Research On Personalized Music Recommendation System Based On Improved Neural Collaborative Filtering

Collaborative filtering technology has been widely used in various recommended tasks and has achieved great success in practice.However,with the rapid development of science and technology in the post-information age,serious information overload and the diversification of data structure are emerging,so collaborative filtering and recommendation methods are facing great challenges.Nowadays,it is impossible to independently rely on collaborative filtering or other recommendation models to complete the recommendation task.Recent years,deep learning method with its powerful data processing ability,feature learning and expression ability,and other advantages quickly occupies a place in the field of recommender system.Almost all industry recommendation system models have applied deep learning technology.Slightly different from most current music recommendation models extracting features from music metadata(timbre,tone,etc.),this paper attempts to extract song content features from the macro attributes of music(genre,artist,etc.).The Word2vec approach yields dense low-dimensional hidden eigenvectors trained by the original high-dimensional sparse eigenvectors,and incorporating classical collaborative filtering,matrix decomposition and content recommendation methods into the framework of deep learning techniques.This paper proposes an improved music recommendation model for neural collaborative filtering to provide personalized song recommendation for each user.In the model structure,on the one hand,all features are grouped and processed to obtain comprehensive group features to enhance model expression ability;on the other hand,interactive features are added to the model through vector outer product to learn more nonlinear information.Experiments were conducted on two music data sets.The experimental results show that the F1 and mAP evaluation index values of the model in this paper are far higher than those of other models,so the recommendation quality is significantly better than other models.In addition,due to the introduction of user information and song attributes,the cold start problem is alleviated,and the use of the Word2vec method also effectively alleviates the data sparsity problem,and the use of deep learning methods greatly improves the generalization ability and robustness of the model.Model versatility and generalization ability need to be further verified by data sets in other different fields,which will be part of future research work.