Research On Personalized Music Recommendation Method Based On Users' Preferences

With the rapid development of Internet technology,the number of items available for users on the Internet has increased exponentially,which has made it difficult for users to select items that they are really interested in,known as information overload problem.The recommender system has emerged as the times require.The recommender system models the users' preferences and provides the personalized recommendation services to users.Music recommendation aims to mine useful information from the interactive data between users and music,and provides personalized music recommendations for users according to their interests,behaviors,social relations and other factors.However,most of the existing music recommendation methods only consider users' preferences at a simple level,which lacks the fully mining of the preferences.In the existing music recommendation,most of the music recommendation algorithms based on collaborative filtering directly use the frequency of use's listening history as the user's ratings of the tracks,ignoring the distribution between these frequencies.In addition,the music attributes,such as melody,rhythm timber and so on,are difficult to extract and process.This results in that the essential characteristics of music can't be effectively applied in music recommendation,and can't provide users with a satisfactory personalized music recommendation.In order to solve the above problems,this paper focuses on the impact of adding tags on user feature extraction.On this basis,a personalized music recommendation method that combines different features of user's preference is proposed.The main research contributes of this paper are as follows:(1)Considering the distribution of the frequency of the user's listening history,a reasonable rating mechanism is established to convert the frequencies into a certain range of scores.In addition,for the problem that the characteristics of music itself are difficult to extract and process,this article uses tags instead of the content features of the music itself.Based on the established scoring mechanism and the different weights of the tags,user's preferences vectors of the user's music types are established and the users' similarity matrix is constructed.Finally,the music playlist is recommended to the target user based on several nearest neighbors' weighted average score of the track.(2)The traditional music recommendation systems rarely consider the multiple aspects of the use's preference,which leading to a sub-optimal recommendation.This paper proposes a novel music recommendation approach,which exploits three computational attributes to describe the user's music taste: novelty,diversity,and mainstream.Based on these three attributes,a virtual friendship of each user is constructed by using an extended random walk algorithm.Finally,more personalized recommendations can be generated by exploiting the music preferences of the target user and his virtual friends.(3)Through extensive experiments,the proposed methods and other related methods are compared and analyzed.The experimental results prove the superiority of the proposed methods.