AI-Assisted Curation: A Framework for Curatorial Materials

This work proposes a framework for the generation of AI-assisted curatorial materials, including an input strategy and an evaluation benchmark, based on literature review and interview. To find the most efficient input strategy and validate the benchmark, 4 methods were tested to generate curatorial text in the Tanka Culture AI Art Exhibition and then the generated texts are evaluated through the proposed evaluation benchmark by LLMs and humans. Our results suggest that Method 3 is the most effective, with a perfect balance of preparation time and text quality. This study contributes a reliable approach to curatorial text creation, offering a valuable tool for both curators and AI technicians, and fostering more efficient and contextually relevant exhibition texts.