【作者】 王翔；

【导师】 丁勇； 吴向阳；

【作者基本信息】 合肥工业大学 ， 物流工程（专业学位）， 2019， 硕士

【摘要】 随着物联网接入用户以及终端设备的飞速增长,用户需求的差异化越来越明显,用户对网络的服务质量(Quality of Service,QoS)也提出了更高的要求。如何通过QoS预测来实现物联网资源的合理分配以及对用户进行个性化服务推荐,逐渐成为学者们研究的热点。推荐系统可以通过行为数据挖掘用户之间的关联性,并且具有解释性强的优点,已经被广泛应用在QoS预测中。然而传统的QoS预测算法未考虑物联网环境下QoS的实时变化性,通过单一的相似度计算方法的改进来进行预测,导致预测精度不高。为了提高QoS预测效果,论文运用协同过滤和人工智能方法,研究物联网环境下的QoS预测方法。具体研究工作如下:(1)基于Service2vec的QoS预测方法研究。首先通过阈值将服务项划分为高响应服务项以及非高响应服务项,针对非高响应服务项集合,设计了一种Service2vec方法,挖掘用户动态调用服务项的潜在特征,构建服务项向量;对调用高响应服务项的用户计算其用户相似性。最后对非高响应服务项采用基于服务项协同过滤的QoS预测方法,对高响应服务项采用基于用户协同过滤的QoS预测方法。(2)基于AutoEncoder的QoS预测方法研究。本文设计了一种基于AutoEncoder的QoS预测方法框架。将AutoEncoder方法应用到用户-服务项的响应时间矩阵,得到用户的特征向量,对其进行谱聚类,同时融入地理因素的影响,对用户的地理位置进行K-Means聚类,最后结合两者的权重进行QoS预测。(3)针对本文提出的两种QoS预测方法,分别选取了对照实验,并在真实的WS-DREAM数据集进行了实验,实验表明,与对比的QoS预测方法相比,本文提出的两种QoS预测方法取得了更好的预测结果。本文针对物联网环境下服务与传统环境服务的不同,提出了两种QoS预测方法,相对于以往QoS预测研究,本文提出的QoS预测方法能够有效提高QoS预测精度,对物联网环境下网络资源的合理分配以及对用户进行物联网环境下的个性化服务推荐具有重要意义。更多还原

【Abstract】 With the rapid growth of IoT access users and terminal devices,the differentiation of user requirements is becoming more and more obvious,and users have higher requirements for the quality of service(QoS)of the network.How to realize the rational distribution of IoT resources and the personalized service recommendation for users through QoS prediction has gradually become a research hotspot of scholars.The recommendation system can mine the association between users through behavior data,and has the advantage of being highly explanatory,and has been widely used in QoS prediction.However,the traditional QoS prediction algorithm does not consider the real-time variability of QoS in the IoT environment,and the prediction is improved by the improvement of a single similarity calculation method,resulting in low prediction accuracy.In order to improve the QoS prediction effect,the paper uses collaborative filtering and artificial intelligence method to study QoS prediction methods in the Internet of Things environment.The specific research work is as follows:(1)Research on QoS prediction method based on Service2 vec.Firstly,the service item is divided into high-response service items and non-high-response service items through the threshold.For the non-high-response service item set,a Service2 vec method is designed to mine the potential features of the user’s dynamic call service items and construct the service item vector.The user similarity is calculated for the user who invokes the high response service item.Finally,the QoS prediction method based on service item collaborative filtering is adopted for the non-high response service item,and the QoS prediction method based on user collaborative filtering is adopted for the high response service item.(2)Research on QoS prediction method based on AutoEncoder.This paper designs a framework of QoS prediction method based on AutoEncoder.Apply the AutoEncoder method to the response time matrix of the user-service item,obtain the user’s feature vector,perform spectral clustering on it,integrate the influence of geographical factors,perform K-Means clustering on the user’s geographic location,and finally combine the two.The weights are used for QoS prediction.(3)According to the two QoS prediction methods proposed in this paper,the control experiments are selected respectively,and experiments are carried out in the real WS-DREAM dataset.Experiments show that compared with the comparative QoSprediction methods,the two QoS prediction methods proposed in this paper are obtained.Better prediction results.In view of the difference between services in the IoT environment and traditional environment services,two QoS prediction methods are proposed.Compared with the previous QoS prediction research,the proposed QoS prediction method can effectively improve the QoS prediction accuracy,and the network resources in the IoT environment.Reasonable allocation and recommendation for personalized service recommendations for users in the IoT environment are of great significance.更多还原