Your neighbors are misunderstood: On modeling accurate similarity driven by data range to collaborative web service QoS prediction

Abstract

Quality of service (QoS) is a set of non-functional attributes of Web services for differentiating enriched Web services with same or similar functionality. Predicting the unknown QoS of Web services for service users is often required for any QoS based service computing because QoS plays a fundamental role in reliable Web service recommendation, composition and selection. Existing collaborative filtering based QoS prediction methods suffer from a serious acclimatization issue caused by the difference of QoS data range, which dramatically degrades the prediction accuracy and even impedes its adaptability. The fact that Web service QoS data exhibit large service effect with different data ranges, is verified on public real-world datasets. In this study, we aim to tackle the problem of QoS prediction while considering the influences of QoS data range in the context of collaborative filtering. In particular, a simple yet effective similarity model called JacMinMax, which is driven by QoS data range, is designed. Furthermore, two neighborhood selection strategies using JacMinMax are proposed, and the obtained neighbors are systemically integrated into neighborhood- and model-based methods for collaborative QoS predictions. Experimental results show that the proposed method efficiently alleviates the influence of the concerned QoS data ranges, and performs better than many state-of-the-art approaches with respect to accuracy.