Enhancing link prediction via network reconstruction

Abstract

To improve accuracy and robustness of link prediction algorithms, this paper proposes a serial ensemble strategy via network reconstruction. First, nine local indices are aggregated by a variant of OWA (Ordered Weighted Averaging) operator.Then, network reconstruction is performed by two different methods, a matrix completion method named ”Low Rank” and OWA operator based ensemble approach. Finally, various kinds of link prediction algorithms are applied in the reconstructed networks. Experimental results on eight real datasets validatethe potential of the serial ensemble strategy. Additionally, matrix completion with network reconstruction is an effective and efficient approach, which is insightful for real-world applications.