Your neighbors alleviate cold-start: On geographical neighborhood influence to collaborative web service QoS prediction

Abstract Predicting the unknown quality of service(QoS) for an active user who has not previously accessed a Web service plays a fundamental role in supporting appropriate service selection, high quality service composition and reliable distributed system construction. Existing Web service QoS prediction methods suffer from the problems of data sparsity and cold-start, which dramatically degrade prediction accuracy and even impede their applicability in real scenarios. Additionally, the potentially positive but inconspicuous relation between geographical region and QoS rating interaction among users and Web services has been underestimated in previous studies. In contrast to those studies, we propose a collaborative Web service QoS prediction approach that incorporates the knowledge of geographical neighborhoods. By analyzing the geographical relationships in the real-world dataset WSDream, we observe that users and Web services are positively correlated with their geographical neighbors. Based on this observation, we first design a bottom up neighborhood clustering method for correlating geographical neighbor selection. Then, we construct two diversified similarity neighborhood regularization terms and systemically integrate them into a matrix factorization model, which achieves the knowledge transfer of geographical neighborhoods in improving QoS prediction accuracy. Experimental results have demonstrated that our approach performs more efficiently than existing methods with respect to accuracy, as well as alleviating the data sparsity and cold-start issues.