Research On QoS Prediction Based On Factorization Machine

With the rapid development of Internet technology,a large number of computer software has tended to be deployed in the cloud environment,and technologies such as Service-Oriented Architecture(SOA),cloud computing and Internet services are becoming more and more popular among consumers.However,with the increasing number of cloud services,lots of cloud services with similar functions have appeared on the Internet.How to choose the suitable service for users among these cloud services with similar functions has become a key issue.Quality of Service(Qo S)is widely used to describe and evaluate the non-functional attributes of cloud services.Therefore,in the case of highly homogeneous cloud services,how to accurately predict Qo S has become one of the hot research topics in the field of Service recommendation.In recent years,many scholars have used Qo S prediction methods based on collaborative filtering,combined with mining user and service contextual information,and have achieved a series of research results.However,the current research work still has some shortcomings:(1)The way to find users or service neighbors is single,and the multi-source information of users and services is not fully considered;(2)It fails to capture the complex feature interactions between users and services and ignores the importance of different feature interactions.To solve the above problems,this paper proposes a multi-source neighbors based factorization machine Qo S prediction model and an attention mechanism based factorization machine Qo S prediction model.The main contributions and innovations are as follows:(1)A multi-source neighbors based factorization machine Qo S prediction model is proposed.This model fully explores the connection between users and services,and obtains the multi-source neighbor sets of users and services from contextual information such as geographic location of users and services,network locations and description documents of services.And through the factorization machine model,it automatically combines the features of users and services with their multi-source neighbors to predict the Qo S value.Experimental results on real data sets verify that the prediction accuracy of this model is better than that of other traditional methods based on collaborative filtering.Meanwhile,the time complexity of this model is low,which is linearly correlated with the size of dataset,and it still performs well in the case of sparse data.(2)An attention mechanism based factorization machine Qo S prediction model is proposed.The model is also based on users and services and their multi-source neighbors,through the factorization machine combine feature to study the low-order and linear feature interaction.At the same time,the attention mechanism is further introduced to obtain the importance of the second-order cross-features of the users and services and its multi-source neighbors,and the key feature combination information is screened out.And the final Qo S prediction is realized through learning the high order and nonlinear feature interaction by deep neural network.Experimental results on a large dataset in the real world show that the model can effectively learn the low-order,high-order,linear and nonlinear interactions of features,and its prediction accuracy is better than the existing Qo S prediction methods.