A Factorization Machine-Based QoS Prediction Approach for Mobile Service Selection

Abstract

Mobile services allow us to access the abundant and various resources (including data and services) on the Internet or devices in the physical world via wireless network technologies. It becomes increasingly popular to create mobile applications by combining existing mobile services. Mobile service selection is an important issue since different services with equivalent functions may have quite different qualities (e.g., performance). Even the same service may present different performances due to the volatility of mobile environments and move of users. Hence, getting aware of the quality of mobile services is a crucial need in service selection. To meet this need, a dozen of quality-of-service (QoS) prediction approaches have been proposed for traditional Web services and mobile services. However, their prediction accuracy and time efficiency still have plenty of room for improvement. This paper proposes a collaborative filtering approach to predict the QoS of mobile services based on factorization machines. Factorization machines significantly improve the traditional collaborative filtering techniques in both accuracy and time efficiency. The proposed approach revamps the classic factorization machine model by incorporating the locations of service users to better fit the mobile environments. The experimental results based on real-world QoS data show that the proposed approach outperforms the other collaborative filtering approaches.