基于无监督学习的异质网络多尺度离群点挖掘研究  

Research on heterogeneous networks multi⁃scale outlier mining based on unsupervised learning

作　　者：朱辉[1] 张莉芸 ZHU Hui;ZHANG Liyun(Guilin University of Technology,Nanning 530001,China;School of Information and Management,Guangxi Medical University,Nanning 530021,China)

机构地区：[1]桂林理工大学,广西南宁530001 [2]广西医科大学信息与管理学院,广西南宁530021

出　　处：《现代电子技术》2024年第12期182-186,共5页Modern Electronics Technique

基　　金：崇左市科技计划:基于大数据技术的崇左市高校毕业生人岗匹配智能平台研究(2023ZC035512)。

摘　　要：现有的异质网络多尺度离群点挖掘算法忽略了数据点之间的顺序关系,无法充分利用数据点在异质网络中的排列顺序信息,从而导致聚类精度下降。对此,提出一种基于无监督学习的异质网络多尺度离群点挖掘方法,对异质网络的多节点、多边特点进行分析。利用季节-趋势时序分解法提取异质网络数据特征。根据数据特征,结合K-means聚类算法与排序算法,将数据点的排序信息添加至聚类过程中,以实现对异质网络数据离群点的挖掘。实验结果表明,利用该方法进行网络数据节点聚类的准确率均能达到80%以上;并且实现了多尺度离群点挖掘后,能够精准地识别出离群点,为后续的网络通信维护提供了良好的保障。The existing multi-scale outlier mining algorithms for heterogeneous networks ignore the sequential relationship between data points and cannot fully utilize the information of the arrangement order of data points in heterogeneous networks,resulting in a decrease in clustering accuracy.Therefore,a heterogeneous networks multi-scale outlier mining method based on unsupervised learning is proposed,and the multinode and multilateral characteristics of heterogeneous networks are analyzed.The seasonal trend time series decomposition method is used to extract data features of heterogeneous networks.In combiantion with K-means clustering algorithm and sorting algorithm,the sorting information of data points is added to the clustering process to achieve outlier mining in heterogeneous network data according to data characteristics.The experimental results show that the accuracy of the proposed method for clustering network data nodes can reach over 80%.After achieving multi-scale outlier mining,it is possible to accurately identify outliers,providing a good guarantee for subsequent network communication maintenance.

关 键 词：异质网络 多尺度 离群点挖掘 无监督学习 K均值聚类 网络数据 离群因子 

分 类 号：TN911-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]