A Two-Phase Learning Approach Integrated with Multi-Source Features for Cloud Service QoS Prediction

Newswise — With the growing popularity of service-oriented computing, a large number of functionally similar cloud services have emerged. Quality of Service (QoS) has become a key factor for users to distinguish these services. However, QoS values are often unavailable due to insufficient user evaluations, provider data limitations, software malfunctions, network failures, and other issues, which poses challenges to scenarios such as resource allocation, service selection, and service composition.

Therefore, Fuzan CHEN, Jing YANG, Haiyang FENG, Harris WU, Minqiang LI from Tianjin University and Old Dominion University jointly carried out a research entitled “A two-phase learning approach integrated with multi-source features for cloud service QoS prediction”. This study proposes a new QoS prediction method called Multi-source Feature Two-phase Learning (MFTL) to address the problem of missing QoS values. MFTL incorporates multiple sources of features that influence QoS, including historical invocation records, user categorical or local information, linear interactions between users and services, and complex nonlinear interactions. It adopts a two-phase learning framework to make effective use of these features: in the first phase, coarse-grained learning is conducted using a neighborhood-integrated matrix factorization model, along with a strategy for selecting high-quality neighbors for target users to capture linear interactions (low-order features); in the second phase, reinforcement learning through a deep neural network is used to capture complex nonlinear interactions (high-order features) between users and services.

The research team conducted several experiments using the WS-Dream dataset to assess MFTL’s performance in predicting response time QoS, with Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as evaluation metrics. The results show that MFTL outperforms many leading QoS prediction methods, especially in sparse data scenarios, and its prediction accuracy is less affected by data sparsity, which has important practical significance for improving the reliability of cloud service selection and composition.

The paper “A two-phase learning approach integrated with multi-source features for cloud service QoS prediction” is authored by Fuzan CHEN, Jing YANG, Haiyang FENG, Harris WU, Minqiang LI. Full text of the open access paper: