Research On Location-aware Service Quality Prediction Methods Based On Deep Neural Networks

With the advent of the era of big data and cloud services,Web services on the Internet are growing exponentially,and Web services with similar functions are also increasing rapidly.How to recommend services to meet users' needs among the massive services with the same functions has become a research hotspot in the field of service recommendation.Service QoS has been widely used in the field of service recommendation as an indicator to measure service non-functional attributes.However,in the current network environment,due to various constraints,it is impossible for users to personally call every service to obtain QoS value.Therefore,QoS prediction becomes an effective way to recommend services for users in a short time.The traditional collaborative filtering method is widely used in QoS prediction,but the method usually considers one-dimensional and two-dimensional linear characteristics and is susceptible to cold start,resulting in poor prediction quality.By learning the linear and nonlinear relationship between users and Web services,the QoS prediction method based on factor decomposition machine alleviates the problems of data sparse and cold start to some extent.However,due to the influence of computational complexity,the factorization machine can only deal with first-order and second-order features,so the quality of QoS prediction is still affected.Therefore,deep learning is introduced into QoS prediction research.The deep neural network is used to learn the high-order linear and nonlinear features between users and Web services,which can effectively alleviate the challenge of data sparse in QoS prediction.In addition,given that QoS is closely related to the location of users and services,users in the same locale have similar experiences on the same Web service.However,the existing QoS methods based on deep learning do not perform well in predicting quality robustness.In view of the shortcomings of the above QoS prediction methods,this paper uses deep learning technology,introduces information entropy and combines with QoS-related location information to propose a location-aware QoS prediction method.The main contributions of this paper are as follows:(1)A Location aware Convolutional Collaborative Filtering(LCCF)QoS prediction method is proposed.LCCF uses collaborative filtering method to find the similar neighbors between users and web services,and uses convolution neural network to find the high-order linear and nonlinear relationship between users and web services,improving the quality of QoS prediction,so effectively alleviates the problem of data sparsity and cold start,and achieves a good balance between coverage and accuracy.The experimental results show that the prediction accuracy of LCCF model is better than the existing QoS prediction methods.(2)This paper proposes a Location-aware Deep Factorization Machine(LDFM)for QoS prediction.LDFM solves the problem of feature bias in the process of information projection by introducing information entropy.Factor decomposition machine and multi-layer perceptron are used to mine the hidden multi-level features of interaction between users and web services from breadth and depth respectively.The experimental results show that LDFM has high operation efficiency,and is superior to the existing QoS prediction methods in prediction stability.