Research On Popularity Prediction And Caching Of Short Video In Mobile Edge Computing

With the rapid development of mobile Internet technology and the continuous growth of the number of Internet users,in the form of fragmented transmission,short videos integrating various social hotspots,skill sharing,public welfare education,advertising and other themes have become an important application in mobile Internet.In order to reduce the initial playback delay of short videos and improve user service experience,edge servers are usually deployed on the mobile Internet for distributed caching of short videos.However,considering the constraints of cost factors,it is impossible to meet the increasing demands of users by deploying a large number of edge servers.Therefore,how to design an effective short video caching strategy and make full use of the caching resources of edge servers to improve the service quality of users has become an important topic.By predicting the popularity of videos,the existing cache strategy preferentially cache those videos with high popularity,and then cache replacement according to the trend of popularity change.However,unlike film and television video,short video has many new characteristics,such as short life cycle and high latency requirements,with regional attribute and social attribute,lead to conventional video prediction method is not suitable for short video popularity prediction.To improve the cache hit ratio,video content providers cache video copies by renting storage resources from network operators.After network operators deploy edge servers,on the one hand,they hope to make full use of storage resources,and on the other hand,excessive increase in the number of cached video copies will lead to higher video caching costs.Therefore,how to balance user service experience and system cost is the problem to be solved in video caching strategy.Based on the above problems,this dissertation concludes as follows:（1）Aiming at the problem of regional popularity prediction of short videos in social networks,this dissertation considers the social attributes and regional characteristics of short videos,and proposes a multi-feature fusion short video regional popularity prediction algorithm.Since there are many factors that affect the regional popularity of short videos,such as video life cycle,topic,associated location,historical playback volume and social interaction,etc,in order to improve the accuracy of short video popularity prediction,short videos are classified and predicted.First,when calculating the popularity of short videos with strong regional attributes,the regional attributes will occupy a larger weight.At the same time,this dissertation sets a life cycle threshold α to represent the potential attractiveness of short videos to users,Secondly,according to social hot topics construct a social hot topic dataset,and calculate the popularity topic weight by correlating video topics with social hot topics.Then,construct a city matrix,set the regional weights of three different relationships in the same city,within the province,and outside the province according to the relationship between the video and the city,and calculate the global popularity of next time slice based on the KNN algorithm based on the historical playback volume and social interaction data of the short video.Finally,based on the regression algorithm,the weight value of each factor’s influence on the popularity of the short video region is calculated,and the popularity of the short video region is predicted according to the constructed popularity model.However,there is no publicly available standard dataset for short video popularity prediction,this dissertation conducts simulation experiments by collecting data sets of short video regional popularity changes on the short video data platform.The experimental results show that the model is significantly better than the popularity prediction algorithm based on KNN model and the popularity prediction algorithm based on user preference.（2）Aiming at the joint optimization problem of user service experience,network operator cost and content provider cost,in order to stabilize user service experience while minimizing caching cost,this dissertation designs a short video caching algorithm based on Lyapunov optimization.First,the algorithm constructs the constraints of the original stochastic optimization problem into a virtual queue,then constructs the optimization objective function and the lyapunov function into a lyapunov drift plus penalty function,and finally decomposes the optimization problem into each time slice.In the case of a stable virtual queue Then,the optimal value in each time slice is solved by the dynamic programming algorithm.Through simulation experiments,it is verified that the short video caching algorithm based on lyapunov optimization achieves better performance in minimizing the cost of content providers and stabilizing user delays.