【摘要】 复杂网络的模式与演化分析具有重要的研究和应用价值,链路预测问题是其中一个研究热点.当前学者们提出了很多基于局部信息的相似性指标和链路预测算法,但是在应用于真实网络尤其是社交网络时,随着对网络宏观的演化模式与微观的链接生成机制的深入研究,链路预测算法的准确性仍有很大的提升空间.本文在现有算法基础上考虑网络同质性,分析局部结构内部的关联模式,将局部共同邻居集合根据全局最短路径信息进行建模,提出一种改进的链路预测方法:局部差异融合算法.该算法不仅保持了链接与节点之间的相似性的密切相关,而且反映了共同邻居集合内部的差异性.在各种真实网络数据集上的实验证实了本文提出算法的有效性.更多还原

【Abstract】 Link prediction is an important research topic for complex network. Understanding the growth of networks and predicting newlinks are important for many tasks. There exist a variety of similarity indices and prediction algorithms based on local information. However,in real world network applications especially in social network,the effectiveness is far from being satisfied. With the study of both macroscopic and microscopic mechanisms in network evolution and link formation,the precision of existing link prediction algorithms can be improved. We consider the homophily in local structure of networks and propose an improved link prediction algorithm called Local Inter-Difference Index( LID),which leverage both the nodes similarity in the network and the inter-difference between nodes in the same common neighbor set. Experiments were applied over several real world network datasets,and the results testified the effectiveness of our proposed algorithm.更多还原