智能系统学报2025，Vol.20Issue(3)：584-593,10.DOI:10.11992/tis.202407001
基于混合邻域图的复杂结构数据集层次聚类算法
Hybrid neighborhood graph-based hierarchical clustering algorithm for datasets with complex structures
摘要
Abstract
Complex structured datasets typically refer to datasets containing clusters of different shapes(including spherical,non-spherical,and manifold shapes),sizes,and densities.The natural neighbor algorithm exhibits limitations in handling datasets with unclear boundaries and varying densities.Particularly,its performance decreases significantly when the dataset contains a significant amount of noise.To address this drawback,we propose a hybrid neighborhood graph-based hierarchical clustering algorithm for datasets with complex structures(HCHNG).We proposed a method of shared natural neighborhood graph,which uses the neighbor relationships to sparse the dataset and reduce the impact of abnormal samples on clustering results.Subsequently,the algorithm divides the dataset into several subgraphs and en-hances the processability of variable density data by merging operations.Concurrently,we propose a new method for defining subgraph similarity,which ensures higher similarity between subgraphs of the same class.Additionally,we im-prove the performance of the natural neighborhood graph in identifying datasets with blurred boundaries.The experi-mental results reveal that the HCHNG algorithms can recognize variable density spherical datasets and complex data-sets containing a large amount of noise.Therefore,our algorithm is more effective than the existing methods in pro-cessing datasets with complex structures.关键词
聚类分析/混合邻域图/共享自然邻居/改进的自然邻域图/共享自然邻域图/子图相似性/复杂数据集/数据挖掘Key words
cluster analysis/hybrid neighborhood graph/shared natural neighbors/improved natural neighborhood graph/shared natural neighborhood graph/subgraph similarity/complex dataset/data mining分类
信息技术与安全科学引用本文复制引用
陈仲尚,冯骥,杨德刚,蔡发鹏..基于混合邻域图的复杂结构数据集层次聚类算法[J].智能系统学报,2025,20(3):584-593,10.基金项目
重庆市教委科学技术研究项目(KJZD-M202300502,KJQN201800539). （KJZD-M202300502,KJQN201800539）
