电子科技 ›› 2019, Vol. 32 ›› Issue (5): 55-62.doi: 10.16180/j.cnki.issn1007-7820.2019.05.011

• • 上一篇    下一篇

银行交易网络的链路预测

马青青,闫光辉,王雅斐,武昱   

  1. 兰州交通大学 电子与信息工程学院,甘肃 兰州 730070
  • 收稿日期:2018-05-04 出版日期:2019-05-15 发布日期:2019-05-06
  • 作者简介:马青青(1993-),女,硕士研究生。研究方向:社区发现。|闫光辉(1970-),男,教授,博士生导师。研究方向:数据挖掘和复杂网络。|王雅斐(1992- ),女,硕士研究生。研究方向:数据复杂网络和节点重要性挖掘。|武昱(1993-),男,硕士研究生。研究方向:复杂网络和张亮分析。
  • 基金资助:
    国家自然科学基金(61662066)

Link Prediction on Bank Transaction Network

MA Qingqing,YAN Guanghui,WANG Yafei,WU Yu   

  1. School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China
  • Received:2018-05-04 Online:2019-05-15 Published:2019-05-06
  • Supported by:
    National Natural Science Foundation of China(61662066)

摘要:

基于银行交易具有动态变化、时效性和重复性的特点,文中通过对银行网络进行清洗和压缩,研究银行网络的基本拓扑统计性质和聚类结构,并得到交易网络满足复杂网络的小世界和无标度特性。针对已有的链路预测算法在动态网络预测中的不足,提出一种自适应的动态链路算法对银行客户交易进行预测。该方法在预测网络的基础上添加了节点重要性与节点连接强弱性两个特性,并将3种预测算法结合随机算法进行了对比分析。随后将这3种算法运用到具有动态交易特性的3类真实数据集中进行实验验证。实验结果显示,新算法的预测精度约为75%。将该算法与经典的预测算法进行比较发现,提出的算法在预测方面的性能提升了5%~10%。

关键词: 复杂网络, 链路预测, 银行网络, 节点重要性, 强弱性, 准确度

Abstract:

Based on the dynamic changes in bank transactions and the characteristics of timeliness and repeatability, the basic topology statistical properties and clustering structure of the bank's network were studied, and obtained the transaction network satisfied with the small-world and scale-free characteristics.Based on the deficiency of existing link prediction algorithms in dynamic network prediction, a new dynamic link algorithm was proposed to predict bank customer transactions. Then, based on the algorithm mentioned above, two characteristics, the three predictive algorithms combined with the random algorithm were compared. These three algorithms were applied to the three types of real data sets with dynamic transaction characteristics for experimental verification. The results showed that the prediction accuracy of the algorithm was about 75%. Finally, comparing the algorithm with the classical prediction algorithm, the proposed algorithm improved the prediction by 5% to 10%.

Key words: complex network, link prediction, bank network, node centrality, strength and weakness, accuracy

中图分类号: 

  • TP391