Font Size: a A A

Research On Personalized Cloud Services Recommendation Based On QoS

Posted on:2016-07-09Degree:DoctorType:Dissertation
Country:ChinaCandidate:X J ZhangFull Text:PDF
GTID:1108330503975922Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Cloud computing is a new type of Internet-based computing, whereby shared resources, software, and information are provided to computers and other devices on demand.Cloud computing has become a scalable service consumption and delivery platform.In cloud environment, due to their various locations and communicationlinks, different consumers will have different Qo S experiences wheninvoking even the same service.With the exponential growth of cloud computing as a solution for providing flexible computing resources, more and more services with same or similar functionalities emerge. Recommending the most satisfied cloud services for consumers has become oneof the most challenging issues in the field of cloud computing. To recommend the most suitable services to consumers according to their needs, it is critical to design a method which could efficiently and precisely predict the quality of services.The main research and contributions of this thesis is described asfollows:i). Combining the model-based collaborative filtering algorithm with the memory-based one, a clustering-based personalized Qo S prediction approach is presented for cloud services. Memory-based CF algorithms areeasy to implement, require little or no training cost, and caneasily take new consumer’s ratings into account. But they have data sparsity and scalability issues. Their online performance is often slow. Model-based CF algorithms can quickly generate recommendations and achieve good onlineperformance. Moreover, they are very efficient. However, the drawback is that the model mustbe performed anew when new users or items are added to thematrix. For dynamic environment and large consumers and services, we design ahybrid clustering model-based and memory-based CF algorithm forpersonalized Qo S prediction(CBHP), which significantly improvesthe prediction accuracy comparing to traditional CFalgorithms as well as the state-of-the-art one.ii). Considering the contextual factor in the process of prediction, a personalizedcontext-aware prediction method for cloud services is proposed. The cloud servicesconsumers are located in different geographic and network environments.Since the consumers invoke services via different communication links, the quality of services they observed are diverse. So we propose a personalized context-aware Qo S prediction approach by considering the location factor into the clustering-based algorithm. The approach will first clusterusers into several regions based on their physical locationsand historical Qo S similarities. Then region-sensitiveservices are identified.After that, the approach is used to automatically predict the Qo S ofthe candidate web services for an active user by leveraginghistorical Qo S information gathered from users of highlycorrelated regions. The experimentspresent that the contexts of consumers and services have influence on performanceof Qo S prediction.iii). Analyzing the uncertainty in the process of prediction, confidence model for Qo S prediction in cloud computing is built. In cloud environment, the number and value of the collected Qo S data may fluctuate due to the distribution and dynamic of the cloud services. A Qo S prediction based on a small set of data or the data that has a largevariationislikely to be unreliable. Moreover,the data collection time will also affect the credibility of prediction. To make the prediction result more accurate and credible, these uncertainties must be deal with in the prediction process. We present a probabilistic model to quantify confidence in Qo S prediction. We doso by integrating three reliability measures: the number of Qo S data items used in prediction, the variation of data in the dataset and the decay of data over time.The experiments show thatour confidence model can help to recommend satisfied services to consumers based on their requirementseffectively.iv). Researching the multiple attributes of Qo S, a recommendation method over multiple attributes for cloud services is introduced. We explain the limitation of existing approaches in dealing with multiple attributes by example.A simple improvement to existing methods is given. When selecting the data used in prediction, data that meet consumer’s requirement for multiple attributes simultaneously will be picked out. Doing so, the problem generated by predicting Qo S through each Qo S attribute individually will be avoided. The qualities data of these attributes are monitored asynchronously, so these asynchronous data need to be deal with when selecting the data used in prediction.In this thesis, we employ a k NN based technique to predict the missed value. Then, we rank the services based on both the predicted Qo S value and the confidence value of prediction. And the first service will be recommended to consumer. The experiments show that the hybridapproach of SIAM and k NN(HSIk NN) can handle asynchronous data effectively and recommend the services accurately. In the experiment we study how the setting of k in HSIk NN may affect the performance of asynchronous data handling.
Keywords/Search Tags:Cloud Service, Quality of Service, Personalization, Service Prediction, Service Recommendation
PDF Full Text Request
Related items