Affiliations: [a]
Department of Equipment Economics and Management, Naval University of Engineering, Wuhan, China
| [b]
Department of Computer Engineering, Naval University of Engineering, Wuhan, China
Correspondence:
[*]
Corresponding author: Kai Su, Mailbox 220, Jiefang Avenue 717, Wuhan, China. Tel.: +86 27 65461835; E-mail: keppelsue@163.com.
Abstract: Strongly promoted by the development of Service-Oriented Computing and Cloud Computing technologies, a large number of functionally equivalent web services emerge on the Internet. Quality of Service (QoS) is becoming a key factor to distinguish different web services. Although many Collaborative Filtering (CF) based approaches are recently proposed to predict the QoS of web services, the prediction accuracy are not satisfactory, since they rarely take full use of the neighbor information and latent feature information contained in the historical QoS data. Especially when the real-world QoS data is highly sparse, the previous works fail to detect the actual relationships between services. In this paper, we present a novel hybrid web service QoS prediction approach that systematically combines the memory-based CF and model-based CF. Firstly a non-negative matrix factorization model for web service QoS prediction is presented. Then an Expectation-Maximization (EM) based approach is designed to learn the model for making further prediction. The service neighbor information, which integrates the direct similarity and transitive indirect similarity of services to handle the data sparsity problem, is introduced into EM based learning process to make the prediction results more accurate. Large-scale real-world experiments are conducted using the WSDREAM QoS dataset. The experimental results demonstrate that our approach can achieve better prediction accuracy than other state-of-the-art approaches.
Keywords: Web service, QoS prediction, collaborative filtering, non-negative matrix factorization, EM, Data sparsity, similarity transition
