QoS prediction for web services in cloud environments based on swarm intelligence search

In the real-world cloud computing scenarios, a user usually invokes few Web services and can provide only a small proportion of QoS (Quality of Service) records. Therefore, for a whole range of services, many records for this user are missing (i.e. incomplete QoS information). These missing QoS values make it difficult to conduct accurate service recommendations. To overcome this difficulty, it is necessary to predict the missing QoS values for the user. Most existing algorithms tackle the QoS prediction problem through aggregating the QoS values of the local similar neighbors of the user, which easily leads to local optimization. In this paper, we model it as a global search optimization problem in the distribution space of QoS values and propose a novel algorithm PSO-USRec. In the algorithm, particle swarm optimization (PSO) is customized and improved by diversifying the initial solutions and smoothing the outlier particles. To validate the effectiveness of the PSO-USRec algorithm, comparison experiments are conducted on a well-known public QoS dataset. The experimental results show that the PSO-USRec algorithm significantly outperforms the state-of-the-art collaborative filtering approaches. It reduces MAE and RMSE by at least 5.42% and 1%, and at most 14.29% and 2.25%, respectively.