Research On Edge Cache Strategy Based On Spatial-temporal Correlation Analysis Of User Experience

The traffic in the mobile edge network has exploded,and mobile users’ requirements for personalized services are gradually increasing.However,due to the limited cache,the communication network will face the problem of communication congestion or even interruption.In order to solve these problems,a feasible solution is to move the active cache.Mobile active edge caching aims to increase network throughput and improve user experience by using information related to user behavior.Generally,caching strategies need to identify and cache the most popular content to make full use of edge storage capacity.Therefore,the evaluation index of the cache strategy is usually the cache hit rate.In this line of thinking,existing caching strategies have considered the popularity of content in terms of time and space from the perspective of users in the entire region.However,the hit operation cannot unilaterally reflect the efficiency of the caching strategy.For example,many users usually request a video and then close it quickly.Therefore,user satisfaction information(ie,access time or content score)should be used to improve smart caching strategies.Based on the above observations,thesis proposes a new edge caching strategy,which considers caching evaluation from both user requests and user satisfaction.First,thesis considers the relationship between combined features and classification results,and proposes an online content request prediction model.Discussed a request probability parameter learning algorithm based on FTRL to effectively solve the over-fitting problem of multi-dimensional features and obtain a smooth solution.In addition,the user prediction error is counted in the user request probability prediction model,which is used as a basis for starting the user request parameter learning model,so as to reduce unnecessary parameter learning.Further,thesis uses actual data sets to construct simulations and analyze the performance of the proposed algorithm.Finally,theoretically the upper bound of the prediction error of the user request probability is derived.Secondly,thesis introduces a collaborative filtering model with additional time/space factors to mine the time/space patterns of user satisfaction evaluation,that is,fully consider the influence of time patterns when studying user preferences,and construct a three-dimensional tensor data model.In order to reduce the burden of manual tuning,thesis uses the complete Bayesian method and considers the probability distribution of the model parameters and hyperparameters,and finally uses the Markov chain Monte Carlo algorithm to obtain the user feature matrix,content feature matrix,and time feature matrix,And then obtain the user’s preference information for a certain content at a certain time.Likewise,actual data sets are used to build simulations and analyze the performance of the proposed algorithm.Then,according to the aforementioned prediction model of user request probability and the prediction model of spatio-temporal correlation user score,an edge caching strategy based on user satisfaction awareness is proposed.This caching strategy can cache online content that is more likely to be requested by users in the current area and that is more in line with user preferences.Experimental results show that the overall cache hit rate is better than existing strategies,and the user satisfaction score for cached content is higher.In addition,the lower limit of user satisfaction for edge caching is theoretically deduced.Finally,thesis designs a caching framework suitable for the proposed new edge caching strategy based on the characteristics of access points and smart user devices.The caching framework takes into account the random mobility of users,can predict user satisfaction on-line and implement content caching and update based on this.The learning of relevant parameters is placed on smart user terminals with high personalization requirements to reduce access points.Computational burden.