Link prediction in directed networks using Hits centrality and biased random walks

The aim of link prediction is to predict missing links or eliminate spurious links and new links in future network. Among various directed network link prediction algorithms, similarity-based algorithms are the most popular and competitive algorithms with low complexity and high prediction accuracy. However, most of the existing similarity-based algorithms only consider directed links and local structure information but ignore higher-order structures and fail to preserve different types of information simultaneously. To cope with this problem, this paper proposes two parameter-free link prediction frameworks, namely Hits Centrality and Bias random walk via Collaborative Filtering (HBCF) and Hits Centrality and Bias random walk via Self-included Collaborative Filtering (HBSCF), which can simultaneously preserve both node significance and higher-order structure information. In addition, to further enhance the ability of the proposed frameworks to capture different types of directed network structures, the two frameworks are fused with some representative directed local and global similarities to propose twelve highly competitive and robust link prediction indexes. Finally, the performance of the proposed twelve indexes are studied by performing several experiments on twenty-six baseline methods and sixteen real-world directed networks. The experimental results demonstrate that the proposed indexes significantly outperforms state-of-the-art baseline methods in most cases.