Label Hierarchy

Label hierarchy research focuses on improving classification accuracy and efficiency by leveraging the inherent structure within label sets, where labels are organized into hierarchical relationships (e.g., a taxonomy). Current research emphasizes developing models that effectively incorporate this hierarchical information, employing techniques like contrastive learning, graph neural networks, and prompt tuning within various architectures including BERT and LLMs, to better align text representations with the label hierarchy. This work is significant because it addresses challenges in multi-label classification, particularly in scenarios with imbalanced data or limited training examples, leading to improved performance in diverse applications such as medical diagnosis and document categorization. The development of data-driven methods for discovering or refining label hierarchies is also a growing area of interest.