时态网络节点相似性度量及链路预测算法  

Node Similarity Measurement and Link Prediction Algorithm in Temporal Networks

作　　者：陈东明[1] 袁泽枝 黄新宇 王冬琦[1] CHEN Dong-ming;YUAN Ze-zhi;HUANG Xin-yu;WANG Dong-qi(School of Software,Northeastern University,Shenyang 110169,China)

机构地区：[1]东北大学软件学院

出　　处：《东北大学学报（自然科学版）》2020年第1期29-34,43,共7页Journal of Northeastern University(Natural Science)

基　　金：辽宁省自然科学基金资助项目(20170540320);辽宁省博士启动基金资助项目(20170520358);中央高校基本科研业务费专项资金资助项目(N172415005-2)

摘　　要：详细分析和阐述了时态网络中的链路预测问题,将时态网络按时间顺序划分为具有相同时间间隔的多层网络快照序列.针对基于共同邻居的相似性指标对网络链路刻画粒度较粗糙的问题,提出了基于邻居节点聚类系数的相似性度量指标NCC和NCCP,并基于此提出时态网络链路预测算法.通过在真实数据集上的对比实验验证了利用邻居节点的聚类信息可以提高预测精度.利用真实邮件数据集验证了所提出的链路预测算法预测效果的优越性,并且实验结果证明越接近预测时间的网络结构对预测结果影响越大.Link prediction in temporal networks was analyzed and discussed in detail.The temporal network was divided into multilayer network snapshot sequences with the same time interval in chronological order.Aiming at solving the problem of rough granularity obtained by the common-neighbor-based similarity index,similarity indexes NCC and NCCP based on neighbor node clustering coefficient were proposed.Then a link prediction algorithm for temporal networks was designed for networks based on these two indicators.The comparison experiments on real datasets showed that the cluster information of neighbor nodes can improve the prediction accuracy.The superiority of the proposed link prediction algorithm was verified by a real mail dataset,and the experimental results showed that the closer the network structure is to the prediction time,the greater the impact on the prediction results.

关 键 词：时态网络 链路预测 多层网络 聚类 相似性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]