Abstract:

The task of the link prediction is to exploit the missing links in the network or to predict the possibility of a link between a pair of nodes. How to predict missing link accurately in incomplete networks is a challenging problem. Considering both the collective influence of nodes and the contribution of edge clustering information, this paper puts forward a new link prediction algorithm named CELP (link prediction algorithm based on collective influence and edge clustering information). Then, combining the community attributes of the node and based on the designed Bayesian network, this paper proposes the extended algorithm CELP* for label network. The experimental results of multiple test networks in various fields demonstrate that compared with the typical link prediction methods and recent partial algorithms, the method improves the prediction accuracy while maintaining the same AUC level, and further confirms the importance of local node information and edge information for link prediction.

Key words: complex network, link prediction, collective influence, clustering information, Bayesian network, similarity algorithm

摘要：

链路预测的任务是挖掘网络中缺失的链接或预测一对节点之间存在链路的可能性。如何有效准确地预测不完整复杂网络中的缺失链接是一个具有挑战性的问题。综合考虑节点的集体影响以及边的聚类信息对所预测边的贡献，提出一种新的链路预测算法CELP（link prediction algorithm based on collective influence and edge clustering information），并结合节点的社区属性，基于设计的贝叶斯网络提出其在标签网络的扩展算法CELP*。来自各个领域的多个测试网络的实验结果表明，与典型的链路预测方法及近期的部分指标相比，所提算法在保持同等AUC水平的同时，提高了预测精度，也进一步肯定了局部节点信息和链路信息对于链路预测工作的重要性。

关键词: 复杂网络, 链路预测, 集体影响, 聚类信息, 贝叶斯网络, 相似性算法