Tensor decomposition for link prediction in temporal directed networks*

Link prediction is a challenging research topic that comes along with the prevalence of network data analysis. Compared with traditional link prediction, determining future links in temporal directed networks is more complicated. In this paper, we introduce a novel link prediction method based on non-negative tensor factorization that takes into account the link direction and temporal information. In the proposed method, the temporal directed networks are modeled as a fourth-order tensor, which considers the temporal correlation coefficient of adjacent snapshots. We obtain link information by the factor matrices of tensor decomposition and score node pairs related to the link information. We give the interpretation and prove the convergence of the proposed method. Experiments are conducted on several temporal directed networks. The experimental results show that compared to several well-known link prediction methods, the proposed method improves the performance of link prediction. It is mainly because we use structural and temporal information effectively.

Automated summary

This paper introduces a novel link prediction method based on non-negative tensor factorization that improves the performance of link prediction in temporal directed networks by effectively considering link direction and temporal information.