Design And Implementation Of A Recommendation System For Live Streaming In A Multi-source Dynamic Environment

With the rapid development of the live streaming industry,watching live streaming has also become one of the major entertainment activities for a large number of users.However,the independence of the major live streaming platforms affects the user experience on the one hand,on the other hand it is difficult to meet the regulatory requirements of the relevant departments,and it is necessary to dynamically aggregate the major live streaming platforms.In the course of the development of the live streaming platform,the platform also encountered the problem of information overload,and the recommendation in the context of live streaming is different from the recommendation of movie and video.New users and streamer introduced during the rapid expansion of the platform will exacerbate the cold start problem in the recommendation system and increase the difficulty of live streaming recommendations.Therefore,it is necessary to establish an effective and unified content perception and recommendation mechanism.In response to the above problems,using distributed spiders and other related technologies,the live streaming room information and barrage information of the relevant live streaming platform are parsed and obtained to realize the aggregation function of the live streaming platform.Furthermore,the live recommendation scenarios are divided,the Neu MF model is used for the conventional recommendation scenarios,and a recommendation algorithm RBUP based on meta-learning is proposed for the cold start problem to implement the recommendation task of the live streaming aggregation platform.The RBUP algorithm uses different global update combination strategies based on the existing Reptile method,uses the idea of learning the learning process in meta-learning,finds the appropriate initialization parameters of the model from the process of multiple recommended tasks,and uses the initialization parameters to improve the model adaptation.The ability to recommend tasks in a cold start scenario reduces recommendation costs and improves recommendation results.Based on the above core ideas,the design and implementation of a live video recommendation system for multi-source dynamic environments was further completed.Functions such as search,information,and login were implemented,and these functions were tested.Using the Movielens dataset to construct a live streaming-related dataset,and using the relevant dataset to perform an algorithm experiment on the proposed recommendation algorithm RBUP,the algorithm compared with the Me LU model improved about 1.9% on the MAE and about 1.2% on the nDCG.