metadata

license: cc-by-nc-4.0
language:
  - fr
metrics:
  - accuracy
  - recall
  - precision
  - f1
library_name: spacy
pipeline_tag: token-classification
tags:
  - spacy
  - token-classification
model-index:
  - name: fr_spacy_custom_spancat_edda
    results:
      - task:
          name: spancat
          type: span-classification
        metrics:
          - name: Span Precision
            type: precision
            value: 0.948
          - name: Span Recall
            type: recall
            value: 0.849
          - name: Span F1 Score
            type: f_score
            value: 0.896

spaCy custom spancat for Diderot & d’Alembert’s Encyclopédie entries

This model identify and classify spans of text for named entities (Spatial, Person and MISC), nested named entities (Spatial, Person and MISC), spatial relations and others from French encyclopaedic entries.

The spans detected by this model are:

NC-Spatial:
NP-Spatial:
ENE-Spatial:
Relation: spatial relations, e.g.
Latlong: geographic coordinates, e.g.
NC-Person:
NP-Person:
ENE-Person:
NP-Misc:
ENE-Misc:
Head:
Domain-Mark: , e.g. Géographie, Histoire

Model Details

Model Description

Developed by: Ludovic Moncla, Katherine McDonough and Denis Vigier
Model type: spaCy Span Categorization
spaCy: >=3.7.2,<3.8.0
Components: tok2vec, spancat
Language(s) (NLP): French
License: cc-by-nc-4.0

Model Sources

Repository: https://github.com/GEODE-project/ner-spancat-edda

Uses

This model can be used to extract entities from any text that are Paeleoecology related or tangential. Potential uses include identifying unique SITE names in research papers in other domains.

Bias, Risks, and Limitations

This model was trained entirely on French encyclopaedic entries and will likely not perform well on text in other languages. Also, the paragraphs used to train the model were chosen based on being already present in the Neotoma database and therefore may have selection bias as they represent what is already known to be relevant to Neotoma and may not correctly manage new, previously missed articles.

How to Get Started with the Model

Use the code below to get started with the model.

pip install https://huggingface.co/GEODE/fr_spacy_custom_spancat_edda/resolve/main/fr_spacy_custom_spancat_edda-any-py3-none-any.whl

# Using spacy.load().
import spacy
nlp = spacy.load("fr_spacy_custom_spancat_edda")

# Importing as module.
import fr_spacy_custom_spancat_edda
nlp = fr_spacy_custom_spancat_edda.load()
doc = nlp("* ALBI, (Géog.) ville de France, capitale de  l'Albigeois, dans le haut Languedoc : elle est sur le Tarn. Long. 19. 49. lat. 43. 55. 44.")

spans = []

for span in doc.spans['sc']:
  print(span)
  spans.append({
      "start": span.start_char,
      "end": span.end_char,
      "labels": [span.label_],
      "text": span.text
  })

print(spans)

# Output
[{'start': 2, 'end': 6, 'labels': ['Head'], 'text': 'ALBI'},
 {'start': 16, 'end': 21, 'labels': ['NC-Spatial'], 'text': 'ville'},
 {'start': 25, 'end': 31, 'labels': ['NP-Spatial'], 'text': 'France'},
 {'start': 33, 'end': 41, 'labels': ['NC-Spatial'], 'text': 'capitale'},
 {'start': 59, 'end': 63, 'labels': ['Relation'], 'text': 'dans'},
 {'start': 93, 'end': 96, 'labels': ['Relation'], 'text': 'sur'},
 {'start': 9, 'end': 14, 'labels': ['Domain-mark'], 'text': 'Géog.'},
 {'start': 46, 'end': 57, 'labels': ['NP-Spatial'], 'text': "l'Albigeois"},
 {'start': 97, 'end': 104, 'labels': ['NP-Spatial'], 'text': 'le Tarn'},
 {'start': 16,
  'end': 31,
  'labels': ['ENE-Spatial'],
  'text': 'ville de France'},
 {'start': 64,
  'end': 81,
  'labels': ['NP-Spatial'],
  'text': 'le haut Languedoc'},
 {'start': 33,
  'end': 57,
  'labels': ['ENE-Spatial'],
  'text': "capitale de  l'Albigeois"},
 {'start': 33,
  'end': 81,
  'labels': ['ENE-Spatial'],
  'text': "capitale de  l'Albigeois, dans le haut Languedoc"},
 {'start': 16,
  'end': 81,
  'labels': ['ENE-Spatial'],
  'text': "ville de France, capitale de  l'Albigeois, dans le haut Languedoc"}]

Training Details

Training Data

The model was trained using a set of 2200 paragraphs randomly selected out of xx Encyclopédie's entries. All paragraphs were written in French and are distributed as follows among the Encyclopédie knowledge domains:

Knowledge domain	Paragraphs
Géographie	1096
Histoire	259
Droit Jurisprudence	113
Physique	92
Métiers	92
Médecine	88
Philosophie	69
Histoire naturelle	65
Belles-lettres	65
Militaire	62
Commerce	48
Beaux-arts	44
Agriculture	36
Chasse	31
Religion	23
Musique	17

The spans/entities were labeled by the project team along with using pre-labelling with early models to speed up the labelling process. A train/val/test split was used. Validation and test sets are composed of 200 paragraphs each: 100 classified as 'Géographie' and 100 from another knowledge domain. The datasets have the following breakdown of tokens and spans/entities.

	Train	Validation	Test
Paragraphs	1800	200	200
Tokens
NC-Spatial
NP-Spatial
ENE-Spatial
Relation
Latlong
NC-Person
NP-Person
ENE-Person
NP-Misc
ENE-Misc
Head
Domain-Mark

Training Procedure

For full training details and results please see the GitHub repository: github.com/GEODE-project/ner-spancat-edda

Acknowledgement

Data courtesy the ARTFL Encyclopédie Project, University of Chicago.

The authors are grateful to the ASLAN project (ANR-10-LABX-0081) of the Université de Lyon, for its financial support within the French program "Investments for the Future" operated by the National Research Agency (ANR).