Leveraging Large Language Models to Address Data Scarcity in Machine Learning: Applications in Graphene Synthesis