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.
展开▼