Automated Assessment of Positive and Risky Messages in Music Lyrics

This study introduces a novel NLP task to assess the intensity of positive and risky messages conveyed in music lyrics across multiple dimensions, including Violence, Substance Consumption, Sex, Consumerism, and Positive Messages. The authors propose an effective multi-task, ordinality-enforced predictive model that outperforms robust task-specific alternatives and provides valuable insights through detailed case studies.