<i>Retracted on January 14, 2026:</i> Enhancing Core-Set Active Learning: Unlocking New Frontiers in Text Classification

This study breaks new ground in Core-Set Active Learning for text classification by introducing five innovative variations that address critical challenges in the field.Our most significant breakthrough comes from applying dimensionality reduction techniques, particularly t-SNE, which yielded a remarkable 2-3% improvement in F1 score on two datasets.This advancement enhances Core-Set's effectiveness for text classification and illuminates a promising path for future research.Our comprehensive evaluation across three diverse datasets, using cutting-edge BERT and SetFit models, provides robust evidence of these improvements.Importantly, we also uncover limitations in adapting computer vision-oriented techniques to text data, exemplified by Core-Set underperforming random sampling by 5% in F1 score on one dataset.These insights are pivotal for researchers and practitioners, offering crucial guidance for refining machine learning strategies in natural language processing.By bridging the gap between Core-Set's theoretical strengths and its practical application in text classification, our study paves the way for more efficient and effective text classification models, potentially revolutionizing how we handle the ever-growing volume of unstructured textual data.