[1M3-GS-10] AI application: Materials

[1M3-GS-10-02] Statistical Analysis of Highly Rated Recipes Based on Flavor Compounds in the Ingredients

〇Hiroshi Uehara^1,3, Daichi Mochihashi^2,3 (1. Rissho University, 2. The Institute of Statistical Mathematics, 3. SOKENDAI)

Keywords:Recipe sharing site, Positive pointwise mutual information, Flavor compounds, Odor perception

Flavors of dishes are perceived through complex combinations of flavor compounds, and the relationship between human odor perception and the combination have not been unveiled. This study statistically constructs features of flavor perceptions represented by flavor compounds, and tries to find the relationship between the flavor features and tasty flavors of dishes. For this purpose, we statistically inferred flavor features from numerous recipes of a recipe sharing site, and evaluated the performance of distinguishing highly rated recipes from those that are not. Then, we analyzed patterns of flavor compounds contributing to the distinctions. The result showed the flavor features outperformed any other vector concerning the recipe information. Further, flavor pattern analysis implied tasty recipes have specific patterns of flavor compounds.