朱俚治

(南京航空航天大学信息中心　南京　210016)



一种加权欧氏距离聚类算法的改进*

朱俚治

(南京航空航天大学信息中心南京210016)

摘要聚类算法是一种无监督学习的算法,能够将相似度强的数据聚类到一个族内,并且能够将属性相异的数据划分到不同的族里。当今聚类算法可分为传统的聚类算法和非传统的聚类算法。传统的聚算法有:基于划分聚类算法、基于层次聚类算法等算法。人们在使用划分法和层次法进行聚类时,是通过计算聚类对象之间属性的距离来实现聚类,因此欧氏距离公式和加权欧氏距离公式在这两种聚类算法中常用的算法。加权欧氏距离计算公式在聚类时着重聚类对象属性权重的选择,所以此文从聚类对象的相似度和对象属性对应的权重,这两个方面来考虑聚类成功的概率。论文作者提出的算法思想是,如果某个聚类对象具有若干的属性,那么首先计算该聚类对象属性的相似度,再根据该属性对应的权重是否为关键权重,如果是此属性对应的权重是关键权重,那么该对象聚类的成功率较高。

关键词加权欧氏距离; 相似度; 权重; 属性; 聚类

Improvement of Weighted Euclidean Distance Clustering Algorithm

ZHU Lizhi

(Information Center, Nanjing University of Aeronautics & Astronautics, Nanjing210016)

AbstractClustering algorithm is an unsupervised learning algorithm, which can cluster the data with strong similarity to a family, and can divide the data into different groups. Clustering algorithms can be classified into the traditional clustering algorithm and the non traditional clustering algorithm. The traditional clustering algorithms are based on clustering algorithm and hierarchical clustering algorithm. When clustering is used to cluster by dividing method and hierarchy process, the distance between the attributes of clustering objects is calculated. So the Euclidean distance formula and the weighted Euclidean distance formula are used in the two clustering algorithms. Weighted Euclidean distance formula in the clustering object class attribute weights of reunion, so this paper from the object cluster similarity and the properties of an object corresponding to the weights, the two aspects to consider clustering success probability. The algorithm proposed by this paper is that if a clustering object has a number of attributes, then it first calculates the similarity of the clustering object attributes, and then according to the weight of the attributes corresponding to the weight is the key, then the success rate of the object clustering is higher.

Key Wordsweighted Euclidean distance, similarity, weight, attribute, clustering

Class NumberTP391.41

1引言

随着信息技术的发展,人们积累的音、视频数据,以及文本、图片等数据越来越多,为了从这些大量的数据中提取对人们有用的信息数据,必须将不同类的数据进行分类和聚类,因此就出现聚类技术[7~9]。分类技术和聚类技术是截然不同的两种技术,分类必须根据已有的知识进行分类,而聚类是一种无监督学习技术。……