基于信息粒数据重构的多关系数据聚类仿真

Clustering Simulation of Multi-Relational Data Based on Information Granular Data Reconstruction

在大数据挖掘过程中,海量数据之间存在着交错复杂的非单一关系,如何准确有效的对多关系数据进行聚类处理,是大数据分析领域亟待解决的难题。传统方法在多关系数据聚类时通常转化为单关系处理,导致出现数据维度增加和数据稀疏等问题,为此提出信息粒数据重构聚类方法。方法首先利用数据的邻域构造信息粒,根据距离关系得到信息粒中数据的相似度,并对信息粒数据采取重构操作,改善数据聚类的细粒度与柔和度;然后基于重构数据与隶属程度修复数据集中的非完整数据,考虑到简化算法和统一约束,引入增益项,并把信息粒限定条件融入目标计算里,从而得到只有隶属程度限定的聚类模型;最后优化聚类过程中的重构数据与隶属程度,保证重构数据的精准,抑制聚类偏差。实验结果表明,信息粒数据重构方法提高了多关系数据的聚类精度,对于不同类型的复杂数据集具有良好的普适性。

In the process of big data mining,there is a complex and non-single relationship between massive data.How to cluster multi relation data accurately and effectively is a problem to be solved in the field of big data analysis.The traditional method usually turns into single relation processing when clustering multi relation data,which leads to the problems of increasing data dimension and data sparsity.Therefore,the information granular data reconstruction clustering method is proposed in the paper.Firstly,the neighborhood of the data was used to construct the information grain,and the similarity of the data in the information grain was obtained according to the distance relationship.The information granularity data were reconstructed to improve the granularity and softness of data clustering.Then,based on the reconstruction data and membership degree,the incomplete data in the data set were repaired.Considering the simplified algorithm and unified constraints,the gain term was introduced,and the information particle restriction condition was incorporated into the target calculation,so that the clustering model with only limited membership degree was obtained.Finally,we optimized the reconstruction data and membership degree in the clustering process,ensured the accuracy of the reconstruction data,and restrained the clustering deviation.The experimental results show that the method improves the clustering accuracy of multi relation data and has good universality for different types of complex data sets.