A Personalized Recommendation Method for Vocal Music Teaching Resources Based on Network Topology

It is impossible to recommend music teaching resources to users because current music recommendation methods only mine the general relationship between users and songs, and cannot distinguish the different preferences of different users for the same song. As a result, this paper proposes a network topology-based personalized recommendation method for vocal music teaching resources that uses neural network topology information and introduces a multi-layer attention mechanism. Using user attributes and song content information, this paper learns song representations from multiple dimensions, as well as mines the preference relationship between users and songs. A user feature-dependent attention network is designed in order to distinguish different users’ preferences for multi-domain features of songs. A song-dependent attention network can be designed by mining the time-series dependencies of user behaviors and distinguishing them from one another. A user’s preference distribution for candidate songs can then be calculated using Softmax, which generates recommendations. These dataset’s experimental results show that the model proposed in this paper is significantly better than the current recommendation model in multiple indicators.