HUMMUS: A Human-Centered Approach to Sequential Recommendation Explainability through Data Humanism

Sequential music recommendation systems operate as ”black boxes,” reducing user agency and failing to support collaborative decision-making in group settings. To address these limitations, we present HUMMUS, an interactive group recommender system that applies Data Humanism principles. HUMMUS visualizes songs as flowers where petals represent audio features, connecting lines reveal algorithmic relationships, and real-time voting enables collaborative playlist creation between users and algorithms. Through a mixed-methods evaluation with 19 participants across collaborative playlist creation scenarios, we demonstrate that this humanistic approach appears to enhance user understanding of recommendations, collaborative engagement, and satisfaction. Participants reported positive emotional responses while maintaining recommendation quality. Our contributions include: a flower-based visualization technique for interpretable audio features, a real-time collaborative voting framework, and evidence that artistic metaphors enhance algorithmic transparency without sacrificing functionality. This research establishes Data Humanism as a viable framework for human-centered recommender systems that prioritize understanding and collaborative discovery alongside algorithmic sophistication.