Computational gastronomy: A study to test the food pairing hypothesis in Turkish cuisine

Abstract

This study adopts computational gastronomy methods to test the food pairing hypothesis. For this purpose, Food Pairing Index values (the average number of shared flavor molecules between ingredient pairs) were calculated for recipes from seven local cuisines in Turkey. These values were examined for correlation with user ratings of recipes, and compared with the FPI values of randomly generated recipes. Additionally, the contribution of individual ingredients to the FPI values and the frequently used ingredients in the cuisines were analyzed. The results revealed that FPI values were not significantly positively correlated with user ratings, suggesting that the food pairing hypothesis cannot be fully supported. Also, comparisons with randomly generated recipes show that FPI values are heavily influenced by the frequency of ingredient usage in kitchens. In particular, positive food pairing behavior was observed in some regional cuisines, while negative food pairing behavior was found in others. However, these patterns were consistent with the similarities observed in other kitchens. Overall, the findings, when evaluated together with previous findings and criticisms, show that the food pairing hypothesis and the current computational gastronomy approach to it are partially successful in food pairing.