Regularizing Model Complexity and Label Structure for Multi-Label Text Classification