Link prediction, aiming to find missing links in an observed network or predict those links that may occur in the future, has become a basic challenge of network science. Most existing link prediction methods are based on local or global topological attributes of the network such as degree, clustering coefficient, path index, etc. In the process of resource allocation, as the number of connections between the common neighbors of the paired nodes increases, it is easy to leak information through them. To overcome this problem, we proposed a new similarity index named ESHOPI (link prediction based on Dempster–Shafer theory and the importance of higher-order path index), which can prevent information leakage by penalizing ordinary neighbors and considering the information of the entire network and each node at the same time. In addition, high-order paths are used to improve the performance of link prediction by penalizing the longer reachable paths between the seed nodes. The effectiveness of ESHOPI is shown by the experiments on both synthetic and real-world networks.

中文翻译：

---

基于高阶路径索引重要性的证据链路预测方法

链接预测，旨在发现观察到的网络中缺失的链接或预测未来可能发生的链接，已成为网络科学的基本挑战。现有的链路预测方法大多基于网络的局部或全局拓扑属性，如度数、聚类系数、路径索引等。在资源分配过程中，随着配对节点的公共邻居之间连接数的增加，通过它们很容易泄露信息。为了克服这个问题，我们提出了一种新的相似性指标 ESHOPI（基于 Dempster-Shafer 理论的链路预测和高阶路径索引的重要性），它可以通过惩罚普通邻居和考虑整个网络的信息来防止信息泄漏和每个节点同时进行。此外，高阶路径用于通过惩罚种子节点之间较长的可达路径来提高链路预测的性能。ESHOPI 的有效性通过在合成网络和真实网络上的实验得到证明。