Weakly supervised token classification model for German recipe texts based on bert-base-german-cased.

Code available: https://github.com/chefkoch24/weak-ingredient-recognition

Dataset: https://www.kaggle.com/datasets/sterby/german-recipes-dataset

Recognizes the following entities:<br> 'O': 0, <br> 'B-INGREDIENT': 1,<br> 'I-INGREDIENT': 2,<br> 'B-UNIT': 3,<br> 'I-UNIT': 4,<br> 'B-QUANTITY': 5,<br> 'I-QUANTITY': 6<br>

Training: <br> epochs: 2<br> optimizer: Adam<br> learning rate: 2e-5<br> max length: 512<br> batch size: 8<br> recipes: 7801<br>

The model was trained on single Geforce RTX2080 with 11GB GPU

Metrics on test set (weakly supervised): <br> accuracy_token 0.9965656995773315<br> f1_token 0.9965656995773315<br> precision_token 0.9965656995773315<br> recall_token 0.9965656995773315<br>