| With the vigorous development of technologies such as cloud computing,Internet of Things,and service-oriented architecture,more and more publicly accessible Web service resources are gathered on the Internet,which inevitably leads to competition among homogeneous services.How to distinguish their quality from the perspective of Quality of Service(QoS)and predict unknown QoS values via collaborative QoS prediction,so as to effectively recommend suitable service resources for users,has become an important research topic.Considering the neglect of the network topology has limited the prediction accuracy of existing research,this work focuses on the research of topology-aware QoS prediction.Based on deep learning and reinforcement learning,two specific neural models are designed for static and dynamic QoS prediction,respectively.For the static QoS prediction,we first attempt to fully consider the Internet network topology and propose a topology-aware neural model(TAN)for highly accurate QoS prediction.In the TAN model,the features of users,services,and intermediate nodes on the communication path are projected to a shared latent space as input features.To jointly characterize the invocation process,the path features and end-cross features are captured respectively through an explicit path modeling layer and an implicit cross-modeling layer.After that,a gating layer fuses and transmits these features to the prediction layer for estimating unknown QoS values.Experimental results on two real-world datasets demonstrate that TAN significantly outperforms state-of-the-art methods on the tasks of response time,throughput,and reliability prediction.Also,TAN shows better extensibility of using auxiliary information.For the dynamic QoS prediction,an intelligent route estimation model for dynamic QoS prediction(IRE4DQP)is proposed.In the IRE4 DQP model,the intelligent route estimation(IRE)is modeled as a Markov decision process,and the policy optimization is achieved by the representation learning based on a graph attention network.Inspired by inverse reinforcement learning(IRL),a neural network for dynamic QoS prediction(DQP)is introduced to approximate the cost function,where the error of QoS prediction is calculated and simultaneously used as the cost function of the IRE model.By this,the IRE model jointly connects with the task of DQP.Trained by the nested algorithms of REINFORCE and maximum entropy IRL,IRE4 DQP can dynamically estimate the possible routes and generate high-precision predictions. |
