基于用户和服务区域信息的个性化Web服务质量预测

Personalized QoS Prediction for Web Services Based on the Region Information of Users and Services

在基于Web服务质量的研究中,服务质量信息均被假设为已知值。然而,在实际中很多服务质量信息是未知的且具有个性化特征。已有研究在进行服务质量预测时只采用全局服务质量信息,未考虑用户和服务的局部区域信息对服务质量预测的影响。因此,综合全局和局部角度对服务质量进行个性化预测具有重要的理论价值和实践意义。基于服务质量的个性化特征以及用户和服务两方面影响因素的视角,提出一种基于用户和服务区域信息的服务质量预测方法。将全局的服务质量信息和局部的区域信息相结合构建预测模型,采用随机梯度下降法对模型进行优化学习,最终能够得到满意的预测结果。实验采用通用的综合数据集WSDream-QoSDataset2,运用Matlab R2010b中机器学习的相关软件包训练预测模型,分析模型中不同参数对服务质量预测结果的影响,并与NIMF方法和Colbar方法进行对比分析。研究结果表明,考虑用户和服务区域因素有助于提高服务质量预测的准确性;服务所在的区域影响比用户区域影响稍大,主要是由于服务器端服务运行条件不同所致;已有服务质量信息的密度对未知服务质量的预测具有一定影响,密度越大,预测效果越好;与NIMF方法和Colbar方法对比,MAE和RMSE评价指标显示所提出的方法具有较高的服务质量预测准确性。同时,采用区域内平均值的方法计算新用户和新服务的服务质量信息,有效地解决了冷启动问题。研究结果有助于提供较为准确的服务质量信息,对基于服务质量的服务选择、服务推荐和服务组合等研究工作具有重要的支持作用,也为面向服务计算的发展提供一些有益的现实启示。

Concerning the Qo S-based studies,the Qo S information is all assumed to be known.However,in practical application,much Qo S information is unknown and have personalized characters.Previous studies have just employed the global Qo S information to make predictions and do not consider the impact of user/service local region information on the Qo S values.Therefore,it is of important theoretical value and practical significance to make personalized Qo S prediction from both global and local perspectives.From the perspective of personalized Qo S and the influence from users and services,this study proposes a novel Qo S prediction approach based on the region information of users and services( QPRIUS).First,a prediction model is established by taking advantage of both the global Qo S information and the local region information of users and services.Then the model is optimized iteratively by means of stochastic gradient descent.Finally,a satisfactory prediction result can be obtained.Experiments utilize the synthetic data WSDream-QoSDataset2 and employ the related software package of machine learning in Matlab R2010 b to train the prediction model.Different parameters in the model are analyzed about their effects on the Qo S prediction accuracy.Furthermore,this study is compared with the methods of NIMF and Colbar.The experimental results indicate that it is beneficial to improve the accuracy of Qo S prediction by considering the region of users and services.Parameter experiments show that the effect of service region is larger than that of user region because there is some difference in the running condition for services in the server-side.The density of existing Qo S information has certain influence on prediction of unknown Qo S.The larger the density is,the better the prediction accuracy will be.In the comparative experiments with the methods of NIMF and Colbar,the MAE and RMSE criteria indicate that the proposed method achieves high Qo S prediction accuracy.This method uses the intra-regional mean one to compute the Qo S information of new users and services and can effectively solve the cold start problem.The findings contribute to provide more accurate Qo S information and have important supporting action for some Qo S-based studies such as service selection,service recommendation,service composition and so on.Moreover,the findings provide some reality enlightenment for the development of service-oriented computing.