Research On Video Caching Strategy In Mobile Edge Caching

With the rapid development of video service,the traffic burden in wireless network is increasing.Mobile edge caching technology can effectively reduce the traffic burden by storing contents on edge servers.However,due to the limited storage and computing capacity of edge device,it is crucial to select cache content to maximize cache efficiency.Therefore,to solve the content selection problem in mobile edge cache,this paper studies the two scenarios of single content provider and multiple content provider respectively,aiming at improving cache hit ratio and optimizing user experience.To solve the problem of content selection in a single content provider scenario,this paper proposes an edge caching strategy based on user preference prediction model(Neural Collaborative Filtering without Content Popularity,NCFCP).Since content popularity and personal preference have different variation rules,we adopt a separate modeling method to improve the accuracy of user preference prediction.To maximize the cache hit ratio,an optimization problem is proposed and decomposed into personal preference prediction and content placement.In terms of personal preference prediction,NCFCP can more accurately model the relationship between users and content.In terms of content placement,a caching mechanism is designed to take user activity into account.Performance evaluations in real data sets show that the algorithm performs 18%and 20%better than the benchmark algorithm on average in terms of cache hit ratio and user satisfaction,respectively.To solve the problem of content selection in multi-content provider scenario,this paper proposes an edge caching strategy based on federated learning(Federated Matrix Factorization for Group,FedMF-Group).The lack of interaction of user data between various content providers hinders the improvement of the algorithm model.Therefore,this strategy improves the traditional centralized matrix decomposition algorithm and adopts the encrypted distributed framework.In a nutshell,FedMF-Group take "edge server-content providers local server" mode,each content provider calculates a gradient for a single user on its own server from the user’s raw and content interaction data,then aggregates the gradients of each user to obtain regional gradients.Then each content provider uploads the encrypted region gradient to the edge server for training.Finally,a content caching mechanism based on group preference degree is designed.The performance evaluation in real data sets shows that the algorithm approaches the centralized cache strategy under ideal conditions in terms of prediction accuracy and cache hit rate.Compared with the encrypted user-level matrix factorization strategy,the time consumption of the algorithm is greatly reduced,which achieves a better fusion of efficiency and security.