GeoEDdA-TopoRel

Dataset Description

• Authors: Bin Yang, Ludovic Moncla, Fabien Duchateau and Frédérique Laforest in the framework of the ECoDA and GEODE projects.
• Data source: ARTFL Encyclopédie Project, University of Chicago
• Git repository:
• Language: French
• License: cc-by-nc-4.0

Dataset Summary

The GeoEDdA-TopoRel dataset is structured in two parts:

1. JSON files containing 2,750 labeled entries from the Encyclopédie of Diderot and d’Alembert, all classified under Geography. This provides a set of features used at different stages of our knowledge graph construction pipeline, which is illustrated in the figure below.

2. TSV files for the matching evaluation. The files entities-selected-for-ground-truth.tsv contains a set of arount 16,000 entities (1,000 selected place named entities + 15,000 head entities). The file ground-truth-matching-valid-pairs.tsv contains correspondences for the 1,000 place named entities (645 correspondences).

Dataset JSON Structure

The dataset is provided as JSON files where each item follows the following structure:

{
  "volume": "17",
  "numero": "2303",
  "head": "TRÉVOUX",
  "text": "TRÉVOUX, (Géog. mod.) ancienne petite ville de France, <...> ",
  "entryType": "Place",
  "cardinality": "Single"
  "placeType": "City"
  "placeNames": [{"start": 48,  "end": 54, "text": "France", "label": "Country"}, {"start": 86, "end": 92, "text": "Dombes", "label": "Region"}, {"start": 118, "end": 126, "text": "la Saône", "label": "River"}, <...>],
  "spatialRelations": [{"start": 94, "end": 117, "text": "sur le bord oriental de la Saône", "label": "Adjacency"} ]
  "segmentedDescriptions": "[]"
}

Each data contains ten fields:

• volume: volume number of the entry from the Encyclopedie ARTFL dataset
• numero: number of the entry (within the volume) from the Encyclopedie ARTFL dataset
• head: entry name
• text: plain text of the entry
• "entryType": type of geography entry, possible values are: Place, Person, Other
• "cardinality": indicates whether the entry describes a single place or multiple places. Possible values: Single or Multiple
• "placeType": "City"
• "placeNames": [{"start": 48, "end": 54, "text": "France", "label": "Country"}, {"start": 86, "end": 92, "text": "Dombes", "label": "Region"}, {"start": 118, "end": 126, "text": "la Saône", "label": "River"}, <...>],
• "spatialRelations": [{"start": 94, "end": 117, "text": "sur le bord oriental de la Saône", "label": "Adjacency"} ]
• "segmentedDescriptions": list of segmented sub-entries if cardinality is equal to multiple, empty list otherwise.

Dataset TSV Structure

• The file entities-selected-for-ground-truth.tsv contains the following fields: book, volume, numero, segment_id, span_text, span_start, span_end, type, subtype, is_head, pref_label, aliases, uri_record, uri_entity
• The file ground-truth-matching-valid-pairs.tsv contains two fields, one for the URI of the source entity (named entity) and the second one for the corresponding target entity.

Acknowledgement

The authors thank the Computer Science Federation of Lyon (FIL) of the CNRS for financially supporting the ECoDA research project.