Research On Music Recommendation Method Considering Interest And Friend Relationship

At present,the music industry is developing rapidly,and more and more music platforms and applications are used.Music has become an important part of people’s life.Users have produced a lot of operation records on the music platform.How to effectively use these data to recommend music that meets their interest preferences has become an urgent problem for each music platform.Many music recommendation algorithms only consider the music listening records of users,but not the influence of user’s friend relationship and user’s interest over time.Therefore,based on the user friend relationship network and the user interest network of music characteristics,a music recommendation method considering interest friend relationship is proposed in this thesis,and the specific contents are as follows:(1)The thesis proposes a music recommendation method considering the friend relationship among users.Through building the user friend relationship network and using Louvain algorithm,the user friend relationship network is divided into some friend relationship communities,and recommend music to users in the community.In a community of friend relationship,the user song scoring mechanism can be established according to the frequency of users’ listening songs,and the user correlation coefficient can be obtained by Pearson similarity.The user influence is calculated in the friend relationship community,and the user contribution degree in the social area of the friend relationship is obtained by combining the user contribution degree with the user correlation coefficient for deciding the User similarity.According to the music records of K users,the based-user collaborative filtering algorithm is then used to recommend music to users.(2)This thesis proposes a music recommendation method considering both interest and friend relationship.Considering that user interest will change with time and affect the effect of music recommendation,time attenuation function is introduced to modify user contribution degree,and user similarity is judged by combining user contribution degree with user correlation coefficient in the friend relationship,and then music recommendation is launched.Then the user interest relationship network is partitioned into some communities of user interest relationship by Louvain-COPRA method.In the interest relationship community,time decay function is introduced to modify the user song scoring mechanism,and then the modified user correlation coefficient is obtained;The user contribution is calculated in the interest relationship community,which is combined with the user contribution in the friend relationship community and the modified user correlation coefficient to get the user similarity,and the user based collaborative filtering recommendation algorithm is used to recommend music to users.(3)Three evaluation indexes,including accuracy,recall rate and F1-Score,are selected to assessed the music recommendation model and three traditional recommendation algorithms are applied on last.fm datasets in this thesis and results show that the proposed model performs better results than other three models,and it also helps to alleviate problems of data sparsity and cold start.