Music Recommendation System Based On Knowledge Graph Enhanced Graph Neural Network And Sentiment Analysis

With the rapid development of music streaming platforms,music recommendation systems play an important role in improving user experience.However,the accuracy of the recommendation results of the existing music recommendation models needs to be improved due to the exposure bias of the training samples,and the recommendation often only considers the user’s historical interaction and attribute information,ignoring the impact of user emotional changes on the recommendation results.Therefore,this paper proposes a music recommendation system based on knowledge graph enhanced graph neural network and sentiment analysis to achieve more accurate and personalized recommendation services.Aiming at the problem of exposure bias in the training samples of the recommendation model,this paper proposes a recommendation model KGIFE based on the knowledge graph enhanced graph neural network.This model adds indirect feedback and item relationship enhancement to improve the recommendation performance on the basis of the original KGIN model,and introduces indirect feedback.It can improve the probability of being recommended for items that have not been interacted with by users,reduce sample exposure bias,and introduce item relationship enhancement to perform similarity fusion of item features in the model and improve item embedding and representation capabilities.Then,the effectiveness of the model is verified by comparison experiments with KGCN,KGAT,KGIN and other models.Aiming at the problem that user emotion changes will affect the recommendation results during music recommendation,this paper constructs a multi-modal fusion hierarchical emotion analysis model HMAMF.The model is divided into a main model and an auxiliary model.The main model combines multi-modal features and performs fusion prediction through the Embrace Net network.The auxiliary model assists the main model training through single-modal decision-making,and extracts single-modal features into the main model to improve the feature extraction ability and analysis and prediction performance of the main model.Then the performance and effect of the model are verified by comparative experiments.Finally,a music recommendation system is designed based on the KGIFE and HMAMF models proposed in this paper.The recommendation model and the sentiment analysis model are combined to recommend music that matches the user’s emotion,improving the accuracy of recommendation and user satisfaction.