基于LLE-PCA的社交网络数据二重降维方法研究  

Research on LLE-PCA-based double dimensionality reduction method of social-network data

作　　者：徐永辉 XU Yonghui(Institute of Information and Navigation,Air Force Engineering University,Xi’an 710077,China)

机构地区：[1]空军工程大学信息与导航学院,陕西西安710077

出　　处：《现代电子技术》2022年第10期69-74,共6页Modern Electronics Technique

基　　金：国家自然科学基金项目(61901509)。

摘　　要：社交网络数据属于多维数据,在社交网络数据挖掘研究中,对海量社交数据进行降维处理是必不可少的步骤。文中针对传统的局部线性嵌入(LLE)降维算法和主成分分析(PCA)降维算法对所处理数据类型有要求的局限性,提出一种新的局部线性嵌入-主成分分析联合二重降维算法(LLE-PCA),并将该算法应用于社交网络数据处理中。该算法将LLE中的重构系数矩阵与PCA中的主成分矩阵相结合,得到在低维空间投影后的新样本方差表达式,新的样本方差表达式可为数据降维的效果好坏提供判断依据。实验结果表明:所提算法对多维社交网络数据的降维效果较好,能明显将不同属性的社交数据分散开,同时兼顾数据的整体和局部特征;此外,相比单一的LLE和PCA算法,LLE-PCA算法缩短了对多维社交数据降维的时间。从降维效果来看,文中算法融合了LLE和PCA算法的优点,在处理同时具有线性和非线性特征的社交网络数据时有更好的兼容性。As the social network data belongs to multi-dimensional data,the dimensionality reduction processing for massive social data is an indispensable step in the research of social network data mining. In allusion to the limitation of the traditional local linear embedding(LLE) dimensionality reduction algorithm and principal component analysis(PCA)dimensionality reduction algorithm for the types of processed data,a new double dimensionality reduction algorithm combining local linear embedding-principal component analysis(LLE-PCA) is proposed,and the algorithm is applied to the data processing of social network. In the algorithm,the reconstruction coefficient matrix in LLE and the principal component matrix in PCA are combined to obtain new sample variance expression after projection in low-dimensional space. The new sample variance expression can provide a criterion for the effect of data dimensionality reduction. The experimental results show that the proposed algorithm has a better dimensionality reduction effect of multi-dimensional social network data,and can obviously separate social data with different attributes,while taking into account the overall and local characteristics of data. In addition,in comparison with single LLE algorithm and single PCA algorithm,LLE-PCA algorithm can shorten the dimensionality reduction time of multi-dimensional social data. Proceeding from the perspective of the dimensionality reduction effect,the proposed algorithm combines the advantages of LLE and PCA,and has better compatibility when processing social network data with both linear and non-linear characteristics.

关 键 词：社交网络数据 二重降维 LLE-PCA算法 数据降维 样本方差 网络数据处理 

分 类 号：TN915-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]