Design Of Service Quality Prediction Model Based On Multi-feature Deep Learnin

With the rapid development of service computing,different service providers provide a large number of services with similar functions but different performance,making it more difficult to recommend appropriate services for users.Therefore,how to select the most suitable services from numerous similar candidate services has become a key issue.Quality of Service(QoS)has become an important reference for service selection and recommendation.QoS is a set of parameters used to describe non-functional attributes of services(such as throughput and response time).It is a key indicator often used in service computing to evaluate service performance.How to accurately predict unknown QoS values based on existing QoS values has become one of the main challenges.There have been many works in this field.Most of them are inspired by collaborative filtering in service recommendation,which predict unknown QoS values by collecting historical information of similar users or services.However,they are easily affected by data sparsity.Meanwhile,feature extraction and prediction accuracy need to be further improved.To solve the above problems,this paper uses deep learning to predict QoS.The main research contents are as follows:1)A deep-learning-based QoS prediction model with user-service interaction graphAt present,existing methods use historical QoS values to make predictions and perform shallow feature extraction.The existing data is not fully utilized.The deep relationships between users and services have not been discovered,which affects QoS prediction accuracy.Thus,a deep-learning-based QoS prediction model with user-service interaction graph is proposed,which makes full use of existing data to describe direct and indirect interaction relationships between users and services to construct user-service interaction graphs.Then user/service feature vector sets are obtained through similarity calculations.Finally,a twostream deep convolutional neural network is designed,based on similar user/service feature vector sets for QoS prediction.It uses deep convolutional unit to learn sample rules and update user/service feature vectors adaptively to further improve the prediction accuracy.2)A deep-learning-based QoS prediction model utilizing multi-stage multi-scale feature fusion with individual evaluationsCurrent methods only use single-scale features to predict QoS values,making the feature information incomplete,resulting in low prediction accuracy.Thus,a deep-learning-based QoS prediction model utilizing multi-stage multi-scale feature fusion with individual evaluations is proposed.It fuses three features: global,local,and individual features.The global and individual feature matrices are obtained through non-negative matrix factorization,from which global and individual features can be extracted.Since the QoS values of services are affected by the geographical location,similar users and similar services are obtained based on the distance similarity calculation,and the local features of users and services are formed.Through deep neural network,these three features are gradually fused in different stages for feature processing and learning.Individual evaluations are used to correct features in each stage,so as to predict QoS values more accurately.3)A deep-learning model for QoS prediction based on feature mapping and inferenceAt present,there are existing methods to extract features through user/service indexes or use matrix factorization to construct features,which makes features lose a lot of information in the transformation of low-dimensional space.Thus,a deep-learning model for QoS prediction based on feature mapping and inference is proposed.The model constructs a feature mapping and feature inference network for feature mapping and feature inference.It maps the features of users and services to two dimensions and uses the dimension-uplifted features of users and services to perform feature fusion to predict QoS values.