NER models from "A Benchmark of Named Entity Recognition Approaches in Historical Documents Application to 19th Century French Directories", DAS22 workshop on document analysis systems.

About
NER models created for the evaluation of Optical Character Recognition (OCR) and Named Entity Recognition (NER) on 19th century French documents, presented in

Abadie, N., Carlinet, E., Chazalon, J., Duménieu, B. (2022). A Benchmark of Named Entity Recognition  Approaches in Historical Documents Application to 19𝑡ℎ Century French Directories. In: Uchida, S., Barney, E., Eglin, V. (eds) Document Analysis Systems. DAS 2022. Lecture Notes in Computer Science, vol 13237. Springer, Cham. https://doi.org/10.1007/978-3-031-06555-2_30

All models can be replicated using the notebooks available on the GitHub repository https://github.com/soduco/paper-ner-bench-das22.

NER models

All models follow the same naming scheme `das22-{step}-{name}` where:
- {step} designate the step in the NER evaluation pipeline where this model has been trained, correspoding to a notebook in `src/ner`.
- {name} is the name of the model. All CamemBERT models are also available on [the HuggingFace hub](https://huggingface.co/HueyNemud).

Note that not all models used in the paper are published here. Specifically, this deposit does not store the models trained on subsets of the gold dataset for the experiment #1 but they can be trained using the notebooks available in `src/ner`.

- **das22-10-camembert_pretrained**: an [off-the-shelf CamemBERT model](https://huggingface.co/Jean-Baptiste/camembert-ner) available on the HuggingFace hub, pre-trained on 845k raw directory entries using a masked language modeling task. This model is intended for further training for NER on the gold (reference and OCR) annotated data. This model can be trained in `10-camembert_pretraining.ipynb`.
 - **das22-20-spacy_best_6373**: a SpaCy NER pipeline trained on the full gold dataset for experiment #1 containing 6373 entries. This model can be trained in `20-experiment_1.ipynb`.
 - **das22-22-camembert_6373**: an [off-the-shelf CamemBERT model](https://huggingface.co/Jean-Baptiste/camembert-ner) fine-tuned for NER on our gold dataset.
 - **das22-22-camembert_pretrained_6373**: the pretrained model `das22-10-camembert_pretrained`, fine-tuned for NER on our gold dataset.
 - **das22-4\*-camembert-\***: CamemBERT pretrained or "simple" fine-tuned on the reference gold dataset or on the noisy OCR gold with projected annotations.