A Reasoning Framework for Precise Cooking Process Based on Graph-Guided Chain of Thought

The growing adoption of intelligent cooking devices has revealed a critical semantic discrepancy between the qualitative descriptors in traditional cooking knowledge systems and the control requirements of these automated platforms. Current systems struggle to address the fundamental demand for culinary automation, particularly in achieving precise parametric quantification, due to the lack of well-defined cooking processes directly translatable to device control instructions. To address this challenge, this paper proposes a framework for precise cooking process reasoning based on a cooking element relationship graph to guide the chain of thought. The framework dissects culinary processes into four core reasoning tasks (ingredient preprocessing shape reasoning, cooking temperature reasoning, cooking time reasoning, and seasoning quantity reasoning), and subsequently constructs a culinary element relationship graph encompassing ingredient physical properties, culinary process parameters, and cuisine characteristics. This structured knowledge representation serves as prior knowledge to guide the initial chain-of-thought generation in large language models, effectively constraining logical deviations in domain-specific reasoning while maintaining culinary operational consistency. The experimental results demonstrate that the framework exhibits a substantial enhancement in four distinct types of reasoning tasks, thereby substantiating the efficacy of the knowledge-guided chain of thought in culinary parameter reasoning. By establishing a digital reasoning paradigm for the cooking process, this study provides a theoretical framework and technical path for the development of precision and personalization of intelligent cooking technology.