Cole T. Peterson, M.S.

Recipes are a unique form of procedural knowledge that is historically lacking systematic analysis despite their relevance in culinary arts. While recipes appear to follow standardized formats, they rely heavily on implicit domain knowledge, cultural context, and unstated assumptions that make them difficult to parse, compare, and quantitatively analyze. No publicly available tools exist to decompose recipes into discrete components that would enable programmatic analysis, comparison across recipe variations, or extraction of underlying cooking principles. This gap prevents systematic study of culinary knowledge and development of tools that could make recipes more accessible to users of varying cooking backgrounds and skills. This research addresses these limitations by developing a novel methodology for representing recipes as directed acyclic graphs (DAGs), partially inspired by binary expression trees (BET) where ingredients and equipment used in a recipe represent operands and cooking actions represent operators. We had three main objectives: (1) create a framework based on existing data structures and programming tools capable of abstracting any recipe regardless of complexity, semantics, or cultural context. (2) Identify quantitative metrics that could be extracted from recipe DAGs to calculate recipe complexity and enable meaningful comparisons between them. And (3) to develop a software toolset for recipe ingestion, manipulation, visualization, and analysis through a standardized API. Our analysis revealed significant complexity and variation within the dataset, with critical path lengths ranging from 4 to 48 edges (average 17.1). The dataset contained 203 unique ingredients across 815 occurrences, with high specialization (136 ingredients appearing in only single recipes). Dominant ingredients include sugar (64.5% of recipes), flour (55.3%), and eggs (52.6%). Co-occurrence analysis revealed fundamental baking patterns, with flour + sugar appearing in 46.1% of recipes. Intermediate product comparison using Jaccard similarity (mean: 0.052) demonstrated that intermediate naming conventions capture functional roles rather than strict compositional specifications. The weighted complexity formula C = 0.5B + 0.35A + 0.15I enabled skill-level classification into beginner, advanced, and expert categories based on branch count, action count, and ingredient count. This work establishes a comprehensive methodology for systematic recipe analysis through graph-theoric approaches, providing a foundation for future applications in recipe recommendation systems, instruction generation, cooking education, and temporal/cultural culinary analysis.