Team led by Professor Kang Jae-woo developed an AI model that reveals the secrets of taste.
Through joint research with ‘SONY AI’, an artificial intelligence is developed to recommend creative combinations of ingredients.
The optimal combination can be recommended due to the addition of the chemical molecular information of 1,561 flavor compound nodes to the existing model.

The research team of Professor Kang Jae-woo of Korea University and ‘SONY AI’ collaborated to jointly develop an artificial intelligence model that recommends creative combinations of ingredients.

The newly developed artificial intelligence model uses the characteristics of chemical ingredients in food ingredients and big data comprised of recipes used in the market to create and recommend optimal combinations of ingredients never used before.

'FlavorGraph', an integrated food graph which consists of a total of 1,561 chemical molecules and information on the relationships between the food ingredients analyzed from 1 million recipes, was used to teach the artificial intelligence model. This model is of great significance in the field of food science in that it can predict new relationships between chemical molecules and food and develop creative and new recipe combination techniques.

In 2019, the research team also developed an artificial intelligence model that recommends combinations of ingredients, but, at the time, only recipe information was used. The key feature of the newly developed model is that, in addition to the recipe pairing information used in the existing model, 1,561 chemical molecules were simultaneously graphically analyzed. This chemical information was added to the accumulated recipe information, thereby allowing the derivation of more accurate and diverse recipes.

This research is the result of joint research between 'SONY AI' and Research Professor Park Dong-hyeon (Institute of Computer, Information and Communication of Korea University) of Professor Kang Jae-woo’s research team, playing a leading role.

'SONY AI' is a subsidiary of SONY and was established in 2019 to develop artificial intelligence technology. Along with games and image sensors, which are the main industries of the SONY group, gastronomy is also included as one of the company’s three major fields of research, and research is underway to apply artificial intelligence to them all.

In 2019, Professor Park developed the ‘Kitchenette’, an artificial intelligence model that recommends combinations of ingredients, and presented a paper at the renowned international conference on artificial intelligence, ‘IJCAI’. This caught the eye of ‘Sony AI’ and at their request, a joint research project was launched. The ‘Kitchenette’ artificial intelligence model became the basis for the currently developed ‘FlavorGraph’.

As a follow-up study with ‘Sony AI’, Korea University is planning to develop an artificial intelligence model that can create a new recipe considering the flavor and nutrition of food using ‘FlavorGraph’. Through this, the plan is to establish a Precision Nutrition program that recommends personalized recipes and diets that consider individuals’ genetic characteristics, lifestyle habits, and health conditions.

The results of this joint study were published last January in the Scientific Report, a sister magazine to Nature magazine.

Related information:
1. ‘SONY AI’ official blog: https://ai.sony/blog/blog-004/
2. Paper on ‘FlavorGraph’: https://www.nature.com/articles/s41598-020-79422-8