Research On QoS Reversed And Cross Prediction Based Web Service Recommendation

With the development of cloud computing, there are increasing numbers of alternative Web services occurring on Internet, which provide the same functionality but differ in quality properties. Facing such a huge collection of services, it becomes more and more difficult for user to find the required service in the Web service registration center or on search engine manually. QoS-based (Quality of Service) service recommendation, with the goal to select web services from each set of functionally-equivalent services to satisfy end-to-end QoS requirements, is becoming an important issue of service-oriented computing.Previous research has addressed this problem by making use of QoS attributes of the candidate services. However, common premise of previous research is that the QoS values of services to the target user are all known. On the contrary, a user can hardly have invoked all Web services, which means that, in real situation, there are many missing QoS values of services to the consumer.In this paper, we propose a novel QoS prediction approach DRaC to handle this issue by using Collaborative Filtering technology to predict the missing values. To improve the accuracy of QoS prediction, we introduced a mechanism named Data Smoothing. All users in training set of prediction system are clustered into several clusters. In order to make the variation of the data smoother, we use the information in each cluster to preprocess the data that will be used as the input of QoS prediction phrase. Meanwhile, a mechanism named Reversed and Cross Prediction is proposed to reduce the influence of data sparsity. Different from traditional Collaborative Filtering-based prediction approaches, DRaC algorithm makes fuller use of neighbors with low similarity to target user or service in the training set. In addition, DRaC learns from users’feedbacks for recommendation results, establishes and maintains a Feedback based Trust model between users automatically. As a result, with the executions of our DRaC prediction approach, he accuracy of QoS prediction can be improved dynamically. Finally, experiment results of QoS prediction demonstrate that our approach outperforms existing methods for accuracy. Moreover, we evaluate the impact of each parameter in our QoS prediction system.