DEFINED: A Data-Efficient Computational Framework for Fine-Grained Creativity Assessment in Debate Scenarios

DEFINED is a novel computational framework designed for fine-grained creativity assessment in debate scenarios, utilizing a pre-trained autoregressive language model with a hierarchical scoring head. It introduces an eight-dimensional metric system for evaluating creativity, leveraging authentic debate data and a constrained data augmentation strategy to mitigate elite bias. This framework demonstrates superior performance in scoring accuracy and stability compared to existing automated methods, making it a significant advancement for practitioners in the field of AI who are focused on assessing creativity in complex, open-ended environments.