The paper introduces TnT-LLM, a two-phase framework that leverages Large Language Models (LLMs) to automate the process of label generation and assignment in text mining. The first phase employs a zero-shot, multi-stage reasoning approach to iteratively produce and refine a label taxonomy using LLMs. The second phase uses LLMs as data labelers to generate training samples for lightweight supervised classifiers. The framework is applied to analyze user intent and conversational domains in Bing Copilot, demonstrating improved accuracy and relevance of label taxonomies compared to state-of-the-art baselines. The evaluation includes human and automatic metrics, showing that TnT-LLM achieves a favorable balance between accuracy and efficiency for large-scale text classification. The paper also discusses practical insights and challenges in using LLMs for text mining in real-world applications.The paper introduces TnT-LLM, a two-phase framework that leverages Large Language Models (LLMs) to automate the process of label generation and assignment in text mining. The first phase employs a zero-shot, multi-stage reasoning approach to iteratively produce and refine a label taxonomy using LLMs. The second phase uses LLMs as data labelers to generate training samples for lightweight supervised classifiers. The framework is applied to analyze user intent and conversational domains in Bing Copilot, demonstrating improved accuracy and relevance of label taxonomies compared to state-of-the-art baselines. The evaluation includes human and automatic metrics, showing that TnT-LLM achieves a favorable balance between accuracy and efficiency for large-scale text classification. The paper also discusses practical insights and challenges in using LLMs for text mining in real-world applications.