Reliability Prediction for Service-Oriented System via Matrix Factorization in a Collaborative Way

The reliability prediction of service-oriented system is a key problem, and has become increasingly important along with the wide utilization of service-oriented architecture. In this paper, we aim to improve the prediction accuracy of reliability in a collaborative way. First, we estimate the failure probability of each component through two independent models extended from Matrix Factorization. For each service and user, we identify the similar neighbors through similarity computation. Then, we build the service neighborhood-based MF model (SN-MF) and user neighborhood-based MF model (UN-MF). In the two models, each unknown failure probability is learned out assisted by similar neighbors' historical failure records collaboratively. Further, we combine the two models together to build an ensemble model, and explicate the way of calculating the final failure probability of the whole system. Afterwards, the failure probability is mapped to reliability with a classical function. Finally, experiments conducted in a realworld dataset demonstrate the effectiveness of our models.