Research And Implementaion Of Personalized Music Recommendation Algorithm

In the Internet era,all kinds of music websites provide thousands of songs for people,which can meet people's needs and bring great convenience.However,the rapid development of digital music has caused overload of music information.Facing with massive music,it becomes difficult for people to find the music with their own interests quickly.Music recommendation system can recommend music which users may like,and help users to find the songs they want quickly.This kind of recommendation service can bring users good user experience,and bring business interests,therefore,music recommendation has become the industry and scholars'important research direction.Collaborative filtering algorithm is widely used in recommendation systems of various fields,but there are problems of cold start and poor extendibility;traditional recommendation algorithms did not consider the specific continuity of music listening behavior and environmental influence,causing the recommendation result not ideal.Aiming at these problems,in this paper,we conducted thorough research on music recommendation algorithms,improved the collaborative filtering algorithm and realized it by parallel implementation based on big data analysis platform;at the same time,based on the characteristics of music listening,we realized a kind of music playlist recommendation algorithm.First of all,this paper researched on collaborative filtering algorithm,and improved the performance of the ALS model method.We put forward collaborative filtering algorithm based user groups that users can be clustered into groups according to users' attributes before training ALS models.At the same time,we researched on distributed computing technology,and realized parallel design 'and implementation of the algorithm based on Spark platform.On Spark platform,clustered collaborative filtering algorithm based user groups improved the accuracy and efficiency of recommendation,and solved the problem of users' cold start.Then,in this paper,we realized a kind of music playlist recommendation algorithm based on space embedding model,considering that music listening has the characteristics of continuity and relevance with context.By building user-song-label space model,the probability relations among users,songs and labels can be mapped as space distances,the algorithm realized personalized and continuous music recommendation,and meet users' current demands of music.At last,based on the above two kinds of recommendation technology,we designed a multi-function music recommendation system in this paper.In view of the practical demands and based on the distributed platforms,the system realized music recommendation according to users' long-term preference prediction and continuous music playlist recommendation according to demands and preferences of the current session.