Research On Location-aware Web Service Quality Prediction Methods Based On Deep Learning

Today,the exponential growth in the number of Web services has resulted in a large number of similarly functioning Web services on the Internet.Users usually choose the best service from a set of candidate services based on Quality of Service(Qo S),but in practice it is not possible to obtain Qo S values for all candidate Web services.Hence,the problem of Qo S prediction is a primary challenge in the field of service recommendation.Collaborative Filtering(CF)has been widely used in the field of Web service Qo S prediction.Memory-based CF methods can only learn low-dimensional linear relationships from users and services,while model-based CF methods cannot capture high-dimensional non-linear relationships between users and services.As a result,deep learning was introduced into the field of Qo S prediction research by building neural networks to learn high-dimensional non-linear relationships between users and services.It has been found that location information is closely related to Qo S values and its incorporation into the model can further improve the accuracy of prediction.However,most of the existing methods still have many shortcomings,such as the feature relationships between users,services and location information need to be further explored,how to take into the interactive learning of both local and global features,better solve the data sparsity problem and further improve the prediction accuracy of the model.This paper addresses these shortcomings and proposes the research on location-aware Web service quality prediction methods based on deep learning,with the following research contents:(1)In this paper,we proposed a deep collaborative filtering model combining local and global location information.In the interaction layer of the model,we cross product the embedding vectors of user and service location information and obtain global location information after using a layer of the attention mechanism.Followed by extracting high-dimensional non-linear relationships of users and services through the multi-layer perceptron and supplementing it with the dot product operation to learn low-dimensional linear relationships of users and services.Extensive experiments on the real dataset WS-Dream validate that the prediction performance of the DCLG model is significantly better than the six existing Qo S prediction methods.(2)The convolutional collaborative filtering model combining local and global features is proposed,which builds on the DCLG model by introducing the convolutional neural networks instead of the multi-layer perceptron to capture global information features while also better focusing on local neighborhood features of users and services.Extensive experiments on the public dataset WS-Dream validate that the proposed model has higher prediction accuracy as well as better convergence.