Souvika Sarkar

The paper proposes a zero-shot multi-label topic classification framework and finds that while knowledge graph augmentation improves performance for smaller language models, it offers diminishing returns for larger, pre-trained models.