Multi-Label Classification with Generative Large Language Models

Multi-label classification is a supervised Machine Learning problem, which can assign zero or more mutually non-exclusive class labels for an instance. It is different from the multi-class classification which assigns exactly one class label out of many predefined class labels for an instance. In this paper, we explore both proprietary and open-source generative Large Language Models (LLMs) for multi-label classification problems. Specifically, we fine-tune these LLMs and provide insights into their behaviors with different prompts and training constraints such as few-shots settings in Aviation Safety and Autonomy domains. We provide recommendations of choosing LLMs for multi-label classifications.