Idea
Our idea is to let a user explore a large dataset of recipes, marking recipes as liked or disliked. From that the app learns the preferences of the user and can predict how likely the user is to like an unseen recipe.
In Practice
To make the exploration fun, we decided to mimic the Tinder user experience. The app shows the user an image, name, ingredients, tags usw. of a recipe and the user then can mark it by swiping to the left or right. The frontend was written with Unity. The app connects to a server backend where the data is stored. After each swipe the labeled dataset is used to train a kernelized Support Vector Machine (SVM). Some data such as ingredients and tags are strings. We convert this to a feature vector by lerning the vocabulary used for each category and counting the occurances of words. The whole machine learning is implemented using sklearn.
For the future
Sadly we had multiple features in mind, that we could not implement due to time constraints.
- Hard Constraints: Use allergies or diet information gathered to filter the proposed recipes
- Show a list of liked recipes
- Different ways for exploration: e.g. show recipes that the user is likely to like or unlikely to like.
Log in or sign up for Devpost to join the conversation.