计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (17): 94-100.DOI: 10.3778/j.issn.1002-8331.1608-0323

• 大数据与云计算 • 上一篇    下一篇

基于张量模型的多因素影响程度分析方法

王卿云1,2,李红燕1,2,洪申达1,2   

  1. 1.北京大学 机器感知与智能教育部重点实验室,北京 100871
    2.北京大学 信息科学技术学院,北京 100871
  • 出版日期:2017-09-01 发布日期:2017-09-12

Multi factors influencing degree analysis method based on tensor model

WANG Qingyun1,2, LI Hongyan1,2, HONG Shenda1,2   

  1. 1.Key Laboratory of Machine Perception, Ministry of Education, Peking University, Beijing 100871, China
    2.School of Electronics Engineering and Computer Science, Peking University, Beijing 100871, China
  • Online:2017-09-01 Published:2017-09-12

摘要: 影响程度分析分为独立影响程度分析和联合影响程度分析。传统的影响程度分析方法难以兼顾二者,并且在分析的过程中受困于影响因素数目过多以及因素之间复共线性的干扰。部分方法甚至难以应对大规模数据集,这些问题无疑阻碍了方法分析准确率的提升以及广泛应用。于是基于此提出了基于张量分解与重建的多因素影响程度分析方法(MAT),消除了影响因素之间的复共线性,全面而准确地分析了单一因素的独立影响程度和多因素的联合影响程度。通过在真实大规模移动通信数据集上的实验,验证了MAT方法的有效性和准确性。

关键词: 影响因素, 影响程度分析, 预处理, 张量分解与重建

Abstract: The influencing degree analysis can be divided into independent influencing degree analysis and the combined influencing degree analysis. The traditional influencing degree analysis method is difficult to take the two into account, and is trapped in the number of influencing factors and multicollinearity between factors. Some methods even cannot be applied to large scale data sets, which undoubtedly hinder the improvement of the accuracy of the analysis and the wide application of the methods. So it proposes the Multi factors influencing degree Analysis method based on Tensor decomposition and reconstruction(MAT), which eliminates the multicollinearity between factors, and realizes the analysis of independent and combined influencing degree at the same time. The experiments on the real large scale mobile communication data set have showed the validity and accuracy of the MAT.

Key words: influencing factors, influencing degree analysis, preprocessing, tensor decomposition and reconstruction