Few-NERD Fine-Grained Named Entity Recognition

# Few-NERD - Fine-Grained Named Entity Recognition (illustrative sample)
# Few-NERD: A Few-shot Named Entity Recognition Dataset
#
# Citation (ACL Anthology BibTeX):
#   @inproceedings{ding-etal-2021-nerd,
#       title = "Few-{NERD}: A Few-shot Named Entity Recognition Dataset",
#       author = "Ding, Ning and Xu, Guangwei and Chen, Yulin and Wang, Xiaobin and
#         Han, Xu and Xie, Pengjun and Zheng, Haitao and Liu, Zhiyuan",
#       booktitle = "Proceedings of the 59th Annual Meeting of the Association for
#         Computational Linguistics and the 11th International Joint Conference on
#         Natural Language Processing (Volume 1: Long Papers)",
#       month = aug, year = "2021", address = "Online",
#       publisher = "Association for Computational Linguistics",
#       url = "https://aclanthology.org/2021.acl-long.248/",
#       doi = "10.18653/v1/2021.acl-long.248", pages = "3198--3213"
#   }
# DOI: 10.18653/v1/2021.acl-long.248
# Dataset: https://github.com/thunlp/Few-NERD
#          https://huggingface.co/datasets/DFKI-SLT/few-nerd
# Licence: CC BY-SA 4.0 (Hugging Face dataset card). Because of the share-alike
# term, the sample sentences in sample-data.json are ILLUSTRATIVE. They were
# written for this design in the style of Wikipedia text and are not copied
# from Few-NERD.
#
# Protocol (paper, Section 4): "We carefully design an annotation schema of 8
# coarse-grained entity types and 66 fine-grained entity types". Each paragraph
# "is independently annotated by two well-trained annotators. Then, an
# experienced expert goes over the paragraph for possible wrong or omissive
# annotations, and make the final decision." Paper example: in "London is the
# fifth album by the British rock band Jesus Jones", London "should be annotated
# as an entity of Art-Music rather than Location-GPE".
#
# Label set (66 fine types, names as listed in the Hugging Face dataset card):
#   art:          broadcastprogram, film, music, other, painting, writtenart
#   building:     airport, hospital, hotel, library, other, restaurant,
#                 sportsfacility, theater
#   event:        attack/battle/war/militaryconflict, disaster, election, other,
#                 protest, sportsevent
#   location:     GPE, bodiesofwater, island, mountain, other, park,
#                 road/railway/highway/transit
#   organization: company, education, government/governmentagency,
#                 media/newspaper, other, politicalparty, religion,
#                 showorganization, sportsleague, sportsteam
#   other:        astronomything, award, biologything, chemicalthing, currency,
#                 disease, educationaldegree, god, language, law, livingthing,
#                 medical
#   person:       actor, artist/author, athlete, director, other, politician,
#                 scholar, soldier
#   product:      airplane, car, food, game, other, ship, software, train, weapon
#
# Adaptations:
#   - Potato span labels are flat, so the two-level hierarchy is encoded in the
#     label names. Each of the 66 labels is "<coarse>-<fine>" and carries its
#     coarse type's color, so the 8 coarse types stay readable on screen. A
#     separate coarse scheme plus a fine-type select would not tie the fine
#     type to a span, so it was not used.
#   - Labels carry no tooltips because the paper does not give per-type
#     definitions in the text this design quotes.
#   - num_annotators_per_item: 2 echoes the paper's two independent annotators.
#     The expert adjudication pass is not modeled.
#   - Sample items are 10 illustrative sentences, not dataset rows.
#
# Model training: the trainer is hf-token-classification. It fine-tunes a
# Hugging Face transformer to tag text spans from the labeled spans, and the
# run writes predicted spans back as prelabels on unlabeled sentences. Install
# it with: pip install 'potato-annotation[train-text]'. Open /admin/training
# with the admin key below, pick the trainer and the "entity_spans" scheme,
# and click Train now. It does not serve or deploy the model.

