Prediction of Service Reliability Based on Grouping

With the wide applications of Internet-based services, service reliability has received more and more attention in the research field of Services Computing. Reliability prediction approaches basically compute reliability according to historical records. However, in most current prediction methods, each user is supposed to return a Boolean feedback representing his/her subjective evaluation on service reliability. This binary feedback cannot express the user's evaluation of reliability precisely. Even though some researchers propose to compute reliability according to feedback from similar users or similar services, accuracy of reliability prediction cannot be guaranteed successfully. To address the problems above, this paper presents a fine-grained reliability prediction method based on grouping (RPMBG). In RPMBG, feedback is divided into several dimensions and each user can choose a service degree for each dimension to represent his/her evaluation on consumed service. A quantifying approach of users' feedback on each dimension will be given with the employment of entropy in information theory. At the same time, a grouping methodology including grouping of users and grouping of services is described with the employments of fuzzy c-means theory, PCC, and Levenshtein Distance. Based on grouping, filtered feedback from similar user group and similar service group will be engaged into reliability prediction. Experimental results demonstrate that our proposed RPMBG has significantly increased the accuracy of prediction result compared with other prediction methods.