| Blockchain is a distributed ledger technology that is widely used for building next generation applications and addressing the lack of trust that exits in such applications.With the cease less evolution of service-oriented computing(SOC)paradigm,blockchain as a service(Baa S)has emerged.Users can build their own applications by selecting the peers of the blockchain.However,unreliable blockchain peers will make users waste resources and even cause the loss of information disclosure.Users must select highly reliable blockchain peers that offer excellent quality of service(Qo S).Due to the significant number of blockchain peers and the sparsity of personalized Qo S data,it is difficult to select the optimal peers.Meanwhile,the blockchain peers store a lot of context information,such a geographic location,IP address and so on.How to extract the reliability related information from these information has become a difficult problem.With the development of cloud computing,the service types of cloud center are increasing exponentially.The selection of cloud services faces the same problem.In order to obtain missing cloud service Qo S data,many researchers use collaborative filtering(CF)technology in various recommendation systems for service selection.Inspired by cloud computing services,we propose two methods to predict the missing Qo S data,and according to the predicted Qo S data,we compute the reliability of blockchain peers:· Aiming at the problem of personalized QoS data sparsity,we propose a blockchain service reliability prediction framework(BSRPF)based on matrix factorization under Baa S.First,in this framework,the user initiates a request to the blockchain peer,and the peer returns the result of the request to the user,and then the user submits the result to the database.These results will be calculated in the database to get Qo S data and generate the original user item matrix.Finally,we enrich the Qo S information by matrix factorization and filling the missing Qo Sdata.· Aiming at the problem of system cold start,we propose an algorithm based on the geographic information and matrix factorization named Geographic Information Matrix Factorization——GIMF.First,there is a series of context information in each user’s Qo S data,including the user’s country and city.The geographical similarity between users is calculated through the latitude of user’s location,and then the invocation similarity between users is recoreded through user’s historical record.After the two similarities are linearly fused,the fused similarity is used for preference propagation to solve the problem of user cold start.The performance GIMF is better than other algorithm under cold start scenario.· Aiming at how to mine the context information in blockchain peers,a modified neural collaborative filtering(MNCF)algorithm based on neural collaborative filtering(NCF)is proposed.Firstly,this method considers the context information of the blockchain nodes,such as the geographic information,the IP address of the blockchain peers,etc.,and mines the reliability information of the blockchain hidden in these context information.Secondly,the relevant factors of calculating the blockchain Qo S data are also taken as the prediction target,and multi task learning is introduced.Multi task learning can improve the accuracy of prediction by averaging the noise and parameter sharing on the main task.On a real-world block chain data set,the accuracy of the three methods proposed in this paper is significantly better than other existing blockchain service prediction algorithms. |
