For the problem of how to query,retrieve,and organize music information quickly and efficiently,the performance of the music retrieval system can be improved through automatic music annotation technology.In this study,a multi-label music automatic annotation system based on generative adversarial networks(GANs) is proposed.The LDA model is used to cluster the music tags to obtain thematic categories,and then the mapping relationship between the audio features and the semantic features of the music is found by the generative adversarial network.For experimental verification,when the method proposed in this paper was applied to the CAL500 dataset in five cross-validation experiments,the comprehensive performance index of the method was greatly improved compared with existing methods.