TnT-LLM: LLMs for Automated Text Taxonomy and Classification

A new two-phase method for text classification using large language models (LLMs) has been developed. The first phase involves generating a taxonomy, while the second phase uses LLMs to create pseudo-labeled training data for more efficient classifiers. This approach leverages the strengths of LLMs as annotators and allows for the deployment of lightweight models that can perform label assignments at scale and in real-time. The method aims to improve text classification accuracy and efficiency for large-scale applications.