Research On A Personalized Music Recommender System Based On SVD++ And Context-Aware Technology

In the era of information overload,the music which is closely related to people's lives has flourished,meanwhile,the musical resources have got very enriching.But people find it increasingly difficult to find music that meets their own tastes,which causes an increasingly urgent need of the personalized music recommendation.While the personalized music recommender systems have been a hot concern of many researchers,the development of the personalized music recommendation is still facing many challenges and problems which continually need further research and optimal solution.In this paper,the research mainly focuses on how to enhance the effectiveness of the personalized music recommendation system,to solve the issues of the accuracy,real-time feature and diversity of the recommendation.The main contents and contributions of our work are summarized as following:(1)In this paper,a framework of a music recommendation algorithm based on the SVD++ model and context-aware technology is proposed.It takes the advantage of the SVD++ model in rating forecasting,while using the context-aware technology to introduce diverse contextual information,such as time,location,and mood,into the recommending process to improve its effectiveness.It also takes full account of the effects of the users' historic rating behaviors,recent listening behaviors and instant feedbacking behaviors,and effectively fuses multiple sorting methods together during the recommending process to generate the Top-N recommendation list.The results of the experiments show that the hybrid music recommendation algorithm have significant effects on decreasing the error of forecasting,improving the accuracy of rating forecasting and easing the sparsity problem of the rating matrixes.(2)In this paper,a hybrid music recommendation algorithm is proposed with the purpose of performing the fusion of diversity in generating the intermediate recommendation result.Besides that,the algorithm fuses three kinds of sorting methods together to generate the final recommendation list,while taking account of the differences between the recommended songs and those negative feedbacking songs to balance the recommendation result.The results of the experiments show that the hybrid music recommendation algorithm can improve the diversity of the recommendation result effectively.(3)For the issue of the real-time feature in the music recommender systems,a method combining the offline calculation with the online processing to generate recommendations is proposed in this paper.By accomplishing most of those complex calculational tasks in the offline calculation module,the amount of the calculation in the online processing module is decreased significantly,thus reducing its response time.Meanwhile the system will adjust the recommendation list immediately according to the users' positive or negative feedback behavior,in order to implement the online real-time recommendation and improve its real-time feature effectively.