recipe_recommender

The app recommends recipes based on the input ingredients and favourite recipes.

Recommender

Content Based Filtering

The recommender uses Content Based Filtering that uses the characteristics of items to make a recommendation for a user (see figure 1).

Figure 1: Content Based Filtering

The figure 1 shows two favorite recipes of a user. A mushroom soup with parsley and a noodles with meatballs in tomato sauce. Based on the ingredients of theses two favorite recipes the user gets a recommendation for a new recipe. The recommendation is a noodle soup with spring onions.

Cosine Similarity

The Cosine Similarity is the metric to measure the similarity between documents or in this case recipes. A value of 0 means no similarity and a value of 1 full similarity. In the following example a 2 dimensional vector space is used.

Figure 2: Cosine Similarity between banana and milk

The figure 2 shows in one dimension a banana and in the other a milk. The angle between both is 90° and the resulting value for the cosine is 0. That means that the banana has no similarity with a milk. But what happens if you mix a banana and milk to a banana shake.

Figure 3: Cosine Similarity between banana shake and milk

Due to the fact that the banana shake is a mixure of a banana and a milk the vector is between those two ingredients/products in figure 3. Therefore the angle for the banana shake between the milk (also banana shake to banana) is 45° and the resulting value for the cosine is 0.707.

Database

The database used for this project is PostgreSQL. The following image shows the schema of the database.

Tables

Ingredients

The ingredients table contains all ingredient used for the recipes.

• id (string): hash generated id based on the ingredient nane
• name (string): name of the ingredient

Recipes

The recipes table contains the basic information of all recipes.

• id (string): given id for the recipe combined of incremented number and origin
• name (string): title of the recipe
• energy_kj (float): kilo joule per portion for the recipe
• energy_kcal (float): kilo calories per portion for the recipe
• fat (float): amount of fat per portion for the recipe
• sat_fat (float): amount of saturated fat per portion for the recipe
• carbs (float): amount of carbs per portion for the recipe
• sugar (float): amount of sugar per portion for the recipe
• protein (float): amount of protein per portion for the recipe
• salt (float): amount of salt per portion for the recipe
• preparation (integer): time to prepare the recipe
• difficulty (string): difficulty to prepare to the recipe

RecipeLinks

The recipe_links table links the recipes to the ingredients with amounts and units.

• id (string): hash generated id based on the recipe_id, ingredient_id, amount and unit
• recipe_id (string): id of the recipe
• ingredient_id (string): id of the ingredient
• amount (float): amount of the ingredient
• unit (string): unit of the amount

RecipeSteps

The recipe_steps table contains the steps to prepare each recipe.

• id (string): hash generated id based on the recipe_id, step and text.
• recipe_id (string): id of the recipe
• step (integer): number of the step
• text (string): text of the current step

Users

The users table contains all the user information.

• id (integer): increasing integer starting from 1
• name (string): name of the user

UserFavorites

The user_favorites table contains all favorite recipes of the users.

• id (string): hash generated id based on the user_id and recipe_id
• user_id (string): id of the user
• recipe_id (string): id of the recipe

Data

This repository is currently not providing any recipe data for the database.

Application

Installing Packages

The necessary packages used to run the application are in the requirements.txt and installed as follows:

pip install -r requirements.txt

Configuration file

The information to connect to the database are stored in the .env file and has the following structure:

POSTGRES_USER='user'
POSTGRES_PASSWORD='password'
POSTGRES_HOST='127.0.0.1'
POSTGRES_PORT='5432'
POSTGRES_DB='database name'

Running the App

The recipe recommender is using streamlit to run the application on a web browser. To run the
application the following command needs to be executed:

streamlit run app.py

If the command is not executed inside the folder of the app, please add the path (~/path/app.py)

Using the APP

The following animation shows the basic functionality of the recipe recommender.

1. Choose the ingredients you want to use
2. Set the amount of recommended recipes you want to receive
3. Push recommend recipes to get recommendations
4. Select a recipe and push eat recipe to see full information