QoS Prediction of Web Services Based on Two-Phase K-Means Clustering

QoS prediction for Web services is a hot research problem in the field of services computing. As one of the most important methods for QoS prediction, Collaborative Filtering (CF) makes prediction based on the historical QoS data contributed by similar users and services. The key issue in this process is to detect the unreliable data offered by untrustworthy users, which has attracted limited attentions so far. The utilization of unreliable data decreases the prediction accuracy greatly. In this paper, we propose a novel credibility-aware QoS prediction method (named CAP) to address this problem. Our method first employs two-phase K-means clustering to identify the untrustworthy users, which clusters QoS values for untrustworthy index calculation in the first phase and clusters users according to their index in the second phase, and then predicts the missing QoS value based on the credible clustering information. The evaluation results demonstrate that CAP provides considerable improvement on the prediction accuracy compared with other approaches and is robust against various percentages of untrustworthy users.