一种三层加权文本聚类集成方法  

A three-level weighted approach for text clustering ensemble

作　　者：李娜 徐森[1] 徐秀芳[1] 许贺洋[1] 郭乃瑄 刘轩绮 周天 LI Na;XU Sen;XU Xiufang;XU Heyang;GUO Naixuan;LIU Xuanqi;ZHOU Tian(School of Information Engineering,Yancheng Institute of Technology,Yancheng 224051,China;Key Laboratory of Computer Network and Information Integration,Southeast University,Nanjing 211189,China;School of Underwater Acoustic Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]盐城工学院信息工程学院,江苏盐城224051 [2]东南大学计算机网络和信息集成教育部重点实验室,江苏南京211189 [3]哈尔滨工程大学水声工程学院,黑龙江哈尔滨150001

出　　处：《智能系统学报》2024年第4期807-816,共10页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(62076215);江苏省高等学校自然科学研究面上项目(21KJD520006);未来网络科研基金项目(FNSRFP-2021-YB-46);盐城工学院研究生培养创新工程项目(SJCX21_XZ018);教育部产学研合作协同育人计划项目(202102594034);中央高校基本科研业务费专项(K93-9-2022-03);江苏高校“青蓝工程”项目.

摘　　要：为了提高聚类集成效果,本文设计了一种对点、簇、划分进行加权的统一框架,提出一种三层加权文本聚类集成方法。首先根据基聚类生成超图邻接矩阵,然后依次对点、簇、划分进行加权获得加权邻接矩阵,最后用层次凝聚聚类算法获得最终结果。在多个真实文本数据集上进行实验,结果表明,与未加权及其他层面加权相比,三层加权方法可以获得更好的聚类效果,三层加权相较于未加权的平均提升幅度为12.02%;与近年来的其他8种加权方法相比,该方法在所有数据集上的平均排名位列第一,验证了本文方法的有效性。To improve the clustering ensemble effect,this paper designs a unified framework for weighted points,clusters and partitions,and proposes a three-level weighted approach for text clustering ensemble.Firstly,the hypergraph adjacency matrix is generated according to the base clustering,and then the weighted adjacency matrix is obtained by successively weighting the points,clusters and partitions.Finally,the final result is obtained by the hierarchical condensation clustering algorithm.Experiments were carried out on multiple real text datasets.The results show that compared with the unweighted results and other level weighted results,this approach has better clustering effect.The average increase of three-layer weighted compared with that unweighted is 12.02%.Compared with the other 8 weighted methods in recent years,the average ranking of this algorithm is the first in all datasets,which verifies the effectiveness of the proposed method.

关 键 词：文本聚类 聚类集成 加权聚类集成 三层加权 加权聚类 多层加权 聚类分析 无监督学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP301[自动化与计算机技术—控制科学与工程]