Li Shibao, Huang Junwei, Liu Jianhang, Huang Tingpei, Chen Haihua
College of Computer and Communication Engineering, China University of Petroleum, Qing'dao 266555, China.
Chaos. 2020 Jan;30(1):013104. doi: 10.1063/1.5094448.
Complex networks have found many applications in various fields. An important problem in theories of complex networks is to find factors that aid link prediction, which is needed for network reconstruction and to study network evolution mechanisms. Though current similarity-based algorithms study factors of common neighbors and local paths connecting a target node pair, they ignore factor information on paths between a node and its neighbors. Therefore, this paper first supposes that paths between nodes and neighbors provide basic similarity features. Accordingly, we propose a so-called relative-path-based method. This method utilizes factor information on paths between nodes and neighbors, besides paths between node pairs, in similarity calculation for link prediction. Furthermore, we solve the problem of determining the parameters in our algorithm as well as in other algorithms after a series of discoveries and validations. Experimental results on six disparate real networks demonstrate that the relative-path-based method can obtain greater prediction accuracy than other methods, as well as performance robustness.
复杂网络在各个领域都有许多应用。复杂网络理论中的一个重要问题是找到有助于链路预测的因素，这对于网络重建和研究网络演化机制是必要的。尽管当前基于相似度的算法研究了共同邻居和连接目标节点对的局部路径等因素，但它们忽略了节点与其邻居之间路径上的因素信息。因此，本文首先假设节点与邻居之间的路径提供了基本的相似性特征。相应地，我们提出了一种所谓的基于相对路径的方法。该方法在链路预测的相似度计算中，除了利用节点对之间的路径外，还利用节点与邻居之间路径上的因素信息。此外，经过一系列的探索和验证，我们解决了确定我们算法以及其他算法中参数的问题。在六个不同的真实网络上的实验结果表明，基于相对路径的方法比其他方法能够获得更高的预测准确率以及性能鲁棒性。
