Research On Edge Caching Strategy Enabled By Recommender System

The vigorous development of live video and short video services not only enriches people’s daily life,but also leads to the explosive growth of data traffic.In order to improve the quality of user experience,edge caching technology caches content in advance in edge nodes that are closer to users to reduce user request response delays.Due to the timevarying user interests and the diversity of video content,how to accurately perceive user interest changes and content trends has become a key challenge to achieve efficient edge caching.Recommendation system technology can analyze users’ personalized user preferences based on massive data,and provide users with "thousands of people and thousands of faces" personalized recommendation services.Therefore,this thesis conducts an in-depth study on edge cache optimization based on the recommender system approach.Aiming at the difficulty of predicting user interest preference due to the time-varying user interest,A user interest mining and evolution model based on user behavior characteristics is proposed.The model uses the gated recurrent unit to remove the data noise from the user’s historical viewing sequence,extracts the main interest preference,and further uses the gated recurrent unit with the attention update mechanism to analyze the evolution process of the user’s main interest,thereby,the prediction result of user interest preference is obtained.The experimental results show that the model proposed in this thesis can predict user interest with an accuracy of 80%.Compared with the caching method based on user preference,the cache hit rate and user hit rate are increased by 21% and 16%on average.Aiming at the difference between user interests and content popularity,A multi-task learning model is proposed,which takes into account the impact of individual user interest preferences and global content popularity on user request behavior by setting different primary/secondary tasks,and further the fusion layer and the alienation layer are used to learn the coupling and difference between multiple tasks,and simultaneously predict the user’s interest preference and the global popularity of the content.On this basis,social attributes such as group user activity and positive content feedback are introduced,and individualized and global features are transformed into group features,so as to provide more efficient content caching recommendations for edge networks from the perspective of user groups.The experimental results show that the model proposed in this thesis improves the accuracy of multi-task prediction by 12% on average,and the cache hit rate and user hit rate increase by 34% and 32% on average compared to the minimum frequency usage strategy.