Tag
This paper systematically investigates reward function design for reinforcement learning to improve the quality of LLM-generated BPMN process models, finding that equal reward weighting outperforms targeted weighting and that design choices interact with model architecture in non-trivial ways.