基于分类方法的Web服务QoS预测技术研究

发布时间：2018-03-06 11:25

本文选题：服务推荐　切入点：协同过滤　出处：《杭州电子科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着互联网技术的飞速发展,面向服务的体系结构(Service Oriented Architecture,SOA)在分布式系统和软件集成领域盛行。在这一大环境下,Web服务的数量在近年内快速增长,这导致用户从海量的Web服务中寻找满足自己需求的服务愈发困难。因此,为了满足每个用户的需求,如何从大规模的Web服务群中选择出具有高质量的服务并且做出个性化推荐是一个非常具有挑战性的任务。基于服务质量(Quality of Service,QoS)的服务推荐是当下Web服务技术领域的热点问题。在为用户做出个性化推荐之前,准确地预测QoS值至关重要。协同过滤(Collaborative Filtering,CF)方法在Web服务推荐系统中得到广泛使用,这种方法利用用户调用服务的历史QoS值来分析每个用户的偏好特征,并找出相似群体,能够非常智能地做出推荐。然而,传统的协同过滤方法没有考虑用户-服务之间的潜在特征,比如网络位置、地理位置,这些信息对于Web服务推荐准确率有着显著的影响。此外,协同过滤算法在大规模数据稀疏的情况下,存在服务质量预测精度不高的问题。针对以上问题,本文提出两个新颖的服务推荐算法:(1)充分利用用户-服务之间的潜在特征,提出了一种基于贝叶斯分类的混合协同过滤服务QoS预测方法,该方法首先通过用户-服务的历史QoS提取出用户服务的特征如用户经度、纬度以及服务的提供商以及地区编号,然后基于提取出的特征使用朴素贝叶斯算法对用户-服务进行分类,最后使用基于混合的协同过滤算法在目标用户分类中找出目标用户最相似的用户对目标服务的QoS值进行预测,从而提高了预测准确度;(2)针对协同过滤算法预测准确度受限于相似用户选择准确度的问题,提出了一种基于DBSCAN共现矩阵的相似用户选择方法,提高了相似用户的选择准确度。并且针对分类器的分类准确率受限于目标特征向量的有效性,提出了用户和服务的频次向量特征,该特征向量能够显著标识用户-服务的个性特征,提高了AdaBoost分类器的分类准确度。根据分类器输出结果的概率近邻模型,并进一步提出了一种聚合模型,该聚合模型综合上述两个概率近邻模型的结果,提高了预测准确度。本文分别使用提出的两种方法在真实的数据集上进行实验,并且与一些著名方法进行了比较,实验结果证明,本文的两种方法准确度均有提升。
[Abstract]:With the rapid development of Internet technology, Service Oriented Architecture SOA (Service Oriented Architecture SOA) is popular in the field of distributed systems and software integration. This makes it more difficult for users to find services that meet their needs from a large number of Web services. How to select high quality service from large scale Web service cluster and make personalized recommendation is a very challenging task. Service recommendation based on quality of Service quality is the technical field of Web service. Before making personalized recommendations for users, It is very important to accurately predict the QoS value. Collaborative filtering Collaborative filtering method is widely used in Web service recommendation system. This method uses the historical QoS value of the user calling service to analyze the preference characteristics of each user and to find out the similar group. Be able to make recommendations very intelligently. However, traditional collaborative filtering methods do not take into account potential features between users and services, such as network location, geographic location, This information has a significant impact on the accuracy of Web service recommendation. In addition, the collaborative filtering algorithm has the problem of low quality of service prediction accuracy when large scale data is sparse. In this paper, two novel service recommendation algorithms: 1) are proposed to make full use of the potential features between users and services, and a hybrid collaborative filtering service QoS prediction method based on Bayesian classification is proposed. In this method, the features of user service, such as user longitude, latitude, service provider and region number, are extracted by the historical QoS of user-service. Then, based on the extracted features, a naive Bayesian algorithm is used to classify the user-service. Finally, a hybrid collaborative filtering algorithm is used to find out the most similar users in the classification of target users to predict the QoS value of the target service. To solve the problem that the prediction accuracy of collaborative filtering algorithm is limited by similar user selection accuracy, a similar user selection method based on DBSCAN co-occurrence matrix is proposed. The classification accuracy of the classifier is limited by the effectiveness of the target feature vector, and the frequency vector feature of the user and service is proposed. The feature vector can clearly identify the personality characteristics of the user-service and improve the classification accuracy of the AdaBoost classifier. According to the probability nearest neighbor model of the classifier output result, a aggregation model is proposed. This aggregation model synthesizes the results of the two probabilistic nearest neighbor models mentioned above and improves the prediction accuracy. In this paper, we use the two proposed methods to carry out experiments on real data sets, and compare them with some famous methods. The experimental results show that the accuracy of the two methods is improved.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09;TP391.3

【参考文献】