Research On Web Services Recommendation And Composition In Dynamic Environment

The development of Internet technology promotes the evolution of computing model.A new computing paradigm,i.e.service computing,came into being.Service computing allows applications of different platforms and different protocols to be integrated and interact.Then,data,software,storage,computing,and other resources are all packaged into services to be delivered to users.With the prevalence of Web service,service computing has attracted more and more attention from industry and academia.So services of various functions emerge in an endless stream.Although the growth of the number and the type of services are rapid,the qualities of services(QoS)are different.Therefore,it is very difficult for users to find the desired services from the mass of services.Accordingly,there is an urgent need for service recommendation systems to recommend services which can give the best experience to users.Meanwhile,as enterprise applications becoming more and more large-scale,a single service can hardly meet the complex requirements from users.The flexibility of service interaction and integration makes it possible for Web services to integrate seamlessly into higher level business processes.Therefore,it becomes a very practical problem to provide composite services to meet complex requirements from users.As a result of the dynamic environment,service recommendation and service composition face great challenges in practice.For example,the provision of Web services is dynamic,that is to say new services may emerge at any time,and some existing services may disappear at any time.The behaviors of services are uncertain,such as service failures may occur during execution due to the network connection or other factors.The values of some QoS attributes are also changeable,such as the increase in network accession may lead to a longer response time.Moreover,compared with the number of users or services,the historical QoS records of service execution are very sparse,which brings great difficulties to QoS prediction-based service recommendation.Therefore,it is of great practical significance to study the approaches for service recommendation and service composition in the dynamic environment.The methods of service recommendation and composition are investigated deeply and systematically in this these,especially in the dynamic environment with dynamic provided services,uncertain service behavior,changeable service QoS values and sparse historical data.Effective machine learning methods are utilized to learn the user-service rating pattern from sparse historical data,and personalized service recommendation methods are designed to efficiently and accurately recommend services for users.Meanwhile,the service composition problem is regarded as a decision optimization problem.Thus,component services are selected in the process of service composition,which can not only execute the composite service which is most adaptive to the dynamic environment,but also satisfy the user’s QoS constraints as well as possible.The main research works of this thesis are summarized as follows:(1)Aiming at recommending accurately for users with the optimal services from the mass of services,a Top-N service recommendation model based on SVM classifier is proposed,namely SVMCF4 SR.According to sparse historical QoS records,SVMCF4 SR can train an SVM classifier effectively,and then the separating hyperplane is used to evaluate the unknown services.Thus,unknown services can be sorted according to the evaluation results,and Top-N services will be recommended to users.(2)To improve the efficiency and effect of personalized service recommendation,an idea based on granularity optimization is introduced to optimize the SVMCF4 SR model.This method can granulate the user space to distinguish the similar users at first,so as to improve the accuracy of the recommendation.And then the training sets are granulated to improve the speed of the model training.(3)In order to achieve the optimal composite service with uncertain service behaviors and QoS values,a method based on the partially observed Markov decision process,namely SC_POMDP,is proposed.According to historical QoS data,SC_POMDP learns the performance of each component service in different states.On the basis of the observed QoS values of the completed service the state of the service composition system is determined,and then the component service most adaptable to the state of the system is executed for the next task.SC_POMDP selects component services during the execution of service compositions to improve the adaptability of the service composition to the uncertain service behaviors and QoS.(4)To solve the problem of service composition with QoS constraints from users,a service composition method based on Markov decision process,namely CSSC-MDP,is proposed.On the base of historical QoS data,CSSC-MDP studies the capability of each component service to satisfy different QoS constraints.Then,according to the real-time satisfaction state of QoS constraints,the best component service is selected to satisfy the user QoS constrain as well as possible.This thesis focuses on service recommendation and service composition which are two key issues in the field of service computing.The service recommendation methods of this thesis can accurately recommend services with optimal performance for users.For complex user requirements,the proposed service composition methods can run the optimal composite service which is most adaptive to the current environment and QoS constraints.The obtained results not only enrich the theoretical research of service computing,but have significant application value as well.