摘要
Web service recommendation is one of the most important fi elds of research in the area of service computing. The two core problems of Web service recommendation are the prediction of unknown Qo S property values and the evaluation of overall Qo S according to user preferences. Aiming to address these two problems and their current challenges, we propose two efficient approaches to solve these problems. First, unknown Qo S property values were predicted by modeling the high-dimensional Qo S data as tensors, by utilizing an important tensor operation, i.e., tensor composition, to predict these Qo S values. Our method, which considers all Qo S dimensions integrally and uniformly, allows us to predict multi-dimensional Qo S values accurately and easily. Second, the overall Qo S was evaluated by proposing an efficient user preference learning method, which learns user preferences based on users' ratings history data, allowing us to obtain user preferences quantifiably and accurately. By solving these two core problems, it became possible to compute a realistic value for the overall Qo S. The experimental results showed our proposed methods to be more efficient than existing methods.
Web service recommendation is one of the most important fields of research in the area of service computing. The two core problems of Web service recommendation are the prediction of unknown QoS property values and the evaluation of overall QoS according to user preferences. Aiming to address these two problems and their current challenges, we propose two efficient approaches to solve these problems. First, unknown QoS property values were predicted by modeling the high-dimensional QoS data as tensors, by utilizing an important tensor operation, i.e., tensor composition, to predict these QoS values. Our method, which considers all QoS dimensions integrally and uniformly, allows us to predict multi-dimensional QoS values accurately and easily. Second, the overall QoS was evaluated by proposing an efficient user preference learning method, which learns user preferences based on users＇ ratings history data, allowing us to obtain user preferences quantifiably and accurately. By solving these two core problems, it became possible to compute a realistic value for the overall QoS. The experimental results showed our proposed methods to be more efficient than existing methods.
基金
supported by the Natural Science Foundation of Beijing under Grant No.4132048
NSFC (61472047),and NSFC (61202435)
