Prof. Jaewoo Kang’s research team develops AI-based recommendation of food pairings
Optimal pairings discovered from analysis of 1 million recipes
Pairings successfully applied to actual food
Results to be presented at leading AI conference IJCAI-19
▲ (from the left) Professor Jaewoo Kang and
PhD student Donghyeon Park of the Department of Computer Science and Engineering
A research team led by Professor Jaewoo Kang of the Department of Computer Science and Engineering under the College of Informatics developed an AI-based model that recommends new and creative food pairings based on a comprehensive analysis of 1 million recipes.
Recently, there has been an explosive increase in data across various domains. Researchers have applying domain-specific AI to make predictions based on data analysis with consideration of specific domain characteristics.
While AI has already been applied to many domains, the realm of food remains largely unexplored. Because food ingredients are comprised of complex combinations of chemical structures, data-based food analysis is perceived as highly challenging.
Prof. Kang’s research team developed a deep learning system that applies Siamese neural networks to food data. More than 1 million recipes were analyzed to acquire knowledge on 300,000 food ingredient pairings. Learning the relationships of food pairings with Siamese neural networks laid the foundation for the discovery and recommendation of novel pairings. Users can visit the website and utilize the search function to search for pairings of the ingredients they have.
The food pairing recommendation model was qualitatively assessed based on how close the scores were to complementary pairings. The model surpassed conventional machine learning models in terms of prediction and recommendation performance.
To demonstrate the practicality of the model, experiments were conducted comparing the model’s recommendations to food pairings suggested by cooking experts. The recommended pairings were found to be consistent with pairings in Karen Page’s “What to Drink with What You Eat,” a well-known food pairing guide book (e.g., red wine with meat, white wine with seafood). The model was also capable of recommending new and creative cocktail recipes (e.g., gin with aquavit, champagne with lemon sorbet).
The key to enhancing competitiveness in the future is to learn and predict new trends from the ever-growing body of domain-specific knowledge. In the world of food, developing new culinary trends and creative recipes is especially important.
The significance of this study lies in it being the first to apply an AI-based analysis of big data to food pairings and establishing a framework for future work by presenting the necessary data and algorithms for model training.
The results will be presented under the title of “KitcheNette: Predicting and Recommending Food Ingredient Pairings using Siamese Neural Networks” at IJCAI-19, a leading AI conference to be held in Macau, China from August 11 to 16.