Study Recovers Expert-Based Artist Similarity Networks from Audio for Music Recommendation

A machine learning paper introduces a method that infers adjacency links between musical artists from the acoustic distributions of their recordings, treating expert critic judgments as a reference standard. The authors position this as a third recommendation signal beyond user interaction data, which struggles in cold-start settings, and intrinsic audio content alone. A construct-validity framework is used to assess whether the recovered networks meaningfully reflect artist similarity.