Multi-Factor Fusion Personalized Movie Recommendation Algorithm Based On Edge Computing

The explosive growth of web data brings new challenges to the design of recommendation algorithms.If there is no timely dynamic perception of changes in user preferences for personalized recommendation,the user experience will be greatly reduced.Recommendation algorithm has been widely used in many fields and has brought great commercial value,especially in the field of movie recommendation.However,with the increase of user scale,the existing recommendation algorithms have problems such as cold start,sparse data,and difficulty in mining the potential feature information of user-item association.Therefore,designing effective movie recommendation algorithm has important theoretical significance and application reference value.The development status of recommendation algorithm at home and abroad is analyzed and summarized in detail.The recommendation algorithms are divided into five categories: collaborative filtering recommendation algorithm,hybrid recommendation algorithm,matrix decomposition recommendation algorithm,machine learning recommendation algorithm and edge computing recommendation algorithm.The core ideas of their representative algorithms are described,makes a depth comparative analysis,and summarizes the advantages and disadvantages,applicable scenarios and existing challenges of various recommended algorithms.For existing movie recommendation algorithms that only consider factors such as rating,label,timestamp or single context information,and do not take into account users’ preferences for different types of movies of different ages,a multi-factor fusion personalized movie recommendation algorithm based on edge computing was proposed.F-SVD algorithm comprehensively considers many factors such as similar users,user rating frequency and user preference for different types of movies in different eras.In the edge server,a new user similarity calculation method F-Pearson was proposed.The user similarity scores are sorted and input into F-SVD algorithm.The F-SVD algorithm uses the rating information of multiple similar users as the benchmark of target user prediction,which can solve the problem of single source data caused by the traditional recommendation algorithm using the rating information of single user.In the cloud server,the Bidirectional Encoder Representations from Transformers model is used to train historical data to solve the difficulty in mining potential feature information.The ratings of predicted users for different types of movies in different eras are input into the F-SVD algorithm.By effectively integrating multiple factors,F-SVD algorithm can mine the potential correlation features between users,users and items,and items in the complex and diverse mass data,so as to improve the accuracy of prediction.In order to verify the effectiveness of the proposed method,different recommendation algorithms were tested on Movielens-Small dataset and Movielens-10 M dataset.Root Mean Squared Error(RMSE),Mean Absolute Error(MAE),PRECISION and RECALL were used as evaluation indexes.Experimental results show that compared with the traditional recommendation algorithm,the proposed algorithm has lower error,prediction accuracy is 89.0%,recall rate is 79.1%,delay is reduced by 9.8%,accuracy is improved by 2.3%.