Multi-label Text Classification Model Based on Multi-level Constraint Augmentation and Label Association Attention

In the multi-label text classification task, a usually corresponds to multiple label categories, and labels have correlation hierarchical structure. However, when hierarchy is unknown, number of various not balanced, which makes it difficult for model classify low-frequency labels. At same time, due existence similar labels, will be distinguish this paper, we propose based on multi-level constraint augmentation association attention. Compared with traditional methods, our method has two contributions: (1) order alleviate problem unbalanced different categories ensure rationality sample generation, data constraints. process uses historical generation information, original information topic constrain generated text. (2) make recognize associated accurately, an interaction mechanism attention filter gate. This combines weight information. considers important weights sentences effectively utilizes co-occurrence relationship between Experimental results three benchmark datasets show that outperforms state-of-the-art methods all main evaluation metrics, especially prediction sparse samples.