基于随机游走增强型矩阵分解的混合服务预测  

Hybrid Random-walk Based Web Service Prediction Enhanced by Matrix Factorization

作　　者：林坚 李俊[1] LIN Jian;LI Jun(School of Mathematical and Electronic Information Engineering,Wenzhou University,Wenzhou 325035,China)

机构地区：[1]温州大学数学与电子信息工程学院

出　　处：《软件导刊》2019年第12期82-88,2,共7页Software Guide

基　　金：国家自然科学基金项目(61402337);浙江省自然科学基金项目(LQ13F020011)

摘　　要：随着Web服务数量的急剧增长,如何在大量功能相似但非功能属性各异的服务中选择满足用户个性化需求的服务是亟需解决的问题。基于QoS(Quality of Service)预测的服务推荐方法成为研究热点。然而,QoS数据的稀疏性和“冷启动”问题阻碍其发展。针对当前主流的QoS预测模型预测精度不高和收敛速度较慢等问题,提出一种基于随机游走模型和矩阵分解技术的混合QoS预测方法。该方法首先基于矩阵分解获得用户及服务的潜因子矩阵,并将用户潜因子矩阵转化为用户相似度矩阵;然后基于用户相似度矩阵并结合Web服务的网络位置信息,使用随机游走模型提高用户相似度矩阵的准确性;最终结合协同过滤方法与矩阵分解模型进行QoS预测。在真实数据集上实验,结果表明,与当前主流的QoS预测方法相比,该方法具有更高的预测精度和效率。With the rapid growth of web services that provide similar functionalities with varied Quality of Service(QoS),it is a major challenge for inexperienced service consumers to select appropriate services.So the basic idea of a service recommendation is to accu?rately predict unobserved QoS for service consumers and recommend the most suitable ones according to QoS preferences.Near?est-neighbor and model-based methods are two major approaches for QoS prediction,so that user-based and matrix factorization algo?rithms are studied.However,existing methods do not fully consider the sparsity problem of QoS data in the real world and thus produce low prediction accuracy and have slow convergence speed.This paper proposes a novel hybrid method based on random-walk model and matrix factorization model,which is mainly composed of calculating the similarity with the decomposed user latent matrix,using the random walk model to increase the accuracy of the similarity,and combining the predictions with self-adaption parameter.Experi?ments based on a real-world dataset show that our method has significantly higher prediction accuracy and better scalability than other service prediction approaches.

关 键 词：QoS预测 随机游走 矩阵分解 混合预测 服务推荐 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]