| With the rapid development of Internet technology,all kinds of music platforms provide people with a large number of songs.However,faced with a large amount of song information,it is difficult for people to find the music they really like just by searching.The personalized recommendation system can screen out music that meets personal preferences for users from a large amount of music information,and improve user satisfaction.Therefore,the personalized recommendation of music has become the research direction of the industry.Aiming at the problems of cold start of items and data sparsity in collaborative filtering algorithm,this thesis proposes an improved collaborative filtering recommendation algorithm based on music tags.This algorithm combines music tags with the traditional collaborative filtering algorithm,calculates the user’s preference for unknown music,and solves the problems of cold start of items and data sparsity.On the basis of solving the problem of data sparsity,the similarity between music is calculated to predict the user’s music preference.Aiming at the real-time problem of collaborative filtering algorithm,a system architecture combining offline processing and online processing is designed.This architecture uses the improved collaborative filtering algorithm in Hadoop to process the historical behavior data of users offline,and selects popular music according to the number of plays to recommend new users.In addition,the real-time behavior of users is simulated through the method of sliding window processing of user behavior data.These behavioral data are processed online in Storm to provide real-time recommendation results for users.This processing method improves the accuracy and real-time performance of the recommendation system.In this thesis,a real-time music recommendation system is constructed according to the real-time requirement of the recommendation system.Experiments on the Last.fm dataset have verified that in music systems based on Hadoop and Storm,the improved collaborative filtering recommendation algorithm based on music tags is more accurate than the traditional collaborative filtering algorithm;the real-time recommendation results generated by Hadoop and Storm processing,its accuracy is higher than the accuracy of offline recommendation results generated by only Hadoop processing. |