| 基于节点匹配度的动态网络链路预测方法 |
| |
| 作者姓名： | 李聪 季新生 刘树新 李劲松 李海涛 |
| |
| 作者单位： | 信息工程大学，河南 郑州 450001 |
| |
| 基金项目： | 国家自然科学基金(61803384) |
| |
| 摘 要： | 现实世界存在众多真实网络，研究真实网络中的动态演化趋势和时序性特征是热点问题。链路预测技术作为网络科学领域重要研究工具可通过挖掘历史连边信息推测网络演化规律，进而对未来连边进行预测。通过分析动态真实网络中的拓扑结构演化，发现通过分析网络拓扑中节点间的交互性和匹配度问题能够更充分捕捉网络的动态特征，提出一种基于节点匹配度的动态网络链路预测方法。该方法对网络节点的属性特征进行分析，定义基于原生影响力和次生影响力的节点重要性量化方法；引入时间衰减因子，刻画不同时刻网络拓扑对连边形成的影响程度；结合节点重要性和时间衰减因子定义动态节点匹配度（TMDN，temporal matching degree of nodes）方法，用于衡量节点对之间未来形成连边的可能性。在5个真实动态网络数据集中的实验结果表明，相比现有 3 类主流动态网络链路预测方法，所提方法在 AUC 和Ranking Score 两种评价标准下均取得更优的预测性能，预测结果最高提升 42%，证明了节点间存在着交互匹配优先级，同时证实了节点原生影响力和次生影响力的有效性。
|
| 关 键 词： | 动态网络 链路预测 节点匹配度 节点重要性 时间衰减因子 |
Link prediction method for dynamic networks based on matching degree of nodes |
| |
| Authors: | Cong LI Xinsheng JI Ushuxin LI Jinsong LI Haitao LI |
| |
| Affiliation: | Information Engineering University, Zhengzhou 450001, China |
| |
| Abstract: | The research of dynamic evolutionary trends and temporal characteristics in real networks which are important part of the real world is a hot research issue nowadays.As a typical research tool in the field of network science, link prediction technique can be used to predict the network evolution by mining the historical edge information and then predict the future edge.The topological evolution of dynamic real networks was analyzed and it found that the interaction and matching between nodes in the network topology can capture the dynamic characteristics of the network more comprehensively.The proposed method analyzed the attribute characteristics of network nodes, and defined a node importance quantification method based on primary and secondary influences.Besides, a time decay factor was introduced to portray the influence of network topology on the formation of connected edges at different moments.Furthermore, the node importance and time decay factor were combined to define the Temporal Matching Degree of Nodes (TMDN), which was used to measure the possibility of future edge formation between node pairs.The experimental results in five real dynamic network datasets showed that the proposed method achieves better prediction performance under both AUC and Ranking Score, with a maximum improvement of 42%.It also proved the existence of interactive matching priority among nodes, and confirmed the effectiveness of both primary and secondary influence of nodes.As the future work, we will add diversified feature information to further deepen the analysis of dynamic real networks and then predict the evolution law more accurately. |
| |
| Keywords: | dynamic networks link prediction matching degree of nodes node importance time decaying parameter |
|
| 点击此处可从《》浏览原始摘要信息 |
|
点击此处可从《》下载全文 |
