In this work, we propose a novel research problem: assessing positive and risky messages from music products. We first establish a benchmark for multi-angle multi-level music content assessment and then present an effective multi-task prediction model with ordinality-enforcement to solve this problem. Our result shows the proposed method not only significantly outperforms strong task-specific counterparts but can concurrently evaluate multiple aspects.