摘要
为了从用户地理空间分布数据中挖掘用户间关联关系,提出了一种基于谱聚类的关联关系挖掘算法。首先定义了关联度,用以衡量用户之间空间分布的相似性,基于关联度构造相似矩阵,再利用谱聚类方法对用户进行聚类分析,聚类结果表征了用户的关联关系。采用Silhouette指标和聚类准确率来衡量用户关系挖掘质量,同时与传统的K-Means方法进行了比较,通过真实数据集实验,结果表明该算法在实验数据集上能达到90%以上的聚类准确率,证明方法有效、可行。
For mining association relationship from user's geographical spatial distribution data,a new method based on spectral clustering is proposed. Firstly,the correlation degree is defined,which is used to measure the similarity of spatial distribution of users,and then the similarity matrix is constructed. Clustering analysis is conducted by using spectral clustering method,and the relationship between users is characterized by clustering results. The Silhouette index and clustering accuracy are used to measure the quality of user relationship mining,meanwhile the traditional K- Means method is compared with the proposed algorithm. Experiments on real data set show that the algorithm can achieve more than 90% of the clustering accuracy,indicating that the method is effective and feasible.
出处
《电讯技术》
北大核心
2016年第1期32-37,共6页
Telecommunication Engineering
关键词
用户行为分析
用户关系挖掘
谱聚类
关联度
K-MEANS
user behavior analysis
user association mining
spectral clustering
correlation degree
K-Means