A Graph Attention-Based Decision-Support Framework for Music Genre Classification and Evolution Analysis
Keywords:
Graph Attention Networks; Music Genre Classification; Self-Supervised Contrastive Learning; Multi-View Representation; Temporal Evolution Mining; Domain Adaptation; Music Information RetrievalAbstract
Music genres develop through intricate patterns of historical transformation, stylistic interaction and mutual influence, yet conventional classification approaches largely disregard these relationships by representing genres as independent, temporally invariant categories. To overcome this limitation, this study introduces the Temporal Multi-view Graph Attention Network (TMGAT), a unified framework for music genre recognition and the identification of evolutionary relationships within large-scale audio collections. The proposed architecture integrates three distinct yet complementary representations: manually engineered acoustic features encompassing Mel-frequency cepstral coefficients (MFCC), chroma and rhythmic descriptors; a frozen pretrained convolutional neural network, PANNs CNN14; and a frozen self-supervised audio representation model, wav2vec 2.0. These representations are consolidated through an adaptive, learnable view-attention mechanism. Before incorporating labelled observations, the framework employs self-supervised contrastive pretraining, combining InfoNCE with SimCLR-inspired data augmentation to establish a coherent shared embedding space. Subsequently, a k-nearest-neighbour similarity graph is constructed to capture relationships among musical tracks, with feature propagation performed through a three-layer, multi-head graph attention encoder incorporating eight attention heads per layer, residual connections and DropEdge regularisation. Genre evolution is further characterised by integrating Fourier-based temporal encoding with a cross-decade attention mechanism, while a gradient reversal layer supports adversarial domain adaptation across datasets. Experimental evaluation using the FMA-large subset (approximately 25,000 tracks spanning 16 genres from 1950–2022), MTG-Jamendo (~55,000 tracks) and GTZAN for cross-corpus validation demonstrates the effectiveness of the proposed approach. TMGAT achieves classification accuracies of 82.5% and 79.6%, alongside macro-F1 scores of 0.821 and 0.787, on FMA-large and MTG-Jamendo, respectively. Under zero-shot cross-corpus transfer, it attains 85.7% accuracy on GTZAN. These outcomes exceed those of eight competitive baseline methods, including PANNs, wav2vec 2.0, MusicNN, CRNN and GAT, with statistically significant improvements confirmed through paired t-tests (p < 0.01). Furthermore, community detection applied to the temporal transition matrix reveals established patterns of stylistic convergence, including Rock–Pop integration, Jazz–Blues associations and the increasing prominence of Hip-Hop–Electronic hybrids after 2000. Collectively, these findings establish TMGAT as an interpretable, graph-based framework capable of examining both genre identity and temporal stylistic development, extending analytical capabilities beyond those available through conventional flat classification models.
Downloads
References
[1] Baevski, A. Z., H.; Mohamed, A.; Auli, M. (2020). wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations. NeurIPS. https://doi.org/10.48550/arXiv.2006.11477
[2] Barakat, A., & Bianchi, P. (2021). Convergence and dynamical behavior of the ADAM algorithm for nonconvex stochastic optimization. SIAM Journal on Optimization, 31(1), 244-274. https://doi.org/10.1137/19M1263443
[3] Bevec, M., Tkalčič, M., & Pesek, M. (2024). Hybrid music recommendation with graph neural networks. User Modeling and User-Adapted Interaction, 34(5), 1891-1928. https://doi.org/10.1007/s11257-024-09410-4
