---
title: ChEMU NER Demo
emoji: ⚗️
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.49.1
python_version: "3.12"
app_file: app.py
pinned: false
license: cc-by-nc-3.0
short_description: BioBERT-based NER for chemical patents (ChEMU 2020)
---

# ChEMU NER Demo

Interactive demo for a BioBERT-based named entity recognition model
fine-tuned on the [ChEMU 2020 Task 1](https://chemu-patent-ie.github.io/)
chemical patent corpus.

The model ([`mpkato/chemu-biobert-ner`](https://huggingface.co/mpkato/chemu-biobert-ner))
extracts 10 reaction-step entity types from patent text:

- Compounds: `STARTING_MATERIAL`, `REAGENT_CATALYST`, `REACTION_PRODUCT`, `SOLVENT`, `OTHER_COMPOUND`
- Conditions: `TEMPERATURE`, `TIME`
- Yields: `YIELD_PERCENT`, `YIELD_OTHER`
- Labels: `EXAMPLE_LABEL`

Held-out dev exact-match micro-F1 ≈ 0.95.

## Usage

Paste any reaction description into the text box and click **Extract entities**.
Three quick examples are provided at the bottom.

## License

Training data (ChEMU 2020) is licensed under CC BY-NC 3.0; this demo and
the underlying model are therefore for **non-commercial research use only**.