Computational gastronomy: an artificial intelligence driven approach to recipes, flavors, nutrition, health, and sustainability

Abstract

Cooking is a quintessential creative pursuit with profound significance for humanity. Food and cooking transcend mere sensory pleasure and have serious nutrition and public health outcomes. Beyond being linked to the culinary and cultural heritage, food systems play a pivotal role in sustainability and are therefore critical for the very survival of life on the Earth. In an era where data-driven and computational insights are transforming every domain, culinary endeavors have primarily been seen through an artistic outlook. This thesis delves into various facets of computational gastronomy, presenting a comprehensive framework for the organization and analysis of cooking recipes, novel recipe generation, algorithmic creativity, molecular informatics, and sustainability. The work presented here bridges gaps in food science, flavor science, and food system studies by addressing the challenges in the structure of recipes, prediction of molecular flavor, and generating novel recipes. The journey of this thesis starts with the development of RecipeDB2, a scalable and structured framework for representing recipes, associated ingredients, and nutritional profiles of diverse cuisines from across the globe. Building upon RecipeDB2, a recipe ontology was introduced, which integrates recipes, ingredients, and flavor molecules. Such an ontology, which incorporates diverse datasets, including RecipeDB2 and FlavorDB2, serves as the backbone for subsequent analyses, enabling semantic reasoning and supporting various machine-learning tasks. To build on this structured representation, we implemented deep-learning-based named entity recognition algorithms on recipes, extracting meaningful entities from unstructured ingredient phrases. After creating a structured database of recipes, we delved into novel recipe generation by fine-tuning large language models in a generic as well as cuisine-specific manner. These machine-generated recipes were further evaluated through the Turing Test for Chefs to assess the efficacy of novel recipe generation strategies. Sustainability is another pivotal theme in this thesis. By evaluating the carbon footprint of recipes, one can assess the environmental impact of tradition-driven culinary choices, thus enabling informed decision-making for a sustainable food future. Apart from culinary science, this thesis delves into molecular informatics by introducing FlavorDB2, a dataset that documents the molecular composition and flavor profiles of natural ingredients. By understanding the chemical properties of flavors, this research deepens the understanding of taste and its potential applications in culinary arts. Aligned with this thread, the thesis explores the application of machine learning and deep learning models to predict the taste (sweet, umami) and toxicity of small molecules, towards addressing challenges in food safety, lifestyle disorders, and culinary innovation. The thesis culminates with a real-world application of computer vision for dish detection in food platters. We showcase the use of object detection techniques for correctly identifying dishes from the Indian platters, as a proof of concept, with potential applications for meal logging, diet management, and personalized nutrition. This thesis offers a multidimensional perspective on computational gastronomy by integrating structured data, natural language processing, large language models, deep learning, molecular informatics, and computer vision. It presents the story of how Artificial Intelligence can mimic human creativity and enhance our understanding of food, enabling innovations that foster better health, and environmental sustainability. By bridging the culinary arts with data science and computation, this work contributes to the growing interdisciplinary field of computational gastronomy, shaping the future of food through computational approaches. Computational gastronomy offers a data-driven approach to food, paving the way for ground-breaking advancements in the culinary landscape. This emerging field defines the traditional, artistic outlook toward food and cooking, demonstrating how the fusion of food, data, and computation can lead to innovative and sustainable culinary solutions.