Research On Cloud Service Recommendation Based On QoS Prediction

The flexible service model of cloud computing is well adopted by the industry.At the same time,it also brings about the prosperity of the cloud service market.Over the years,the cloud service market is full of massive and nearly identical functions cloud services with different quality of service(QoS).Users are often confronted with choices in massive and homogeneous cloud services.Recommendation system is the preferred technical solution to solve the information overload problem.Therefore,it is an effective way to address the current situation that to produce a set of ordered cloud service sets actively from the user's perspective based on QoS and recommendation technology.Service recommendation suffers from the problem of sparse QoS information,which is mainly caused by the characteristics of cloud services.Missing QoS infomation prediction is a straightforward way to tackle the sparse QoS problem,which is benift for higher accuracy of service recommendation.In order to produce orderly cloud service set,it is necessary to rank current candidates.Traditionally,only one attribute of QoS is concerned when cloud service is ranking.However,QoS is a set of multi-dimensional criteria.It can satisfy users' omprehensive preferences on cloud services that to add multiple attributes of QoS.Based on the above analysis,we study service recommendation based on user experience.Firstly,in order to sort the spase QoS problem,a NearestGraph method is proposed to complete one-dimensional QoS information.Then,based on the dense QoS information and the multiple attributes of QoS,a Multi-Ranking method is proposed to achieve the effect that service recommendation results meet users'comprehensive preferences.The main work of this thesis can be summarized as follows:(1)Research on missing value prediction based on one-dimensional attributes of QoS.According to the stability of users and cloud services in the cloud service market,we use QoS values to measure the status of roles and put forward the concept of role stability.Based on memory-based collaborative filtering algorithm,we use graph structure to expose the internal relationship of roles in cloud service market.And we integrate role stability by extending nearest neighbor graph.Thus,we present a NearestGraph algorithm.The experimental results also confirm the effectiveness of NearestGraph,which is in line with the original intention of NearestGraph.For the role of instability,the accuracy of missing value prediction is higher.(2)Research on ranking of cloud services based on multi-dimensional attributes of QoS.In this thesis,we analyze the shortcomings of KRCC by comparing the ranking similarity in statistics.And we give an improved SRCC based on the features of the current cloud service market.In order to meet users'non-functional comprehensive requirements on cloud services,we add the multiple attributes of QoS,and improve the existing methods of selecting users' nearest neighbors.So far,we present a Multi-Ranking algorithm.In the experimental part,the Multi-Ranking is comparing with the existing cloud service ranking algorithms.And we explore the performance improvement of Multi-Ranking,which proves the effectiveness of the improved SRCC.