This model is a fine-tuned DistilBERT classifier that predicts a single food category for a given product/ingredient name.
Labels (7):
vegetable
fruit
dairy
meat
beverage
snack
grain
Typical output:
a category label (optionally with a confidence score)
Eval Results
Synthetic split (train/validation/test from the same generated distribution)
Accuracy: ~0.99
Macro F1: ~0.988
Manual unseen set (n=20)
Accuracy: 0.70
Note: The manual unseen set is small and intended as a quick sanity check for generalization to unseen product names.
Dataset
This model was fine-tuned on a custom product classification dataset.
The dataset consists of product/ingredient names mapped to 7 coarse food categories:
vegetable
fruit
dairy
meat
beverage
snack
grain
The data includes a mix of manually labeled and generated samples.
All entries are short product titles or ingredient names in English.
Note:
The dataset is domain-specific (food products only) and does not cover non-food categories.
Model performance depends on vocabulary coverage and label consistency.
Intended Use
Food/product categorization for search, filtering, or analytics
Categorizing short product titles, ingredient names, and simple food phrases
A lightweight baseline model for food taxonomy experiments
Not Intended For
General “anything in the world” categorization (electronics, cosmetics, etc.)
