This paper proposes a solution to the challenges of data collection and cost, which rely on human annotators for training and validating machine learning models for text classification. Existing methods rely on manual labeling by experts, which is time-consuming and expensive. Furthermore, this approach incurs ongoing costs, especially when model retraining is required (data/model changes). Therefore, this paper proposes several approaches that utilize large-scale language models (LLMs) to validate the predictive accuracy of classifiers. This approach replaces human annotators, ensures model quality, and supports high-quality incremental learning.
