DWIE Document-Level Entity Annotation and Linking

# DWIE: Document-Level Entity Mention Annotation and Linking
# Based on: Klim Zaporojets, Johannes Deleu, Chris Develder, and Thomas
#   Demeester (2021), "DWIE: An entity-centric dataset for multi-task
#   document-level information extraction." Information Processing &
#   Management 58(4), 102563.
#   DOI: 10.1016/j.ipm.2021.102563 — arXiv: https://arxiv.org/abs/2009.12626
#   Dataset: https://github.com/klimzaporojets/DWIE
#
# Task: mark every entity mention in a full news article, assign an entity
# type, and link the mention to its Wikipedia/Wikidata entry via the entity
# search modal. In DWIE, links are assigned at the level of entity clusters
# (all coreferent mentions of an entity share one KB link, Wikipedia version
# 2018-11-15), and entities absent from the knowledge base are left as NIL.
# Deliberate simplifications: DWIE's 311 hierarchical, multi-label entity
# types are collapsed to seven top-level types, and linking is performed per
# mention (annotators keep cluster consistency by linking every mention of
# the same entity to the same KB entry, or leaving all of them unlinked for
# NIL entities).
#
# Annotation instructions adapted from Section 3 of the paper; the paper does
# not publish the verbatim annotator guidelines.

annotation_task_name: "DWIE Document-Level Entity Linking"
task_dir: "."

data_files:
  - sample-data.json

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

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

port: 8000
server_name: localhost

annotation_instructions: |
  You are shown a complete news article. Mark **every mention** of an entity in
  the article — an entity can be any physical or abstract object, typically
  denoted by a proper noun (people, organizations, locations, events,
  ethnicities and nationalities, numeric values, and other named objects).

  For each mention:

  1. Highlight the mention text and assign the entity type that best describes
     it (PERSON, ORGANIZATION, LOCATION, EVENT, ETHNICITY, VALUE, or MISC).
  2. Open the link icon on the span and search Wikidata for the entity the
     mention refers to. Link the mention to the matching entry.
  3. Annotate at the level of the underlying entity, not the surface string:
     all mentions of the same real-world entity (e.g. "Angela Merkel", "the
     chancellor", "Merkel") must be linked to the **same** knowledge-base
     entry.
  4. If the entity does not exist in the knowledge base, leave the mention
     unlinked (a NIL entity). Do this consistently for every mention of that
     entity.

  Judge each mention in the context of the whole document — earlier mentions
  often disambiguate later, shorter ones.

annotation_schemes:
  - annotation_type: span
    name: entity_mentions
    description: "Mark every entity mention, assign its type, and link it to the knowledge base"
    labels:
      - name: PERSON
        tooltip: "Individual people, named or referred to by role"
      - name: ORGANIZATION
        tooltip: "Companies, institutions, governments, agencies, teams"
      - name: LOCATION
        tooltip: "Countries, cities, regions, geographic features, facilities"
      - name: EVENT
        tooltip: "Named events such as elections, wars, tournaments, disasters"
      - name: ETHNICITY
        tooltip: "Ethnic groups, nationalities, and peoples"
      - name: VALUE
        tooltip: "Numeric and temporal values: money, quantities, dates"
      - name: MISC
        tooltip: "Other named entities: products, works, laws, languages, etc."
    entity_linking:
      enabled: true
      auto_search: true
      required: false
      knowledge_bases:
        - name: wikidata
          type: wikidata
          language: en

html_layout: |
  <div style="padding: 15px; max-width: 860px; margin: auto;">
    <div style="background: #f1f5f9; border: 1px solid #cbd5e1; border-radius: 8px; padding: 6px 12px; margin-bottom: 8px; font-size: 13px; color: #475569;">
      <strong>Article:</strong> {{title}}
    </div>
    <div style="font-size: 16px; line-height: 1.8;">{{text}}</div>
  </div>

allow_all_users: true
instances_per_annotator: 30
annotation_per_instance: 2
allow_skip: true