Online Service Recommendation For Dynamic QoS Data

With the rapid development of the Internet and the associated information technology,the services with the same functional properties but different non-functional properties have emerged in large numbers.The designers of service-oriented applications need to choose the suitable services from a broad pool of functionally identical or similar Web services when building applications,which means that users have more opportunities to find the services meet their needs,but also cost more time to find suitable services.Therefore,service recommendation plays an important role in the process of building service-oriented applications.However,most of the traditional service recommendation methods work off-line,which predict the unknown QoS by modeling the static QoS value.The off-line working ignores the fact that QoS is dynamic.Due to QoS performance is highly related to many dynamic factors which may change over time.For example,when time varies,workload and network environments of services or users may change and QoS will be affected accordingly.Thus,QoS recommendation approaches are required to process the unknown QoS values timely and accurately.In order to improve the prediction accuracy,efficiency and recommend quality,the main works of this thesis are as follows:Firstly,according to the dynamic characteristics of QoS and noise problems,this thesis proposes the prediction approach for dynamic QoS,the approach considering the factors that cause the change of QoS(network noise,client and server network status,etc.)while predicting the unknown QoS.In addition,the approach adjust the learning rate while updating the model.According to the experiments,the approach has improved the prediction accuracy of QoS.Secondly,this thesis puts forward the online recommendation approach based on the methods of learning to rank.In order to improve the quality of recommendation,the approach uses the method of learning to rank to construct a sorted list and performs the optimization process on the sorted list with considering the characteristics of QoS.At the mean time,the approach employs the online learning method to improve the efficiency of recommendation.Finally,based on the methods and the theories above,this thesis designs and realizes the simulation system,and makes further experiments and analysis of results.The experiments results also reflect the correctness and feasibility of our approaches.