The Design And Prototype Implementation Of Tag-based Personalized Music Recommendation System

The Internet era provides a variety of network services for people,music service is one of them.Major music sites provide a large number of songs to meet people's music needs.With the pursuit of personalized 90 s to become the main consumer of the Internet,the personalization of music services will open up a broader market.There are millions of music in various genres at home and abroad,but there is also a serious problem between users and music — information asymmetry.As a branch of the information filtering system,the recommendation system can predict the user's preferences,boost the net flow and stimulate consumption.The personalized music recommendation system can effectively explore the long tail and provide music recommendation list for users.The content-based recommendation algorithm and collaborative filtering-based algorithm are two exmaples of music recommendation system algorithm.Both algorithm have advantages and disadvantages,which provides good ideas to design the algorithm of this thesis.This thesis designs a tag-based personalized music recommendation system in order to provide personalized music playlists for users.First of all,it summarizes the domestic and international development of music recommendation and analyses the existing recommendation algorithms.Next it designs a collaborative filtering algorithm based on directed tags,using several steps to improve the tag-based collaborative filtering algorithm,and it verifies that the algorithm has good performance through contrast experiments.Then designs the functional modules of the personalized music recommendation system according to the demand analysis.The main research of this paper are as follows:(1)Using several steps to improve the tag-based collaborative filtering algorithm,and the cognitive order is added to the ternary relationship between users,music and tags.The user interest feature directed graph and the music feature directed graph are formed according to the sequential relationship between them,which can reflect the users' classification preference,thus more accurately simulating users' interest,and make the recommendation result more consistent with the personalized needs of users.(2)The music feature directed graphs is divided into several disparate clusters according to certain rules which ensures that the feature directed graphs between clusters are different,meanwhile the musical feature directed graphs in each cluster are similar.When matching the similarity of them,we can just query in the cluster which has the highest similarity with the target to match a sufficient amount of applicable music.This step improves the efficiency of matching user interest and music.(3)This thesis designs a personalized music recommendation system,including the basic functions and recommendation function.The system can crawl the data and accomplish pre-processing of data such as de-duplication,word segmentation,keyword extraction,etc.The recommendation engine contains two algorithms—the DTSCF algorithm and the recommendation algorithm based on LDA-MURE.The system can select the appropriate algorithm according to the actual situation of data to provide a continuous personalized music playlist for user,which satisfies the users' demands for music.