基于时序分析的Web服务QoS协同预测

Collaborative Web Service QoS Prediction Based on Time Series Analysis

Qo S预测是Web服务选取、动态服务组合和面向服务系统性能计算的重要基础.针对环境要素随时间变化导致的Web服务Qo S波动问题,提出一种基于时序分析的Qo S预测方法.该方法基于历史Qo S数据采用融合基于用户和基于项目的协同过滤方法计算不同时间片Qo S属性评价矩阵中的缺失项,进而构建Web服务的Qo S属性评价时间序列.在此基础上,设计了基于重近轻远原则的预测算法QARSPre,利用时间片步长控制序列权重,同时构造均值绝对偏差序列动态调整具有较大波动序列的权重,削弱因序列波动对预测结果准确性的影响.实验结果表明QARSPre优于传统的Qo S预测方法,能够适应不同数据集的变化.

Quality-of-Services（ Qo S） prediction is an important basis for service selection,dynamic service combination and service-oriented systems computation. To deal w ith the Qo S fluctuation of Web services caused by environment factors change w ith time,a collaborative prediction approach of Web service Qo S using attribute time series analysis w as proposed. M issing values in Qo S attribute rate matrix（ QARM） are calculated by a collaborative filtering approach that systematically combines user-based approach and item-based approach,then Qo S attribute rate serials（ QARS） are built based on QARM. A prediction algorithm QARSPre is designed in principle of larger w eight on the nearer but smaller w eight on the farther,and time slice steps are used to control slice w eights. M oreover,mean absolute error serials are constructed dynamically adjust the w eight for w eakening the influence of QARS fluctuation on prediction accuracy. Experiment results indicate that QARSPre performs better than other related w ork on accuracy and is applicable to different datasets.