A Hierarchical Matrix Factorization Approach for Location-Based Web Service QoS Prediction

With the rapid growth of population of service-oriented architecture (SOA), services are playing an important role in software development process. One major issue we should consider about Web services is to dig out the one with the best QoS value among all functionally-equivalent candidates. However, since there are a great number of missing QoS values in real world invocation records, we can hardly do a detailed comparison among those selectable Web services. To address this problem, we propose a location-based hierarchical matrix factorization method to make efficient and accurate QoS prediction. In our method, we consider both global context and local information. We first apply matrix factorization (MF) on global user-service records and obtain a global prediction matrix. After that, we use MF to predict QoS values on some user-service groups, which are clustered by K-means algorithm. Then we combine global and local predicted QoS values to provide our final prediction. Extensive experiments show the effectiveness of our hierarchical approach which outperforms other popular methods.