文献类型：期刊文献

英文题名：A Categorical Representation of Multi-Feature Fusion for Mixed Attribute Data Clustering

作者：Liang, Zupeng Li, Qiude Ji, Shengfen Hu, Sigui Yu, Yang Pan, Zhongwen Yang, Tingting Pan, Yinglong

第一作者：Liang, Zupeng

机构：[1] School of Biology and Engineering, Guizhou Medical University, Guizhou, Guiyang, 561113, China; [2] School of Mathematics and Statistics, Guizhou University, Guizhou, Guiyang, 550025, China; [3] School of Foreign Language, Guizhou Institute of Technology, Guizhou, Guiyang, 550003, China; [4] School of Economics, Hunan University of Finance and Economics, Hunan, Changsha, 410205, China

第一机构：School of Biology and Engineering, Guizhou Medical University, Guizhou, Guiyang, 561113, China

年份：2024

外文期刊名：SSRN

收录：EI(收录号：20240035688)

语种：英文

外文关键词：Cluster analysis - Clustering algorithms

摘要：The categorical attributes in mixed data are represented as high-quality numerical attributes, which have been widely concerned by the data processing community. The existing categorical representation methods do not subdivide the categorical attributes into ordinal and nominal attributes (i.e., these methods do not consider the order of the categorical attributes), resulting in the loss of the ordering relationship within the ordinal attributes, which may fail to achieve the desired representation performance. To this end, we take ordinal, nominal and numerical attributes as the object of study, fuse various intra- and inter-attribute feature relationships from these three attributes, and propose a Categorical Representation of Multi-Feature Fusion (CR-MFF) framework that is applied to clustering analysis. Specifically, 1) in the intra-attribute representation, the ordering information inside the ordinal attribute and the probability information inside the nominal attribute are extracted; 2) in the inter-attribute representation, the interactive information among ordinal, nominal and numerical attributes as well as that between attributes of the same type is considered; 3) we fuse multiple feature information from the above two representation methods. Consequently, the proposed CR-MFF not only fully extracts the complex feature information of categorical attributes but also mines the ordinal relationship between these attributes. Extensive experiments on 20 mixed, categorical, and nominal datasets with diversified characteristics demonstrate that our CR-MFF significantly improved the clustering performance of the spectral clustering method, compared with 8 state-of-the-art competitors. ? 2024, The Authors. All rights reserved.