GeoEDdA: A Gold Standard Dataset for Geo-semantic Annotation of Diderot & d'Alembert's Encyclopédie - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2024

GeoEDdA: A Gold Standard Dataset for Geo-semantic Annotation of Diderot & d'Alembert's Encyclopédie

Résumé

This paper describes the methodology for creating GeoEDdA, a gold standard dataset of geo-semantic annotations from entries in Diderot and d’Alembert’s eighteenth-century Encyclopédie. Aiming to explore spatial information beyond toponyms identified with the commonly used Named Entity Recognition (NER) task, we test the newer span categorization task as an approach for retrieving complex references to places, generic spatial terms, other entities, and relations. We test an active learning method, using the Prodigy web-based tool to iteratively train a machine learning span categorization model. The resulting dataset includes labeled spans from 2,200 paragraphs. As a preliminary experiment, a custom spaCy spancat model demonstrates strong overall performance, achieving an F-score of 86.42%. Evaluations for each span category reveal strengths in recognizing spatial entities and persons (including nominal entities, named entities and nested entities).
Fichier principal
Vignette du fichier
GeoExT24_GeoEDdA_camera-ready_Moncla.pdf (787.46 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Licence : CC BY - Paternité

Dates et versions

hal-04511909 , version 1 (19-03-2024)

Licence

Paternité

Identifiants

  • HAL Id : hal-04511909 , version 1

Citer

Ludovic Moncla, Denis Vigier, Katherine Mcdonough. GeoEDdA: A Gold Standard Dataset for Geo-semantic Annotation of Diderot & d'Alembert's Encyclopédie. Second International Workshop on Geographic Information Extraction from Texts (GeoExT) to be held at the 46th European Conference on Information Retrieval (ECIR 2024), Mar 2024, Glasgow, United Kingdom. ⟨hal-04511909⟩
2 Consultations
3 Téléchargements

Partager

Gmail Facebook X LinkedIn More