Research On Cache And Recommendation Strategy Optimization In Mobile Network

Over the past few decades,there has been a phenomenal growth in mobile computing,resulting in exponential growth in data traffic.By deploying popular content in edge nodes close to users,mobile edge caching technology can reduce user access latency and repeated transmission of the same content,and alleviate the burden on backhaul.In the mobile edge caching system,the introduction of recommendation mechanism can reshape users’ requests,improve the homogeneity of user request behavior,and thus improve the cache hit ratio and caching revenue.However,the cache capacity of MEC nodes is limited,hence effective caching strategies need to be designed to optimize the efficiency of edge node caching.This thesis mainly studies the mobile edge caching technology and recommendation mechanism in the mobile network.The deep learning model in the recommendation system is used to predict user preferences,and the LSTM neural network is used to predict the information of user’s location.Based on the predicted user behavior information,the cache strategy of the base stations and the recommendation strategy of the users are optimized with the aim of maximizing the benefits of the operator cache and minimizing the delay of the content transmission delay.Specific content and innovation points are as follows:(1)For the multi-base station collaborative cache scenario,a recommendation and cache optimization algorithm based on user preference prediction is proposed.Under the limitation of recommendation list size,cache capacity and channel bandwidth,the optimization problem of minimum total delay is constructed by combining optimization recommendation,user access and cache strategy.The problem is non-convex nonlinear,and the coupling between the three optimization variables makes the optimization problem difficult to solve.This thesis proves that the optimization problem is NP-hard.In order to solve the optimization problem,the joint optimization problem is decoupled into three sub-problems,namely,recommendation optimization problem,user access optimization problem and cache optimization problem.Considering the user experience,the range of files in the recommendation list is restricted to avoid the aversion caused by the recommendation mechanism.It is proved that the relaxed user access optimization subproblem is a non-convex optimization problem and user access decisions are obtained based on the alliance game algorithm.It is proved that the cache optimization subproblem is equivalent to a monotone submodular function maximization problem with multiple knapsack constraints,which is solved by an advanced greedy algorithm.In order to obtain user preference information,the Deep Crossing model of deep learning recommendation system is applied to predict user preference.The simulation results show that,compared with the traditional model,using the Deep Crossing model to predict user preferences can achieve higher accuracy,and the proposed algorithm can also effectively reduce the total system transmission delay.(2)Aiming at the problem that user mobility affects cache hit ratio,a mobile edge caching algorithm based on user location prediction is proposed.The algorithm divides a day into several time periods.After training the LSTM model with the historical location data of users,the user location information is predicted,and then the user group distribution of the base station service area in the next time period is obtained.Then the user group distribution information,preference distribution and the user request information of the previous time period are combined to calculate the file request distribution of the next time period.Finally,under the limitation of cache capacity and taking cache strategy as optimization variable,the optimization problem with maximum benefits for operators is constructed.The problem can be simplified to 0-1 knapsack problem,and the optimal solution of each base station cache strategy is obtained by dynamic programming algorithm.Simulation results show that the proposed algorithm can effectively reduce the backhaul cost and improve the caching revenue of the mobile operators.(3)A Stackelberg game algorithm of content providers and operators based on user behavior prediction is proposed to solve the problem of competition between content providers and operators to maximize their own interests.Firstly,LSTM model and Deep Crossing model are used to predict user preference and user location.In order to solve the pricing problems of content providers and cache decision-making problems of operators,non-cooperative Stackelberg game framework is adopted,in which the content provider acts as leader and the operator acts as follower.Considering user payment,file size,backhaul cost and file request distribution,the utility functions of content providers and operators are defined respectively.The iterative method is used to obtain the optimal unit rent of content providers,and the dynamic programming algorithm is used to obtain the optimal cache decision of operators.In order to further improve the cache income of operators,this thesis introduces the recommendation mechanism and optimizes the recommendation decision based on simulated annealing algorithm.Simulation results show that the proposed algorithm can converge at Stackelberg equilibrium point to maximize the benefits of both sides of the game,and the introduction of recommendation mechanism can also effectively improve the benefits of the operator.