An Efficient and Precise Training Data Construction Framework for Process-supervised Reward Model in Mathematical Reasoning | Synapse