Modeling Music Emotion Judgments Using Machine Learning Methods.

Area of Science:

• Cognitive Science
• Music Psychology
• Computational Neuroscience

Background:

• The study addresses the long-standing debate between cognitivist and emotivist theories of music emotion.
• Investigates the interplay between cognitive appraisal and physiological responses in music-induced emotions.

Purpose of the Study:

• To develop and compare machine learning models for predicting emotion judgments from music.
• To determine the relative contributions of perceived and felt emotions to overall emotional experience.

Main Methods:

• Collected emotion judgments and physiological data (five channels) from 60 participants listening to 60 music excerpts.
• Utilized various machine learning (ML) methods, including neural networks, linear regression, and random forests.
• Extracted audio features for perceived emotion models and physiological features for felt emotion models.

Main Results:

• Models supported a hybrid theory, indicating emotion judgments are influenced by both perceived and felt emotions.
• Neural networks outperformed other ML methods, offering flexibility and interpretability.
• Arousal judgments were primarily driven by felt emotion, while valence judgments were mainly influenced by perceived emotion.

Conclusions:

• Machine learning models provide valuable insights into the complex mechanisms of music-induced emotion.
• A hybrid approach, integrating cognitive and physiological data, is essential for a comprehensive understanding of music emotion.
• Neural networks represent a powerful tool for modeling nuanced emotional responses to auditory stimuli.