Collaborative Filtering Model for Chinese Music Genre Recommendation System

Luo Wenhao, Azuraliza Abu Bakar
dx.doi.org/10.17576/apjitm-2025-1402-14

Abstract

The rapidly increasing amount of information can meet the diverse needs of users, but it also faces the challenge of data screening. In the face of massive music resources, users often can?t make quick and appropriate choices. Therefore, music recommendation system has become an effective solution tool in this context and has been applied by many large streaming media platforms. However, the current mature music recommendation system still has challenges including lack of personalized recommendation, cold start and data sparsity problems. This article aims to achieve high-precision personalized Chinese music recommendation by using a hybrid scheme of content-based and collaborative filtering algorithms. Our specific objectives are three folds, (i) to propose features and score about Chinese music items to achieve personalized recommendation. (ii) to propose a content-based method to solve the cold-start problem. (iii) to formulate an efficient similarity computation to overcome the data sparsity problem. Four machine learning algorithms were employed: K-nearest neighbors algorithm, Singular Value Decomposition, Latent Factor Model, and Non-negative Matrix Factorization. Experimental results show that the hybrid solution combining content-based and KNN algorithms performs best in addressing cold start, personalization, and data sparsity problems.

keyword

Chinese music recommendation, hybrid collaborative filtering model, genre base

Area

Pattern Recognition