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.

翻译：本文提出一项新颖的研究问题：评估音乐产品中的正面与风险信息。我们首先建立了一个多角度、多层次音乐内容评估的基准，随后提出一种带有序性约束的有效多任务预测模型来解决该问题。实验结果表明，该方法不仅显著优于各类强任务特化模型，还能同时对多个方面进行评估。