Large Language Model Data Augmentation for Text-Pair Classification Tasks

In recent years, large language models (LLMs) have demonstrated remarkable capabilities across various natural language processing tasks.This study explores the application of LLMs for data augmentation in text-pair classification tasks, such as semantic textual similarity and natural language inference.We propose a novel framework that leverages LLMs to generate diverse and contextually relevant paraphrases and transformations of text-pairs, enhancing the training data without manual annotation effort.Our experiments on widely-used benchmarks show that the augmented data not only improves model performance but also increases its robustness to out-of-domain examples.We perform extensive ablation studies to understand the contribution of different augmentation strategies and analyze the trade-offs between data diversity and noise.Additionally, we assess the generalization capabilities of models trained with augmented data across multiple architectures and dataset sizes.The results suggest that LLM-driven data augmentation is a promising approach to overcome data scarcity, reduce overfitting, and enhance the adaptability of text-pair classification systems.