TRQP: Trust-Aware Real-Time QoS Prediction Framework Using Graph-Based Learning

QoS prediction algorithm requires to be real-time to be integrated with most real-time service recommendation or composition algorithms. However, real-time algorithms are prone to compromise on the solution quality to improve their responsiveness, which we aim to address in this paper. The collaborative filtering (CF) technique, the most widely used QoS prediction method, consider the influences of all users/services while predicting the QoS value for a given target user-service pair. However, the presence of untrustworthy users/services, whose QoS invocation patterns are different from the rest, may lead to degradation in prediction accuracy. Moreover, in many cases, the quality of the prediction algorithms often deteriorates to ensure faster responsiveness due to their inability to capture non-linear, higher-order, and complex relationships among user-service QoS data. This paper proposes a trust-aware QoS prediction framework leveraging a novel graph-based learning approach. Our framework (TRQP) is competent enough to identify trustworthy users and services while learning effective feature representation for finding a rich collaborative signal in an end-to-end fashion. Our experiments on the publicly available WS-DREAM-1 dataset show that TRQP is not only eligible as a real-time algorithm but also is well capable of handling various challenges associated with QoS prediction problems (e.g., extracting complex non-linear relationships among QoS data) and outperformed major state-of-the-art methods.