Research On Qo S Prediction Based On Deep Feature Learning

Due to the elastic on-demand service model and the characteristics of allowing wide access to the network,the number of cloud services on the Internet has exploded,resulting in a large number of homogenizations with similar functions but different quality of service(QoS)in the cloud service market.In this case,it is difficult for users to ensure that the selected cloud service can meet their complete needs in a specific environment.Therefore,combined with recommendation technology,accurate and personalized prediction of QoS has become a necessary condition to help users to choose their cloud services.In recent years,cloud service recommendation based on QoS prediction has continuously attracted attention in the field of service computing,and the research results are very fruitful.However,the ever-increasing diversity and dynamic of cloud environment pose many challenges on QoS prediction in service recommendation: such as how to explore more factors that affect QoS prediction,and how to construct more powerful prediction models to improve QoS prediction accuracy,etc.In response to the above problems,based on the multi-source information in the service invocation process,this article fully explores the various factors that affect QoS prediction,and proposes a QoS prediction method based on multi-source feature extraction;on this basis,a joint deep network based on convolutional neural network is proposed to further learn multisource feature interaction and realize QoS prediction.The main work and innovations of this paper are summarized as follows:(1)This paper proposes a novel feature extraction method based on latent factor embedding to extract multi-source features of users and services,and to realize QoS prediction by combining with multi-layer perceptron.First,embedding layer is applied to capture deep explicit features of users and services from contextual data.Then the Doc2 Vec algorithm is introduced to learn the semantic vector of the service description document to avoid the instability caused by the random initialization.Furthermore,a latent factor embedding method is proposed to extract implicit features from QoS matrix in order to integrate deep attribute information of users and services.Finally,the implicit features and explicit features are further combined to generate multi-source features,and the QoS prediction is completed by introducing a multi-layer perceptron.Experimental results show that the multi-source feature extraction method based on latent factor embedding can improve the QoS prediction accuracy and effectively alleviate the impact of data sparseness,which provide a new solution for capturing key features from different multi-source information.(2)This paper proposes a convolutional neural network-based joint deep network(JDN)for learning multi-source feature interaction and realizing QoS prediction.First,a single-hidden layer(SHL)neural network is introduced to learn feature sequence from the original multi-source features adaptively to reduce the influence of feature arrangement order on the learning.Then the significant local feature interactions are captured through the convolutional layer and pooling layer,and the global feature interactions are further captured by perception layer based on the mixed features of original multi-source features and learned local feature interactions to realize QoS prediction.Extensive experiments are conducted on a real-world dataset to confirm the effectiveness of the proposed method,and the results demonstrate that our JDNMFL outperforms traditional collaborative filtering methods and several state-of-the-art models significantly.