基于多变量自动回归的评分数据异常检测算法  

Multivariate autoregression based algorithm for anomaly detection in rating data

作　　者：孙连海 卢颖[2] SUN Lian-hai1, LU Ying2(1. School of Computer Sciences, Chengdu Normal University, Chengdu 611130, China; 2. Computer Science and Engineering College, Xi＇an Technological University, Xi＇an 710021, Chin)

机构地区：[1]成都师范学院计算机科学学院,四川成都611130 [2]西安工业大学计算机科学与工程学院,陕西西安710021

出　　处：《计算机工程与设计》2018年第6期1629-1632,1652,共5页Computer Engineering and Design

基　　金：中国博士后科学基金面上基金项目(2014M560730);四川省科技厅应用基础基金项目(2015JY0071);成都师范学院高层次引进人才专项科研基金项目(YJRC2014-9);成都师范学院自然科学类培育基金项目(CS14ZD02)

摘　　要：将用户对产品评价数据视为用户基本行为和异常行为的组合,应用多变量自动回归方法进行用户基本行为的发现和异常行为的检测。将用户的评价值分布表示为累积分布,将评价值表示为基本行为和异常行为的线性组合的生成过程。针对评价值中混入的异常行为的稀疏性,基于噪音数据的高斯分布,给出相应的最小化目标函数。实验结果表明,该算法能很好地描述时序评价序列中用户的基本行为和异常行为,与相关算法相比能更好的对未来的评价值进行预测。Assuming the rating data as the combination of base behaviors and anomaly behaviors,and multivariate autoregression was used to discover the base behaviors and detect anomaly behaviors.The rating distribution was represented as cumulative distribution,and the rating score was represented as the generative process of the linear combination of base and anomaly behaviors.Aiming at the sparsity of anomaly in rating data,a minimum object function was proposed based on Gaussian distribution of noise.Experimental results show that the proposed algorithm can describe the base behavior and anomaly in time-series rating data efficiently,and it is more accurate while predicting future rating scores than related works.

关 键 词：自动回归 异常检测 稀疏性 产生过程 高斯噪声 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]