APEX: Large-scale Multi-task Aesthetic-Informed Popularity Prediction for AI-Generated Music

Researchers introduce APEX, a machine learning framework that predicts popularity of AI-generated music by analyzing both engagement metrics and aesthetic quality across 211k songs from platforms like Suno and Udio. The model demonstrates strong generalization capabilities when tested on unseen generative music systems, suggesting that aesthetic dimensions are crucial predictors of music popularity in the AI-generated music landscape.

The emergence of APEX addresses a critical gap in understanding AI-generated music markets, where traditional popularity predictors like artist reputation and label backing no longer apply. As platforms like Suno and Udio democratize music creation, the industry lacks frameworks for evaluating what drives listener engagement in this rapidly expanding space. APEX's multi-task learning approach uniquely combines engagement signals—streams and likes—with five perceptual aesthetic quality dimensions, creating a more comprehensive popularity prediction model than engagement-only approaches.

This research reflects broader industry trends where AI-generated content platforms are maturing and require sophisticated analytical tools. The dataset of 211k songs (10k hours) from major generative platforms provides empirical evidence that aesthetic quality matters alongside popularity metrics, validating what users intuitively understand about music consumption. The model's strong performance on out-of-distribution data from eleven unseen generative systems suggests these learned representations capture fundamental properties of music quality that transcend architectural differences.

For the music and AI industries, APEX creates immediate practical applications. Content platforms can use similar frameworks to surface higher-quality AI-generated music, improving user experience and reducing listener churn from low-quality content saturation. Artists using AI tools gain insights into quality factors that drive engagement. Investors in generative music platforms should recognize that aesthetic quality measurement becomes a competitive moat—platforms implementing quality filters may achieve better retention and user satisfaction than those relying solely on raw generation volume.

Looking ahead, the standardization of music quality metrics could enable secondary markets for AI-generated music, licensing frameworks, and algorithmic curation systems that outperform current random-discovery models.

• →APEX framework predicts AI-generated music popularity using both engagement metrics and five aesthetic quality dimensions across 211k songs.
• →Aesthetic quality features improve generalization across multiple generative music systems unseen during training.
• →The model addresses a market gap where traditional artist-reputation signals don't apply to AI-generated music platforms.
• →Research suggests quality measurement could become a competitive differentiator for generative music platforms.
• →Framework enables better content curation and user retention strategies for AI music platforms.