annotation_task_name: "Few-NERD - Fine-Grained Named Entity Recognition"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "text"

output_annotation_dir: "annotation_output/"
export_annotation_format: "jsonl"

port: 8000
require_password: false
admin_api_key: "demo-admin-key"

annotation_schemes:
  - annotation_type: span
    name: entity_spans
    description: "Highlight each named entity and choose its coarse-fine type"
    labels:
      # art
      - name: "art-broadcastprogram"
        color: "#8b5cf6"
      - name: "art-film"
        color: "#8b5cf6"
      - name: "art-music"
        color: "#8b5cf6"
      - name: "art-other"
        color: "#8b5cf6"
      - name: "art-painting"
        color: "#8b5cf6"
      - name: "art-writtenart"
        color: "#8b5cf6"
      # building
      - name: "building-airport"
        color: "#78716c"
      - name: "building-hospital"
        color: "#78716c"
      - name: "building-hotel"
        color: "#78716c"
      - name: "building-library"
        color: "#78716c"
      - name: "building-other"
        color: "#78716c"
      - name: "building-restaurant"
        color: "#78716c"
      - name: "building-sportsfacility"
        color: "#78716c"
      - name: "building-theater"
        color: "#78716c"
      # event
      - name: "event-attack/battle/war/militaryconflict"
        color: "#ef4444"
      - name: "event-disaster"
        color: "#ef4444"
      - name: "event-election"
        color: "#ef4444"
      - name: "event-other"
        color: "#ef4444"
      - name: "event-protest"
        color: "#ef4444"
      - name: "event-sportsevent"
        color: "#ef4444"
      # location
      - name: "location-GPE"
        color: "#22c55e"
      - name: "location-bodiesofwater"
        color: "#22c55e"
      - name: "location-island"
        color: "#22c55e"
      - name: "location-mountain"
        color: "#22c55e"
      - name: "location-other"
        color: "#22c55e"
      - name: "location-park"
        color: "#22c55e"
      - name: "location-road/railway/highway/transit"
        color: "#22c55e"
      # organization
      - name: "organization-company"
        color: "#f59e0b"
      - name: "organization-education"
        color: "#f59e0b"
      - name: "organization-government/governmentagency"
        color: "#f59e0b"
      - name: "organization-media/newspaper"
        color: "#f59e0b"
      - name: "organization-other"
        color: "#f59e0b"
      - name: "organization-politicalparty"
        color: "#f59e0b"
      - name: "organization-religion"
        color: "#f59e0b"
      - name: "organization-showorganization"
        color: "#f59e0b"
      - name: "organization-sportsleague"
        color: "#f59e0b"
      - name: "organization-sportsteam"
        color: "#f59e0b"
      # other
      - name: "other-astronomything"
        color: "#06b6d4"
      - name: "other-award"
        color: "#06b6d4"
      - name: "other-biologything"
        color: "#06b6d4"
      - name: "other-chemicalthing"
        color: "#06b6d4"
      - name: "other-currency"
        color: "#06b6d4"
      - name: "other-disease"
        color: "#06b6d4"
      - name: "other-educationaldegree"
        color: "#06b6d4"
      - name: "other-god"
        color: "#06b6d4"
      - name: "other-language"
        color: "#06b6d4"
      - name: "other-law"
        color: "#06b6d4"
      - name: "other-livingthing"
        color: "#06b6d4"
      - name: "other-medical"
        color: "#06b6d4"
      # person
      - name: "person-actor"
        color: "#3b82f6"
      - name: "person-artist/author"
        color: "#3b82f6"
      - name: "person-athlete"
        color: "#3b82f6"
      - name: "person-director"
        color: "#3b82f6"
      - name: "person-other"
        color: "#3b82f6"
      - name: "person-politician"
        color: "#3b82f6"
      - name: "person-scholar"
        color: "#3b82f6"
      - name: "person-soldier"
        color: "#3b82f6"
      # product
      - name: "product-airplane"
        color: "#ec4899"
      - name: "product-car"
        color: "#ec4899"
      - name: "product-food"
        color: "#ec4899"
      - name: "product-game"
        color: "#ec4899"
      - name: "product-other"
        color: "#ec4899"
      - name: "product-ship"
        color: "#ec4899"
      - name: "product-software"
        color: "#ec4899"
      - name: "product-train"
        color: "#ec4899"
      - name: "product-weapon"
        color: "#ec4899"

annotation_instructions: |
  Highlight every named entity in the sentence and choose the most specific
  type, written as coarse-fine (for example art-music or location-GPE).
  The eight coarse types are art, building, event, location, organization,
  other, person, and product. Each has fine types, and every coarse type
  except "other" has an "other" fine type (such as person-other) for
  entities that fit no listed fine type.
  Choose the type that fits the entity in context. In the paper's example,
  the album title "London" is art-music, not location-GPE.

user_config:
  allow_all_users: true
max_annotations_per_user: 50
num_annotators_per_item: 2

model_training:
  enabled: true
  writeback:
    schemas: ["entity_spans"]
    include_unlabeled: true
    min_confidence: 0.6