JUCS - Journal of Universal Computer Science 30(3): 383-401, doi: 10.3897/jucs.106314
An SVR-based and Location-aware Method for Mobile QoS Prediction
expand article infoLifang Ren, Jing Li§, Wenjian Wang§
‡ Shanxi University of Finance and Economics, Taiyuan, China§ Shanxi University, Taiyuan, China
Open Access
With the rapid development of intelligent mobile communication technology, the num­ber of mobile services and the number of mobile users are both continuously increasing. So, the services used by a user can only account for a very small proportion of the existing services, which results in a sparse user­service quality of service (QoS) matrix. However, QoS is critical for service selection and service recommendation. Therefore, predicting the unknown values of the sparse QoS matrix is essential. However, due to the sparsity of QoS data, the QoS predic­tion accuracy is difficult to improve. Faced with the problem, this paper intends to utilize the outstanding generalization ability and only support vectors dependent property of support vector regression (SVR) to overcome the difficulty brought by the sparsity of data and predict the un­known QoS more accurately. Moreover, it is evident that in the mobile environment, QoS values are closely related to the locations of the invoking users. Therefore, this paper intends to improve the accuracy of QoS prediction by incorporating not only the information of similar users but also the information of nearby users into feature vectors. On the other hand, the known QoS values of nearby users can be used to roughly estimate the unknown QoS values of the cold­start user, so as to alleviate the cold­start problem to some extent. Thus, a location­aware SVR­based method for QoS prediction (SVR4QP) is proposed. Compared with some classical QoS prediction algorithms, the experimental results show that in 1/3 of the cases, SVR4QP is moderate; in 1/6 of the cases, SVR4QP is suboptimal; and in half of the cases, SVR4QP is optimal. Compared with some novel mobile QoS prediction methods, the experimental results show that in 1/4 of the cases, SVR4QP is moderate; in half of the cases, SVR4QP is suboptimal; and in 1/4 of the cases, SVR4QP is op­timal. All these indicate that SVR4QP has comparatively more accurate mobile QoS prediction.
Mobile service, QoS prediction, Locatio-naware, Support vector regression