一种基于免Hessian优化的二阶隐特征Web服务QoS预测模型  

A Hessian-free optimization-based second-order latent-factor Web-service QoS prediction model

作　　者：李艳[1] 罗辛[1] 

机构地区：[1]重庆大学计算机学院,重庆400044

出　　处：《中国科技论文》2016年第14期1649-1654,共6页China Sciencepaper

基　　金：高等学校博士学科点专项科研基金资助项目(20120191120030);国家自然科学基金资助项目(61202347);中央高校基本科研业务费专项资金资助项目(106112015CDJXY180005);重庆市新产品创新青年科技人才培养计划项目(cstc2014kjrc-qnrc40005)

摘　　要：提出1种基于免Hessian优化的二阶隐特征Web服务QoS(quality-of-service)预测模型。使用单一隐特征向量对待解隐特征进行建模,从而得到完全的Hessian矩阵;使用Hessian矩阵的Gauss-Newton逼近代入优化过程,以避免负曲面问题;使用共轭梯度下降方法对目标函数进行迭代求解,避免对Gauss-Newton逼近矩阵求逆;根据免Hessian优化方法的原则,在共轭梯度下降迭代过程中使用高效方法计算Gauss-Newton逼近矩阵和随机向量的乘积。在2个大规模Web服务用户端QoS数据集上的实验结果表明,该二阶预测模型具备很高的预测准确度和良好的执行效率。The principle of the Hessian-free optimization is adopted to build an efficient second-order LF-based Quality-of-Service （QoS） predictor. More specifically, a） the desired LFs were modeled into a unique LF vector to obtain the full Hessian expression ;b） the resulting Hessian matrix was subsequently replaced by its Gauss-Newton approximation to avoid the problem of negative curvatures; c） conjugate gradient descent （CGD） was employed to avoid inversing the Gauss-Newton approximation; and d） the product between the Gauss-Newton approximation and an arbitrary vector involved in the CGD process was computed followed the principle of Hessian-free optimization with high efficiency. Experimental results on two industrial QoS datasets indicate that the newly proposed predictor is highly accurate with fine computational efficiency.

关 键 词：服务计算 服务质量 隐特征 协同过滤 二阶优化 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术]