| With the continuous development of the Web 2.0 era,people have more and more channels to acquire new things and the ways to obtain new knowledge are more and more convenient.However,people are faced with the problem of information overload caused by ocean of data.In order to solve the problem of obtaining user’s personalized demand information from huge amounts of information,recommendation system came into being.Cloud-based software service,as a new type of open software model with rapid development,is facing the increasingly serious problem of homogeneity.In order to solve this problem,many scholars began to apply recommendation system to the field of cloud services.Quality of Service(QoS),as a non-functional attribute describing cloud services,can easily see the pros and cons of cloud services.Therefore,QoS prediction of cloud services has become one of the hot research issues in the field of cloud service recommendation.In the existing research,collaborative filtering algorithm is usually used for QoS prediction,but the influence of context information such as network location on the prediction results is often ignored.Because it is easy to find similar neighbors through network location information,QoS values can be predicted more accurately according to the neighbor information.In addition,in some studies,only the single influence of the user side or the service side is considered,but both of them have a certain proportion of influence on the predicted results.Based on the above problems,two network locationaware collaborative filtering algorithms are proposed in this paper to predict the missing values of QoS.The main innovations are as follows:A matrix factorization algorithm based on regularization of mixed context information is proposed,which introduces the network location information,geographic location information of users and the network location information,geographic location information,WSDL information of services.The improved Pearson correlation coefficient is used to calculate the similarities between users and the similarities between services.The effect of these information is to further filter the similar neighbors of users and services,and eliminate the interference of some non-similar items.In order to further reduce the possible influence of low similarity between target users and neighbors,candidate services and neighbors,we add the differences between users and neighbors,services and neighbors as regularization terms into the matrix decomposition.Experimental results show that our algorithm has obvious advantages over some classical algorithms.A matrix factorization algorithm based on local information and latent relations is proposed.Due to the limited number of services invoked by users in a large number of cloud services,the ability of the recommendation system to predict missing QoS values through existing invocation records is not outstanding.Therefore,we will integrate the neighbors’ information,which obtained by network location,geographic location and WSDL.The information is added to the collaborative filtering algorithm to make up for the problem of low prediction accuracy when the matrix is sparse.In order to mine the potential relationships between users and services and their neighbors,and give full play to the role of similar neighbors,we build local matrices based on user-based neighbors and service-based neighbors respectively.Then combines global information and local information,and adjusts the local information of users and services through parameters,which affect the overall algorithm.In this way,we can predict the missing QoS values in a large number of invocation relationships between users and services.Through a large number of comparative experiments,our algorithm has obvious advantages in predicting accuracy. |
