摘要
从两路数据聚类分析到三路数据聚类分析实质上是由平面分析到立体分析的过程。三路数据聚类方法研究的核心之一是如何把传统的两路截面数据聚类技术向三路数据聚类扩展的问题。本文基于Tucker模型的思路,提出一种先对三路数据执行矩阵分解,而后进行聚类分析的三路数据聚类方法。这种方法不但能够通过核心矩阵反映三路数据三个模式信息联系的强度大小,而且还可以在一个分解框架下对三路数据的三个模式同时进行聚类分析。实证分析结果表明,本文提出的聚类方法不但灵活、易于理解,同时也有着良好的判别性和实用性。
Clustering analysis from two-way data to three-way data is a process of analysis from plane to three-dimensional. One of core problems about three-way data clustering method is how to extend the traditional clustering technology of the two-way data to the three-way data. Based on Tucker3 model, this paper puts forward a three-way data clustering method which performs matrix decomposition and then takes the component decomposition of three ways as the synthetic weights. Through the core matrix data, this method can not only reflect the strength of the correlation information in addition to three ways, but also can perform clustering analysis at the same time in one framework of three-mode com- ponent decomposition. Empirical analysis results show that the proposed clustering method is not only flexible and easy to be understood, but also have a good discrimination and practicality.
出处
《数理统计与管理》
CSSCI
北大核心
2016年第1期71-80,共10页
Journal of Applied Statistics and Management
基金
国家社科基金项目(15BTJ033)
国家自然科学基金项目(71271206)
江苏高校哲学社会科学研究一般项目(2014SJB406)
江苏师范大学博士学位教师科研支持项目(12XWR016)资助
关键词
三路数据
相似性测度
综合集成
Tucker分解
聚类方法
three-way data, similarity measure, comprehensive synthesis, tucker3 decomposition, clus- tering analysis
