Design And Implementation Of A QoS Prediction-based Recommendation Platform For Cloud Services

The rapid development of Internet technology has led more and more enterprises and individual developers to invest in the wave of construction of Internet-related industries.Cloud computing services such as computing,storage and applications have flourished,greatly satisfying the upfront resource device construction needs of enterprises and individual developers.QoS is an important non-functional attribute for measuring cloud services,and plays an important role in the recommendation of cloud computing resources.In recent years,researchers have proposed numerous algorithms to achieve cloud service recommendation based on QoS prediction,including collaborative filtering,matrix factorization and improved algorithms based on them.However,they have always faced the problems of data sparsity and cold start.For example,when the number of services invoked by a user is small,the presence of a large number of default values in the user service invocation matrix will make prediction difficult.And when a new user or service appears,it is also difficult for existing methods to make predictions as there is no relevant record of user service calls.To address the above challenges,this thesis proposes a context and neighbourhood-aware QoS prediction algorithm and a deep learning QoS prediction algorithm based on federated learning,and completes the design and implementation of a cloud service recommendation platform based on the proposed prediction algorithms,which is able to recommend suitable cloud service resources for users,the details are as following.First,this thesis proposes a deep-learning QoS prediction method based on context and neighborhood awareness.We incorporate the attribute features of users and services into the prediction process and learn the potential and complex non-linear relationships between attribute features through neural networks to effectively solve the data sparsity problem and the cold start problem.At the same time,the population features of user and service neighborhoods are extracted by convolutional neural networks,effectively improving the prediction capability.The experimental results show that the method can effectively improve the prediction accuracy.In the mean absolute error MAE metric,the algorithm proposed in this thesis improved the performance of two QoS attributes,response time and throughput,by an average of 12.15%and 17.64%respectively,compared with the neural network MLP method in the comparison algorithm;in the root mean square error RMSE metric,it improved by an average of 6.64%and 13.55%respectively.Second,to address the issue of data privacy protection during the training process,this thesis proposes a context-aware deep learning QoS prediction algorithm based on federated learning.User data is trained locally in each client for model training,and the aggregation and update process of model parameters is performed through a central server,which can effectively protect the data privacy of users in each client and reduce the training pressure on the central server.Experiments on public datasets also demonstrate that the algorithm has competitive prediction accuracy while effectively protecting user data privacy,with an average improvement of 25.44%and 31.38%in mean absolute error MAE performance compared to the user and item-based collaborative filtering UIPCC algorithm for both QoS attributes,response time and throughput,respectively and probability factor matrix factorization algorithm PMF,the algorithm improved the average absolute error MAE performance by an average of 17.77%and 5.71%,respectively.Third,based on the proposed QoS prediction algorithms,the design,and implementation of the cloud service recommendation platform are completed.Firstly,the framework design of the cloud service recommendation platform is completed,and different QoS prediction algorithms are selected for different user types.For old users of the platform,QoS prediction is performed for all services in the candidate cloud service collection using context and neighborhood-aware QoS prediction algorithms,and based on the prediction results,suitable services are selected for recommendation.For new users of the platform,the QoS prediction of the candidate cloud services is carried out using a context-aware deep learning QoS prediction algorithm based on federation learning,which can effectively cope with the cold start problem.Finally,the platform is built and tested.