TASTEset

This is a dataset for food entities recognition problem. It consists of 1,000 manually annotated recipes' ingredients. Recipes were scraped from the following websites:

• https://www.allrecipes.com
• http://food.com
• https://tasty.co
• www.yummly.com

using this tool.

The dataset encapsulates 15 entities:

FOOD, UNIT, QUANTITY, PHYSICAL QUALITY, PROCESS, COLOR, TASTE, PURPOSE, PART, TRADE NAME, DIET, EXAMPLE, EXCLUDED, EXCLUSIVE, POSSIBLE SUBSTITUTE.

The dataset is available in the CSV format, here. It contains two columns:

• ingredients - list of recipe's ingredients
• ingredients_entities - entities manually annotated in the list of ingredients

The ingredients_entities are of the following format:

• "span": list of tuples, each tuple contains of the start and end character id. If more tuples are present, it means that the entity is discontinuous.
• "type": entity type
• "entity": entity

An example from the dataset:

List of ingredients

• 1 box powdered sugar
• 8 oz. soft butter
• 1 (8 oz.) peanut butter
• paraffin
• 12 oz. chocolate chips

More details on the dataset can be found in our paper.

Project environment

Python 3.6.9 has been used. If you have other version, perhaps you might need to adjust the requirements.txt. You can also consider using pyenv.

1. If you use virtualenv

python -m venv tasteset_env
. ./tasteset_env/bin/activate
pip install --upgrade pip  # upgrade pip to newest version
pip install -r requirements.txt
python -m spacy download en_core_web_sm

2. If you use pyenv

pyenv virtualenv 3.6.9 tasteset_env
pyenv activate tasteset_env
pip install --upgrade pip  # upgrade pip to newest version
pip install -r requirements.txt
python -m spacy download en_core_web_sm

To freely import all scripts, please add the followings to the PYTHONPATH:

export PYTHONPATH=PYTHONPATH:<path-to-tasteset-repo>/src

Baselines

Below are F₁-score results of state-of-the-art models on 5-fold cross-validation. You can run your own experiments, please refer to the following folder.

Note that these results do not include the rarest entities: EXCLUDED, EXCLUSIVE, POSSIBLE SUBSTITUTE.

Entity	BERTbase-uncased	BERTbase-cased	BERTlarge-cased	BioBERT	FoodNER	BERT with CRF	LUKE
FOOD	0.921 +- 0.006	0.927 +- 0.010	0.932 ± 0.006	0.823 ± 0.020	0.928 ± 0.008	0.928 ± 0.006	0.923 ± 0.004
QUANTITY	0.982 +- 0.002	0.986 +- 0.001	0.984 ± 0.003	0.893 ± 0.012	0.985 ± 0.001	0.984 ± 0.003	0.981 ± 0.003
UNIT	0.979 +- 0.008	0.983 +- 0.006	0.982 ± 0.007	0.884 ± 0.020	0.982 ± 0.008	0.982 ± 0.007	0.977 ± 0.008
PROCESS	0.928 +- 0.008	0.938 +- 0.012	0.939 ± 0.010	0.817 ± 0.013	0.934 ± 0.006	0.930 ± 0.008	0.903 ± 0.008
PHYSICAL QUALITY	0.824 +- 0.032	0.830 +- 0.017	0.829 ± 0.025	0.751 ± 0.043	0.822 ± 0.025	0.816 ± 0.024	0.795 ± 0.020
COLOR	0.907 +- 0.022	0.926 +- 0.016	0.917 ± 0.022	0.786 ± 0.047	0.921 ± 0.016	0.931 ± 0.008	0.898 ± 0.024
TASTE	0.848 +- 0.030	0.854 +- 0.027	0.876 ± 0.014	0.793 ± 0.039	0.854 ± 0.028	0.854 ± 0.057	0.791 ± 0.046
PURPOSE	0.832 +- 0.056	0.875 +- 0.054	0.872 ± 0.050	0.724 ± 0.097	0.857 ± 0.056	0.814 ± 0.078	0.790 ± 0.048
PART	0.744 +- 0.014	0.764 +- 0.058	0.831 ± 0.058	0.785 ± 0.050	0.789 ± 0.054	0.820 ± 0.070	0.783 ± 0.077
TRADE NAME	0.444 +- 0.107	0.538 +- 0.034	0.612 ± 0.110	0.145 ± 0.069	0.543 ± 0.053	0.557 ± 0.113	0.544 ± 0.031
DIET	0.668 +- 0.101	0.687 +- 0.095	0.711 ± 0.099	0.650 ± 0.086	0.668 ± 0.122	0.751 ± 0.059	0.661 ± 0.089
EXAMPLE	0.422 +- 0.069	0.466 +- 0.071	0.518 ± 0.125	0.216 ± 0.156	0.430 ± 0.097	0.414 ± 0.097	0.537 ± 0.052
all	0.933 +- 0.007	0.941 +- 0.006	0.943 ± 0.004	0.842 ± 0.018	0.939 ± 0.005	0.938 ± 0.003	0.928 ± 0.005

Please refer to our paper for models' details.