Flavor analysis and region prediction of Chinese dishes based on food pairing

Abstract

There are thirty-four provincial administrative regions in China, and each region possesses its own distinct food culture. However, existing studies on flavor-based research often treat Chinese dishes from various regions as a single entity, resulting in biases when examining regional distinctions and the variety of recipes. To explore the potential correlation between food ingredients and regions, this paper performs flavor analysis and region prediction for Chinese typical dishes. A new dataset including 1534 recipes combined with a compound-ingredient list to construct a compound-ingredient-recipe network. The results indicate that in terms of food ingredient selection, Hainan shows the highest degree of flavor compound sharing, while Shanghai does the lowest. While all the regions studied tend to exhibit negative food pairing according to the food pairing hypothesis, a region with a high degree of flavor sharing is more able to become prominent representation of the corresponding flavor branch through the analysis of representative ingredients. Across the 14 regions studied, the frequency of ingredient usage within a region plays a crucial role in maintaining its distinct flavor profile. Concerning the individual contributions of ingredients to flavor compound sharing, the positive contributions are linked to rice and natural spices, including Welsh onion, ginger, and garlic, while the negative contributions are primarily associated with flour, pork, and traditional Chinese condiments such as soy sauce, cooking wine, and grain vinegar. Multiple deep learning experiments demonstrate the effectiveness of combining ingredient and flavor information to region prediction for recipes.