A recommender system for multimodal data about food recipes. This is a project for the Artificial Intelligence course at Università degli Studi di Milano Bicocca (a.a. 2021/2022), and the work was done by me and my classmate Kevin Manella.
The dataset we used, "Food Ingredients and Recipes Dataset with Images", is available at Kaggle website here.
For the task of multimodal embedding we used the following resources:
- Image embedding: VGG16 implemented with Tensorflow and Keras;
- Paragraph embedding: Doc2Vec implemented with the Gensim library.
A graph that shows the multimodal embedding approach we implemented:
A graph that shows the approach implemented to get similar recipes given the embeddings:
An example of promising results on data already present in the original dataset:
An example of promising results on new data found on the web (Google images):