[4] Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 2008(10), P10008. https://doi.org/10.1088/1742-5468/2008/10/P10008
[5] Bogdanov, D., Won, M., Tovstogan, P., Porter, A., & Serra, X. (2019). The mtg-jamendo dataset for automatic music tagging. In. ICML. https://mtg.github.io/mtg-jamendo-dataset/
[6] Chen, T., Kornblith, S., Norouzi, M., & Hinton, G. (2020). A Simple Framework for Contrastive Learning of Visual Representations. https://doi.org/10.48550/arXiv.2002.05709
[7] Choi, K. F., G.; Sandler, M.; Cho, K. . (2017). Convolutional Recurrent Neural Networks for Music Classification. ICASSP. https://doi.org/10.1109/ICASSP.2017.7952585
[8] Defferrard, M. B., K.; Vandergheynst, P.; Bresson, X. (2017). FMA: A Dataset for Music Analysis. ISMIR. https://doi.org/10.48550/arXiv.1612.01840
[9] Fu, W. S., K.; Xu, K.; Zhang, Q. C. . (2026). CryoLVM: Self-supervised Learning from Cryo-EM Density Maps with Large Vision Models. https://doi.org/10.48550/arXiv.2602.02620
[10] Ganin, Y. U., E.; Ajakan, H.; Germain, P.; Larochelle, H.; Laviolette, F.; Marchand, M.; Lempitsky, V. (2016). Domain-Adversarial Training of Neural Networks. Journal of Machine Learning Research 17(1-35). https://doi.org/10.48550/arXiv.1505.07818
[11] Grover, A., & Leskovec, J. (2016). node2vec: Scalable Feature Learning for Networks Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, California, USA. https://doi.org/10.1145/2939672.2939754
[12] Hamilton, M., & Pearce, M. (2024). Trajectories and revolutions in popular melody based on U.S. charts from 1950 to 2023. Scientific Reports, 14(1), 14749. https://doi.org/10.1038/s41598-024-64571-x
[13] Han, J., Huang, W., Rong, Y., Xu, T., Sun, F., & Huang, J. (2023). Structure-aware dropedge toward deep graph convolutional networks. IEEE Transactions on Neural Networks and Learning Systems, 35(11), 15565-15577. https://doi.org/10.1109/TNNLS.2023.3288484
[14] Hao, Y., Jing, X.-Y., Chen, R., & Liu, W. (2023). Learning enhanced specific representations for multi-view feature learning. Knowledge-Based Systems, 272, 110590. https://doi.org/10.1016/j.knosys.2023.110590
[15] Hershey, S., Chaudhuri, S., Ellis, D. P., Gemmeke, J. F., Jansen, A., Moore, R. C., Plakal, M., Platt, D., Saurous, R. A., & Seybold, B. (2017). CNN architectures for large-scale audio classification. In 2017 ieee international conference on acoustics, speech and signal processing (icassp) (pp. 131-135). IEEE. https://doi.org/10.1109/ICASSP.2017.7952132
[16] Kipf, T. N. W., M. (2017). Semi-Supervised Classification with Graph Convolutional Networks. ICLR. https://doi.org/10.48550/arXiv.1609.02907
[17] Kong, Q. C., Y.; Iqbal, T.; Wang, Y.; Wang, W.; Plumbley, M. D. (2020). PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition. IEEE/ACM TASLP, 28, 2880–2894. https://doi.org/10.1109/TASLP.2020.3030497
[18] Liang, D., Yang, F., Zhang, T., & Yang, P. (2018). Understanding mixup training methods. Ieee Access, 6, 58774-58783. https://doi.org/10.1109/ACCESS.2018.2872698
[19] Liu, Q., & Liu, X. (2022). An improved multi-view attention network inspired by coupled P system for node classification. PLoS ONE, 17(4), e0267565. https://doi.org/10.1371/journal.pone.0267565
[20] Logan, B. (2000). Mel Frequency Cepstral Coefficients for Music Modeling. ISMIR. https://ismir2000.ismir.net/papers/logan_paper.pdf
[21] Loshchilov, I. H., F. (2019). Decoupled Weight Decay Regularization (AdamW). ICLR. https://doi.org/10.48550/arXiv.1711.05101
[22] Nowak, K., & Zieliński, T. (2025). Graph Neural Network for Music Style Classification. Frontiers in Artificial Intelligence Research, 2(1), 26-34. https://doi.org/10.71465/fair142
[23] Ostermann, F., Vatolkin, I., & Ebeling, M. (2023). AAM: a dataset of Artificial Audio Multitracks for diverse music information retrieval tasks. EURASIP Journal on Audio, Speech, and Music Processing, 2023(1), 13. https://doi.org/10.1186/s13636-023-00278-7
