改进的贝叶斯矩阵修复方法  

Improved matrix completion method based on Bayes theory

作　　者：王社会[1,2] 杨俊安[1,2] 尹海波[1,2] 

机构地区：[1]电子工程学院404教研室,合肥230037 [2]安徽省电子制约技术重点实验室,合肥230037

出　　处：《计算机应用》2014年第A01期127-130,共4页journal of Computer Applications

基　　金：安徽省自然科学基金资助项目(1308085QF99)

摘　　要：海量数据在采集和传输过程中由于多种原因会不可避免地造成数据矩阵元素的缺失,对获得的残缺矩阵进行直接分析可能会出现错误的结果,因此在数据分析之前需要对残缺矩阵进行修复。常用的贝叶斯修复方法假设属性间完全相关,朴素贝叶斯方法则假设各属性相互独立,因此在处理属性间关系并不完全相关或完全独立时往往无能为力。基于贝叶斯理论,提出了一种改进的矩阵修复方法,采用关联度系数来衡量数据之间关联性,综合考虑了数据中部分属性存在关联关系而部分属性又相互独立的复杂情况。实验结果表明该方法能有效提高残缺矩阵修复的正确率,且对时效性几乎没有影响。Massive data may inevitably miss some elements in the process of acquisition and transmission because of various reasons. Correct results could not be achieved by being analyzed these obtained incomplete matrices directly, so these incomplete data sets must be completed before analyzing them. The general Bayesian method assumes that all attributes of data set are relevant, and the naive Bayesian way considers all the attributes are absolutely independent, so the ways cannot deal with the data set whose attributes are not absolutely relevant or absolutely independent. Based on Bayesian theory, this paper put forward a new method to complete missing matrix. By adopting the correlation coefficient to calculate the correlation between data, the new method reasonably considers the complicated conditions on which part of the attributes are relevant and the others are independent. The experimental results indicate that this method can effectively improve the accuracy to repair incomplete matrices, and timeliness is hardly affected.

关 键 词：数据缺失 矩阵修复 贝叶斯理论 属性关联 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]