Context-Aware Music Recommendation Based On Latent Factor Model

Recommender system has special advantages in solving the problem of"Information Overload" and providing personalized services.In recommender system,latent factor model is a popular recommendation algorithm widely used in many fields.However,due to its short development time,it still has the problem of low recommendation accuracy.In order to solve the problem of low accuracy of traditional latent factor model recommendation algorithm applied to music streaming media platform,this paper makes full use of the text data in the comments section of Netease cloud music platform where users express their personal feelings,and improves the traditional latent factor model by mining the user emotion characteristic factors,aiming to enhance the performance of user score prediction and improve the accuracy of recommendation algorithm.This paper proposes an emotion latent factor model(EmotionLFM)which integrates user’s emotional features.The user’s emotional feature factors are obtained from the emotional analysis of songs scored by the BERT-XGBoost combination model.The BERT-XGBoost combination model uses the neural network framework Pytorch in Python language to call the best pretraining model to extract the deep features of the comment text,and then uses the XGBoost algorithm library to analyze the emotional tendency.EmotionLFM model uses the comment sentiment score obtained from the BERT-XGBoost combination model to construct the user sentiment correction coefficient vector,and then modifies the user hidden category factor preference matrix P in the latent factor model according to the vector,so as to improve the score prediction performance of the latent factor model.In this paper,the user and song information of Netease cloud music platform are selected as the experimental data.Firstly,the popular song comment text crawled based on Scrapy framework and the user’s implicit behavior log data are processed into the experimental data set of the model.Then,the improved latent factor model recommendation algorithm EmotionLFM is compared with the traditional latent factor model recommendation algorithm by using the experimental data set on Spark distributed platform.The experimental results show that the song recommendation algorithm based on EmotionLFM model improves the recommendation effect and the recommendation accuracy of the music recommendation system.