Research On QoS Prediction Methods In Service-Oriented Recommender Systems

With the rapid development of Internet technology,Web service recommendation and its application research have become the focus of attention in academia and industry.At the same time,with the advent of the Web 2.0 era,the number of registered Web services on the Internet has exploded.In many serviceoriented applications scenarios,web services have the same or similar business functions,and service quality is different.How to recommend high-quality services to target users in the same or similar web services is an important hot research problem in the field of service computing research.At present,domestic and foreign scholars have conducted extensive and indepth research on QoS prediction in service recommendation,and have achieved significant research results.The research content is mainly divided into two major research hotspots:(1)research on service QoS prediction in non-temporal awareness scenarios;(2)research on service awareness of temporal QoS prediction.In the problem of service QoS prediction in non-temporal awareness scenarios,the existing research results mainly adopt the idea of collaborative filtering algorithms.However,the existing collaborative filtering-based method uses all the services invoked by a target user(or all users invoking a target service)to calculate the QoS prediction base value of the target user calling the target service,and the deviations caused by services(or users)that are not related to the target service(or target user)during the base value calculation has not been considered.Based on the shortcomings in the existing research,this paper proposes a QoS prediction method based on an enhanced collaborative filtering algorithm.Before performing collaborative filtering calculation,this method uses RBS similarity to detect neighbor services(users)and trim redundant services(or users)that are not related to the target service(or target users),thereby improving the reliability of the base value calculation.Compared with existing methods through experiments,this method significantly improves the accuracy of QoS prediction.In the problem of service QoS prediction in temporal awareness scenario,the existing research results mainly use the sequence prediction model of recurrent neural network and the sequence prediction model of statistics.However,the sequence prediction model based on the recurrent neural network has not considered that the characteristics of users and services will change with time in the process of QoS prediction,and the ability to express the temporal characteristics of users and services needs to be improved.Driven by this research,this paper proposes a GRU(Gated Recurrent Unit)-based temporal awareness service QoS prediction method.This method uses binary invoking records to reconstruct the feature representation module of users and services,so that it has the ability to express temporal features.At the same time,this method fuses similarity feature of users(services)and user(service)binary feature,it enables the feature expression to perceive the context.Compared with the existing methods through experiments,this method significantly improves the accuracy of temporal awareness service QoS prediction methods.Under the best experimental settings,the accuracy of QoS prediction is improved by 21%.