Research On Personalized Music Recommendation Method Based On Deep Learning

With the development of the Internet,digital music compared to traditional physical music has been greater and faster development.In the face of increasing mass music database,relying on traditional search methods to find their own interest in music has become increasingly unable to meet the needs of users.In this case,the recommendation system is introduced to the field of digital music gradually.Relying on the recommended ability of the recommendation system,users do not need to actively search will be able to get the music they wanted.But with the continuous improvement of user needs,the traditional recommendation of digital music has been unable to meet people's needs.In the process also produced a series of problems to be solved.For example,due to the lack of user data,the newly added music set can not be effectively recommended,this called cold start problem.And then,for example,in the recommended results,the user received the recommendation set focused on the same type of music,making it difficult to find new interest of user.In view of the above problems,this paper proposes a personalized music recommendation method based on deep learning.Firstly,this paper designs the music information forecasting model based on the music floor feature set and the deep belief network.Based on the research on the music classification of different dimensions,four kinds of underlying features of music are selected.Every music is composed of 40-dimension eigenvectors,and then the music feature set is constructed.Cause of good performance in audio field,deep belief network is used as the music information prediction model.And its structure and parameters are redesigned.The multi-dimensional information prediction of music is completed.The method can be a good supplement to the new music information to solve the cold start problem;Secondly,this paper on the basis of improved Apriori algorithm to complete the overall recommendation of the design work.According to the information mining of the candidate set to find in the associated music and join the recommended candidate.Such music,although doesn't match the user's interest,it has a strong relevance.In considering the recommended accuracy on the basis of improving the recommended results of diversity.With this method,we can find the user's new interest.While the improved Apriori algorithm in the efficiency and accuracy has been greatly improved.Experiments show that this method can complete the music recommendation work.