深度学习时代聚类研究核心问题传统方法与深度聚类综述

打开文本图片集

中图分类号：TP181 文献标识码：A 文章编号：2096-4706（2025）23-0040-09

Abstract：As thedigitization process ofhuman activitiesaccelerates，the scaleandcomplexityofdata grow exponentialy， presenting multi-dimensional chalenges to traditional clustering methods.Basedonaclasification framework encompasing “data characteristics-driven，algorithm performance and robustness，evaluationand interpretability，and scenario-driven dimensions”，thispapersystematicallysortsout thecoreissuesandtheirsolutions inthfeldofclusteringfromboth traditional methodsanddeepclustering perspectives，and analyzes theadvantages andlimitationsofvarious methods.However，thedep clustering isdpendentonlarge-scaledata，andhas highcomputationalcomplexityandpoorinterpretabilityItprospectsfuture research directions includingfew-shot，lightweight，and interpretabledeepclustering.This work providesasystematiceference for research in the field and contributes to the development of deep clustering theories and applications.

Keywords： clustering; Deep Learming; deep clustering; unsupervised learning; algorithm optimization

0 引言

聚类作为无监督学习核心技术，基于“组内相似、组间相异”原则实现样本自动分簇、挖掘数据内在结构。（剩余25327字）