Text Classification based on Active learning and Meta-learning

To tackle the limitations of traditional text classification methods that depend on manually engineered features and deep learning techniques that necessitate extensive labeled data, this paper proposes a novel approach that integrates meta-active learning for text classification. This method minimizes reliance on labeled data and enhances rapid adaptation to new tasks. The most valuable sample annotations are selected by TextGCN to accelerate new domain understanding. When our propose model is combined with adversarial domain adaptation meta-learning, its adaptation and performance are optimized. Experimental results indicate that our proposed model outperforms several baseline models across three benchmark datasets, showcasing its effectiveness.