Research On The Application Of Context-aware Collaborative Filtering In Network Service QoS Prediction

With the rapid development of network service technology,network services in the number and type of more and more.When the user face in the choice of network services,they will face a lot of similar services,how to choose the best performance of the service and be applied to become an urgent problem to be resolved.Due to the openness of the network environment,the fluctuation of the load of the service itself,the uncertain and volatile environment factors such as the uncertainty of the user's needs,the quality of service(QoS)of the network service is uneven,how to accurately predict the network Service QoS is particularly important in the field of service computing and cloud computing.The traditional prediction techniques and applications are based on the historical statistical data to predict the future trend.In recent years,the concept of collaborative forecasting has been proposed for the application of QoS prediction.Collaborative QoS prediction referred to collaborative filtering to assist the choice of auxiliary service.It is based on the principle that similar users tend to observe a comparable quality for the same target service,and forecasts the unknown QoS value given the current user and service.Collaborative QoS prediction can help users to complete the QoS prediction task by using the existing historical data in the large service systems,and thus avoid the intuitive QoS measurement.However,existing methods which exploit user-based collaborative filtering,item-based collaborative filtering,or matrix factorization technique,only consider the relationship between the user and the service.The use of contextual information on the QoS prediction is insufficient,thus the prediction effect is difficult to reach the expected level.Inspired by this,we have proposed a context-based matrix factorization method for collaborative QoS prediction where a deviation model is taken as a way to realize context-aware in matrix decomposition.It is assumed that each contextual factor will produce a deviation in the quality of service,and introduces the variables to model these deviations.These deviations are further learned through optimization from data.WSDREAM dataset is utilized to judge the prediction accuracy in terms of MAE and RMSE.The methods of statistical methods,heuristic collaborative filtering and matrix factorization are compared and analyzed.It is proven by experiment that in the sparse data,the matrix factorization based on context is more accurate than the other models in response time and throughput.