Scholarlink

<p>Music Emotion Recognition (MER) is vital for understanding emotional responses to music and advancing aesthetic education, enhancing artistic sensitivity and cognitive growth. However, current MER methods frequently fail to effectively model cross-modal emotional dependencies and the dynamic variations in expressive cues. This research proposes a multimodal deep learning (DL) framework that integrates the Multimodal Music Emotion dataset including 14 audios file (70 records), textual data (2000 records), and visual elements (2000 records) modalities for accurate MER and comprehensive evaluation of aesthetic education effectiveness. Preprocessing audio signals includes noise reduction, and features are extracted using Mel-Frequency Cepstral Coefficients (MFCCs) to be acoustic and timbral structures. Textual lyrics are preprocessed through tokenization, and word embeddings capture semantic and emotional context in features. Visual data is resized, and features Visual sentiment is interpreted using a convolutional neural network (CNN) that extracts face expression cues. A feature-level fusion approach combined with an attention mechanism highlights emotionally dominant attributes across modalities. The proposed Adaptive Golden Jackal-driven Dynamic Graph Neural Network (AdapGJ-DGNNet) dynamically constructs graph relationships among multimodal feature nodes and adaptively updates edge weights using the AdapGJ strategy, enabling optimal graph connectivity and robust temporal-spatial emotion modeling. The DGNNet structure facilitates learning non-linear correlations between modalities and enhances generalization under complex emotional variations. The proposed AdapGJ-DGNNet achieves 90.12% accuracy, 85.47% recall, 87.05% F1-score, and a ROC–AUC of 0.989, outperforming all baseline methods, with results validated through an efficient Python-based implementation. This research highlights the potential of intelligent multimodal MER systems for advancing aesthetic education.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A model for music emotion recognition and aesthetic education effect evaluation based on deep learning and multimodal data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月29日

查看全文
  • Quan Yuan

摘要

Music Emotion Recognition (MER) is vital for understanding emotional responses to music and advancing aesthetic education, enhancing artistic sensitivity and cognitive growth. However, current MER methods frequently fail to effectively model cross-modal emotional dependencies and the dynamic variations in expressive cues. This research proposes a multimodal deep learning (DL) framework that integrates the Multimodal Music Emotion dataset including 14 audios file (70 records), textual data (2000 records), and visual elements (2000 records) modalities for accurate MER and comprehensive evaluation of aesthetic education effectiveness. Preprocessing audio signals includes noise reduction, and features are extracted using Mel-Frequency Cepstral Coefficients (MFCCs) to be acoustic and timbral structures. Textual lyrics are preprocessed through tokenization, and word embeddings capture semantic and emotional context in features. Visual data is resized, and features Visual sentiment is interpreted using a convolutional neural network (CNN) that extracts face expression cues. A feature-level fusion approach combined with an attention mechanism highlights emotionally dominant attributes across modalities. The proposed Adaptive Golden Jackal-driven Dynamic Graph Neural Network (AdapGJ-DGNNet) dynamically constructs graph relationships among multimodal feature nodes and adaptively updates edge weights using the AdapGJ strategy, enabling optimal graph connectivity and robust temporal-spatial emotion modeling. The DGNNet structure facilitates learning non-linear correlations between modalities and enhances generalization under complex emotional variations. The proposed AdapGJ-DGNNet achieves 90.12% accuracy, 85.47% recall, 87.05% F1-score, and a ROC–AUC of 0.989, outperforming all baseline methods, with results validated through an efficient Python-based implementation. This research highlights the potential of intelligent multimodal MER systems for advancing aesthetic education.

Graphical Abstract