Design And Implementation Of A Short Video Recommendation System Based On Multiple Granular Recall

In recent years,with the advancement of technology,mobile communication networks and mobile intelligent terminals have developed rapidly.With the support of high-speed networks,short video data shows an exponential growth trend,and the short video industry has become one of the main traffic portals of the Internet.Rich short video content changes the user's habit and provides users with a more convenient social way.Short video types are diverse and rich in content,due to their special forms of expression,it is difficult for short video websites to provide personalized recommendation services for users.Therefore,the short video system incorporating recommendation algorithm emerges as the times require.Personalized recommendation algorithm can mine the user's preferences according to the user's personal information and historical interaction information,provide the user with the favorite content.The pros and cons of the recommended algorithm directly affect the user experience.Different recommendation algorithms and feature processing methods have different recommendation effects in different application scenarios.How to extract effective features from a large amount of short video data to obtain the optimal short video recommend algorithms,recommend content of interest to users,and avoid information overload of short video websites,which has become an urgent problem to be solved.The main work of this paper includes the following aspects:(1)To crawl the short video data of Douyin,use Fiddler to capture the data,analyze the data interface,analyze the meaning of each parameter in the data package,simulate the client to send data requests to the target server through the program,analyze the response content of the server,and finally get user-related data,short video-related data and interactive-related data.(2)To solve the problem of low accuracy of single algorithm in short video recommendation process,a short video recommendation algorithm based on multi-granularity recall is proposed.User features and short video features are extracted from the collected short video data,and the algorithm is constructed and trained.Firstly,the algorithm uses coarse-grained recall strategy and collaborative filtering algorithm to generate coarse-grained recall for users.Secondly,it uses fine-grained recall strategy and XGBoost algorithm to predict and sort short videos and generate fine-grained recall for users.Finally,it uses accurate recommendation strategy to improve HOP-Rec algorithm and generate recommendation list for users.This algorithm can accurately recommend short video for users,and achieves remarkable recommendation effect.(3)Designed and implemented a short video recommendation system,which is based on C/S architecture,SSM framework at the server,Android platform at the client,and multi-granularity short video recommendation model are embedded to recommend personalized short video content for users.