative Matrix Factorization (NMF)-based method is proposed to solve the link prediction problem in dynamic graphs. The method learns latent features from the temporal and topological structure of a dynamic network and can obtain higher prediction results. We present novel iterative rules to construct matrix factors that carry important network features and prove the convergence and correctness of these algorithms. Finally, we demonstrate how latent NMF features can express network dynamics efficiently rather than by static representation, thereby yielding better performance. The amalgamation of time and structural information makes the method achieve prediction results that are more accurate. Empirical results on real-world networks show that the proposed algorithm can achieve higher accuracy prediction results in dynamic networks in comparison to other algorithms.

Highlights

With the widespread reach of the Internet, social networks have become popular and people can use them to build wider connectionsŒ1 4
We propose a method for link prediction based on non-negative Matrix Factorization (NMF) in dynamic networks
Empirical results on real-world networks show that the proposed algorithm can achieve higher quality prediction results in dynamic networks, in comparison to other algorithms

Summary

Introduction

With the widespread reach of the Internet, social networks have become popular and people can use them to build wider connectionsŒ1 4. When we build a network structure to topologically approximate complex systems, missing or redundant links may unavoidably occur owing to time and cost restrictions in experiments conducted for constructing the networks. In biological networks like disease-gene networks, protein-protein interaction networks, and metabolic networksŒ9, links indicate the interaction relations between the organism and the diseases represented by the nodes they connect to. Such implicit interaction relations can be discovered by a biological experiment, their excessive cost makes large-scale experiments impossible. To reduce the huge cost of biological experiments, link prediction can be employed at the preprocessing stage to discover the potential interaction relations in biological networks. In diseases-gene networks, link prediction can detect the hidden links between disease and gene to discover the cause of the disease and to discover a treatment and new drugs for the diseaseŒ10

Objectives

Methods

Results

Conclusion