501 world recipes as an open dataset: per-serving nutrition, step timings, ingredient scaling rules (CC BY-SA 4.0)

A monumental collection of 501 home-cooking recipes from 127 countries has been made available as an open dataset, providing comprehensive details for each dish. The data, released under a Creative Commons Attribution-ShareAlike 4.0 International license, includes per-serving nutrition information, such as calories, protein, fat, and carbohydrates, for all 501 dishes.

The dataset contains over 3,200 step-by-step instructions, with each step broken down into minutes, ensuring accurate timing for each dish. Every ingredient is given a unique identifier and scaling rules, allowing users to adjust quantities for different serving sizes with precision. For instance, meat typically scales linearly with servings, while salt and spices are adjusted using a damping mechanism, mimicking the way a real kitchen would scale the recipe.

All ingredients are linked to stable IDs and feature scaling rules, including linear, damped, or fixed scaling. This precise ingredient scaling ensures that users can build a portion calculator directly on top of the data, eliminating the need for manual calculations. The photos used in the dataset are sourced from Wikimedia Commons, each accompanied by author and license information.

The dataset's creators have emphasized the importance of the scaling rules, as many other recipe datasets store ingredient quantities in a format that demands manual scaling, leading to inaccurate results. By providing explicit scaling rules, this dataset simplifies the process of adapting recipes to different serving sizes, making it an invaluable tool for meal planners, recipe developers, and data enthusiasts alike.

The dataset, available in both JSON and CSV formats, can be downloaded from the provided links. Users are encouraged to credit "UniTools — theunitools.com" when using the data and to share derivatives under the same Creative Commons Attribution-ShareAlike 4.0 International license. The creators have also included honest caveats regarding the dataset's limitations, clarifying that the nutrition values are computed from ingredients, not lab-measured, and that the dataset is maintained by a single person.

Users are invited to report any errors they find by opening an issue on the repository, ensuring the dataset remains accurate and up-to-date.