[24] Park, D. S. C., W.; Zhang, Y.; Chiu, C.-C.; Zoph, B.; Cubuk, E. D.; Le, Q. V. (2019). SpecAugment. Interspeech. https://doi.org/10.21437/Interspeech.2019-2680
[25] Pons, J. S., X. (2019). MusicNN: Pre-trained convolutional neural networks for music audio tagging. arXiv. https://doi.org/10.48550/arXiv.1909.06654
[26] Rossi, E. C., B.; Frasca, F.; Eynard, D.; Monti, F.; Bronstein, M. (2020). Temporal Graph Networks for Deep Learning on Dynamic Graphs. ICML Workshop. https://doi.org/10.48550/arXiv.2006.10637
[27] Sumbodo, W., Pambudi, R., & Setiadi, R. (2025). Design and Analysis of Pick and Place Robot Gripper in the Fertilizer Industry. Reports in Mechanical Engineering, 6(2), 53-64. https://doi.org/10.31181/rme511
[28] Tan, A., Wang, Y., Wu, W.-Z., Ding, W., & Liang, J. (2025). Multi-View Fusion Graph Attention Network for Multilabel Class Incremental Learning. Information fusion, 123, 103309. https://doi.org/10.1016/j.inffus.2025.103309
[29] Tzanetakis, G. C., P. . (2002). (2002). Musical genre classification of audio signals. IEEE Transactions on Speech and Audio Processing, 10(5), 293-302. https://doi.org/10.1109/TSA.2002.800560
[30] van den Oord, A. L., Y.; Vinyals, O. (2018). Representation Learning with Contrastive Predictive Coding. arXiv. https://doi.org/10.48550/arXiv.1807.03748
[31] van der Maaten, L. H., G. (2008). Visualizing Data using t-SNE. JMLR, 9, 2579–2605. https://www.jmlr.org/papers/v9/vandermaaten08a.html
[32] Vaswani, A. S., N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, Ł.; Polosukhin, I. (2017). Attention Is All You Need NeurIPS. https://doi.org/10.48550/arXiv.1706.03762
[33] Veličković, P., Cucurull, G., Casanova, A., Romero, A., & Lio, P. (2017). Graph Attention Networks. https://doi.org/10.48550/arXiv.1710.10903
[34] Wang, W., Wang, T., & Cai, Y. (2022). Multi-view attention-convolution pooling network for 3D point cloud classification. Applied Intelligence, 52(13), 14787-14798. https://doi.org/10.1007/s10489-021-02840-2
[35] Wang, Z. M., M.; Xiao, X.; Liu, C.; Ma, C.; Zhang, Y.; Wang, X.; Krishnaswamy, S.; Wang, T. (2026). CTR-LoRA: Curvature-aware and Trust-Region Guided Low-Rank Adaptation for Large Language Models. ICASSP, 1396-1400. https://doi.org/10.1109/ICASSP.2026.CTRLORA
[36] Xiao, X. W., Z.; Mo, M.; Liu, C.; Ma, C.; Li, Y.; Krishnaswamy, S.; Wang, X.; Wang, T. (2026). Self-supervised visual prompting for cross-domain road damage detection. . WACV, 3514–3524. https://openaccess.thecvf.com/WACV2026
[37] Yang, L., Zhang, L., & Yang, W. (2021). Graph adversarial self-supervised learning. Advances in neural information processing systems, 34. https://proceedings.neurips.cc/paper_files/paper/2021/hash/7d3010c11d08cf990b7614d2c2ca9098-Abstract.html
[38] Yao, H.-Y., Zhang, C.-Y., Yao, Z.-L., Chen, C. P., & Hu, J. (2024). A recurrent graph neural network for inductive representation learning on dynamic graphs. Pattern Recognition, 154, 110577. https://doi.org/10.1016/j.patcog.2024.110577
[39] Zhang, Y., & Li, T. (2025). Music genre classification with parallel convolutional neural networks and capuchin search algorithm. Scientific Reports, 15(1), 9580. https://doi.org/10.1038/s41598-025-90619-7
[40] Zhao, P., Zang, W., Liu, B., Kang, Z., Bai, K., Huang, K., & Xu, Z. (2021). Domain adaptation with feature and label adversarial networks. Neurocomputing, 439, 294-301. https://doi.org/10.1016/j.neucom.2021.01.062
[41] Zhou, Y., Zheng, Q., Bai, S., & Zhu, J. (2023). Semantically consistent multi-view representation learning. Knowledge-Based Systems, 278, 110899. https://doi.org/10.1016/j.knosys.2023.110899
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Decision Making: Applications in Management and Engineering
This work is licensed under a Creative Commons Attribution 4.0 International License.
