AINeutralarXiv – CS AI · Mar 44/102
🧠
Boosting Meta-Learning for Few-Shot Text Classification via Label-guided Distance Scaling
Researchers propose a Label-guided Distance Scaling (LDS) strategy to improve few-shot text classification by leveraging label semantics during both training and testing phases. The method addresses misclassification issues when randomly selected labeled samples don't provide effective supervision signals, demonstrating significant performance improvements over state-of-the-art models.