The recipe similarity network: a new algorithm to extract relevant information from cookbooks.

Abstract

This study integrates network science and intersection graph theory to analyse the structural properties of recipe networks in Catalan cuisine. Using three distinct cookbooks, two traditional and one haute cuisine, we construct the recipe similarity networks by linking recipes based on shared ingredients, with link weights reflecting ingredient similarity. We introduce a new, ad hoc, similarity measure that overcomes some limitations of traditional similarity metrics. We explore how different methodological approaches, such as the substitution of recipes/ingredients with their composing ingredients and link weight normalisation, influence network structure and node centrality. Our analysis reveals that recipe similarity networks are highly interconnected but show structural differences across cuisines, particularly in haute cuisine, which features more specialised recipes. Node centrality metrics identify key recipes that define culinary traditions, such as “Allioli” in traditional Catalan cuisine and “Becada con brioche de su salmis” in haute cuisine. We also develop a community detection algorithm based on link removal and clique identification, which uncovers tightly-knit recipe groups. This study advances the field of computational gastronomy by providing a methodological foundation that can be integrated with artificial intelligence techniques to support recipe personalisation, food recommendations, and gastronomic innovation.