Research On Coded Distributed Computing Scheme For Privacy Protection In Information Recommendation

In recent years,the rapid development of mobile Internet technology has confronted users with a serious information overload problem,i.e.,users are unable to quickly find information that is truly useful for them from the vast amount of network data,and information recommendation technology is an effective way to solve this problem.The traditional centralized information recommendation technology performs data processing by a single computer,but this method gradually fails to meet the computation demands brought by the increasing amount of data in the network.Therefore,distributed information recommendation technology that uses multiple computers to collaborate on computation tasks has become a key research direction.However,this research faces two major challenges.Firstly,the distributed computing nodes can be attacked to reveal the computation data which leads to the leakage of the users’ privacy.Secondly,the computing nodes can also become stragglers,resulting in a significant increase in computation time.Related researches show that linear coding techniques can effectively solve these two problems.However,the use of linear coding to solve these two problems at the same time may bring additional computation overhead,which increases the computation time and communication load and reduces the computation efficiency.Therefore,in this paper,we design specific coding distributed computing schemes for different stages of the matrix factorization algorithm,with the objectives of protecting the users’ privacy,solving the straggler problem,and further improving the computation efficiency.The details are as follows.In this paper,we study a weakly secure coded distributed computing scheme that can be used in the matrix factorization algorithm,which ensures that the computation nodes cannot obtain the original computation data.Specifically,we construct data redundancy by encoding the original data to ensure data security to avoid the leakage of the users’ privacy.Then,we send encoded blocks which are encoded by different encoding coefficients to different computation nodes to solve the straggler problem.We verify the effectiveness of the proposed scheme in protecting the users’ data privacy through theoretical analysis,and prove that the proposed scheme can solve the straggle problem and reduce the computation latency through simulation experiments,and prove that the proposed scheme can reduce the computation time of each iteration in the gradient descent solving process of matrix factorization algorithm through real experiments.In addition,we investigate the privacy coded distributed computing scheme for the users’ interest privacy protection,which can ensure that the users’ interest preferences are not known by the computing nodes when using the data obtained from matrix factorization to make recommendations to users.Firstly,we use encoding to introduce data redundancy to computing nodes to solve the straggle problem.Then,we design a coded computing scheme using the idea of obfuscation,which can ensure that the users’ interest privacy is not leaked during the computation process.Finally,we demonstrate the effectiveness of the proposed scheme in protecting the users’ interest privacy,and prove that the proposed scheme can not only solve the straggle problem and reduce the computation latency,but also further reduce the communication load and improve the computation efficiency.