A Network Embedding Approach for Link Prediction in Dynamic Networks

Abstract

Dynamic networks and their evolving nature have gained the attention of researchers with its ubiquitous applications in a variety of real-world scenarios. Learning the evolutionary behavior of such networks is directly related to link prediction problem as the addition and removal of links or edges over time leads to the network evolution. With the rise of large-scale dynamic networks like social networks, link prediction in such networks or otherwise temporal link prediction has become an interesting field of study. Existing techniques for enhancing the performance of temporal link prediction leverages the notion of matrix factorization, likelihood estimation, deep learning and time series based techniques. However, building a framework for temporal link prediction that preserves the non-linear varying temporal properties of dynamic networks is still an open challenge. Here, we propose a unified framework that incorporates Network Representation Learning (NRL) and time series analysis for temporal link prediction. Our experimental results on various real-word datasets show that the proposed framework outperforms the state-of-the-art works.KeywordsDynamic networksTemporal networksLink predictionNetwork Representation Learning (NRL)Time series