Research And Implementation Of Short Video Hybrid Recommendation System Based On User Preferences

As the Internet technology develops,short videos are rapidly accepted by people by virtue of their fast propagation,high interactivity and high information content,and watching short videos has become a way for people to relax and entertain themselves in their leisure time.However,with the increasing number of short videos,it is difficult for people to find short videos that meet their preferences,In order to solve this information overload problem,recommendation systems have emerged,which not only help users filter out their favorite content,but also increase the revenue of personalized recommendations on platforms.However,the problems of cold start,data sparsity,and real-time that have existed in the development of recommendation systems so far have had a negative impact on the quality of recommendations,and a single recommendation algorithm can no longer meet the needs of today’s recommendation systems.To this end,this thesis investigates short video recommendation systems based on users’ preferences,with the following main contents:(1)The offline recommendation algorithm is designed for the cold start and data sparsity problems.The tag-based short video recommendation algorithm is improved using Term Frequency-inverse Document Frequency(TF-IDF)method to alleviate the problem of lack of freshness of popular short videos recommended by the tag-based recommendation algorithm for users.A dynamic weighted hybrid recommendation algorithm built by improved tag-based recommendation algorithm and Alternating Least Square(ALS)-based collaborative filtering algorithm is used to obtain the user’s preference level for unknown short videos,which solves the cold start and data sparsity problems.(2)The real-time recommendation algorithm is designed for the real-time problem.The tag-based short video similarity matrix is dynamically weighted and mixed with the short video similarity matrix generated by the algorithm based on ALS to obtain a new mixed short video similarity matrix,and then the priority of short videos is obtained using the constructed priority calculation model.In order to reduce the time for users to get the recommendation results,Kafka and Spark Streaming computing frameworks are introduced to ensure the real-time recommendation results.(3)Designed and implemented a short video hybrid recommendation system.Based on the requirement analysis of the short video recommendation system,a short video hybrid recommendation system based on user preferences was implemented.The test results show that the system has achieved the expected results in terms of both functionality and performance.This thesis uses the proposed dynamic weighted hybrid recommendation algorithm to complete the design and implementation of a hybrid recommendation system for short videos,which not only meets the user’s demand for recommendation system functions,but also provides a solution to the problems of cold start,data sparsity and real-time.