Neighborhood Regularization Based Matrix Factorization For QoS Prediction Of Network Services

The emergence and rapid development of the Internet has brought great convenience to people, it does meet the needs of all kinds of information in this era of information, but the development of the network is faster than people's expectations. A substantial increase in the amount of network information makes users can not directly and effectively get the information they really need, and reduces the efficiency of the information, which causes a problem of information overload. Faced to a large number of the network service with similar features, the user may feel difficult to judge the degree of service to meet their needs, so that they should make a comprehensive comparison of these quality of service to make the best choice. However, due to the limitation of time, cost and other factors, service providers would not layout of a large number of software sensor to monitor the quality of information of each service in the cloud environment, and users may not carry out large-scale testing one by one to experience the difference of the performance. Therefore, So the research of effective service QoS prediction and the corresponding recommendation technology, and the guide for users to select the required services, has become an urgent issue.Aiming at the shortcomings of the existing service quality prediction method, this paper puts forward using implicit neighbor relationships to optimize QoS prediction model on the basis of probability matrix factorization model (NRMF). Assume that similar users (or services) for the same service (user) tend to observe similar quality of service, we first discover implicit neighbor respectively based on services and users, and then build the implicit neighborhood regularization items and use them in the probability matrix decomposition, finally reform the objective optimization function, so as to achieve optimal prediction model parameters.Through the experiment, the proposed NRMF method is superior to other QoS prediction models based on the neighborhood relationship. Depending on the experiment and analysis we draw the following conclusions:(1) the prediction model optimized by user and service neighborhood relationship, can significantly improve the predictive accuracy of service quality. User-neighborhood-based regularization is more important than service-neighborhood-based regularization;(2) Increasing the number of neighbors is helpful to improve the prediction accuracy, but a large number of neighbor not conducive to the results forecast;(3) For the number of hidden factors, the larger value can reduce the prediction errors, but may be easy to cause an over-fitting problem in the process of optimization.