Abstract
Web services is the corner stone of many crucial domains, such as cloud computing and the Internet of things. In this context, QoS prediction for Web services is a highly important and challenging issue. In fact, it allows for building value-added processes as compositions and workflows of services. Current QoS prediction approaches, such as collaborative filtering methods, mainly suffer from the problems of data sparsity and cold-start obstacles. In addition, previous studies have not explored in depth the impact of geographical characteristics of services/users and QoS rating on the prediction problem. To address these difficulties, we propose a deep learning-based approach for QoS prediction. The main idea consists of combining a matrix factorization model based on a deep auto-encoder (DAE) and a clustering technique based on the geographical characteristics to improve the prediction effectiveness. The overall method proceeds as follows: First, we cluster the QoS data using a self-organizing map that incorporates the knowledge of geographical neighborhoods; by doing so, we allow for the reduction of the data sparsity while preserving the topology of input data. Besides that, the clustering step effectively handles the cold start problem. Second, for each cluster, we train a DAE that minimizes the squared loss between the ground truth QoS and the predicted one. Third, the missing QoS of a new service is predicted using the trained DAE related to the closest cluster. To evaluate the effectiveness and robustness of our approach, we conducted a comprehensive set of experiments based on a real-world Web service QoS data set. The experimental results showed that our method achieves a better prediction performance compared to several state-of-the-art methods.