alchemy-ner

You can directly create a project based on this template to develop your own NER models.

To use alchemy, please initialize this submodule after cloning this repo by:

git submodule update --init --recursive

We implement the src.task.ner.NerTask based on AlchemyTask. Within NerTask, meta information, data processing, output processing and evaluation are defined.

We also implement some baselines for the convenience of comparison.

Implemented baselines

• Tagging model (src.models.tagger.Tagger): A simple bert-base sequence tagger with BIO taggin scheme.
• BERT-CRF model (src.models.tagger.CRFTagger): A simple bert-base CRF tagger with BIO taggin scheme.
• Biaffine (src.models.biaffine.Biaffine): A span-based model to detect nested named entities. [code] [paper].
• Propose-and-Refine (src.models.pnr.PnRNet): A two-stage set prediction model to detect nested named entities. [code] [paper].

Meta information

Data processing

Output processing

Evaluation