Cooking recipe recommendation application.

🇫🇷 Français

The dataset used for recipe recommendation is Food.com Recipes and User interactions.

A score is applied for each recipe based on the user. Several factors are taken into account to determine this score:

• The ingredients available to the user
• The user's eating habits
• The season
• The user's diet (allergies, vegetarian, halal, etc.)

Arbitrarily, the score of a recipe is calculated as follows:

score = (ingredients_score + habits_score + season_score) * diet_score

Note ingredients_score could be the percentage of ingredients in the recipe that the user has available. diet_score would be equal to 0 if the recipe contains an ingredient that the user does not wish to consume. Otherwise, it would be equal to 1.

The user can enter the ingredients he has available.

Different clusters of recipes are created according to the users' interactions with the recipes.

The user can enter the recipes he likes during the application's onboarding. These recipes are used to determine which recipes are likely to appeal to the user.

Graph representing the clusters of recipes

As the users' interactions are dated, it is possible to determine the average and/or median period of the year when each recipe is most cooked.

The user simply indicates during the application's onboarding the ingredients he does not wish to consume.

• Normalization of the name of the recipes (using a generative AI)
• Grouping ingredients by diet
• Image recognition to add ingredients available to the user