ASUN：一种基于最大最近邻比率的任意形状层次聚类方法

打开文本图片集

中图分类号：TP301.6 文献标识码：A 文章编号：1006-8228（2025）10-31-07

ASUN： A Hierarchical Clustering Method for Arbitrary Shape Using Maximum Nearest Neighbor Ratio

Zheng Aiyu

（SchoolofComputeriencedchoogyanUnesityofiencendcholoani3ooina）

Abstract：Althoughtraditionalhierarchicalclusteringalgorthmscanoutputcompletedendrograms，theyusuallyfailtoeffectively identifyarbtrarilyshapedclusters.Meanwhile，mostexistinghierarchicalclusteringalgorithmsdesignedforarbitrarilyshaped clustersrelyonreconstructedclusterstructures，whichofendependonmorecomplexconceptsandparametersandstruggleto producecompletedendrograms.Thispaperproposesasingle-parameterhierarchicalclusteringalgorithmforarbitrarilyshaped clustersbasedontheMaximumNearestNeighborRationamedASUN，whichnotonlyensuresclusteringaccuracyforarbitrarily shapedclustersbutalsoproduceshighqualitydendrograms.First，wedfineasimilaritymeasurebasedonthek-NearestNeighbor Pair（kNNP）relationshipbetweenclusters.ThekNNPrelationshipcharacterizestheintrinsicconectvityoflocalboundaries betweenclusters，andthissimilartymeasuremaximizestheaggregationdegreeoflocalboundariesoftwoclusters.Second，we elaborateindetailtheimplementationprocessoftheASUNalgorithmanditsoptimizedversion，f-ASUN.Finally，through comparativeexperimentswithsimilaralgorithmsonbothsyntheticandreal-worlddatasets，wevalidatetheefectivenessand efficiency of the proposed algorithm.

Keywords：ArbitraryShapeClusters;Dendrogram;LocalBoundary；MaximumNearestNeighborRatio;SingleParameter

0引言

聚类分析是数据挖掘和机器学习中的关键任务之一，其应用涵盖信息检索、计算机视觉和模式识别等领域。（剩余12081字）