The coming of the digital revolution has brought the network era, which has caused big changes to the music spread situation and enlarged both the range and degree of music propagation. The present research paper proposes one hybrid music recommendation method which is integrated inside one autoencoder. This content in the basic sense can be divided into two parts: one is the analysis of data content, and the other is experiments about the recommendation method. Firstly, a music feature extraction scheme applicable to music recommendation scenarios is designed; secondly, a music hybrid recommendation model is constructed to realize music track recommendation for different users by incorporating the autoencoder structure. Finally, the Jitterbug popular BGM features are extracted from the time domain and frequency domain in the Jitterbug popular BGM audio dataset as well as in the GTZAN public dataset, respectively, to analyze the distribution law, and the model of this paper is compared with other models in the dataset for comparison experiments. The results show that Jitterbit popular BGMs have fundamental tone periodicity, i.e., they maintain a low frequency and energy overall. And this paper’s model has the optimal model efficacy with 9.06% and 4.11% improvement in Recall metrics and 4.67% and 2.28% improvement in F1 metrics on the two datasets, and it also has the optimal anti-sparse ability. The model performs well in the accuracy of music recommendation and can provide technical support for the digital music dissemination path.
