AERQP: adaptive embedding representation-based QoS prediction for web service recommendation

Over the last few years, abundant and diverse Web services have migrated to the cloud. However, the disparity of the cloud environment renders quality of service (QoS) prediction harder. Based on analyzing the problems of inaccurate semantic representation and inadequate service invocation modeling in QoS prediction within the cloud, we propose an adaptive embedding representation-based QoS prediction method (AERQP) for Web services recommendation. First, the optimal embedding dimension of an explicit feature is determined dynamically by a policy network. Next, the embedding representation is remapped based on linear transformations. Then, global feature interactions are learned through a deep network with multi-head external attention as the core to fully model service invocations and realize accurate QoS prediction. Last, the experiment results indicate that AERQP improves an average of 44.8% and 16.9% on mean absolute error and root mean square error, respectively, compared to baseline methods.

Automated summary

This study proposes an adaptive embedding representation-based QoS prediction method for Web services recommendation. By analyzing the problems of inaccurate semantic representation and inadequate service invocation modeling in QoS prediction within the cloud, this method improves the accuracy of QoS prediction.