Generating a Question Answering Dataset About Geographic Changes in a Knowledge Graph

verfasst von: Michalis Mitsios, Dharmen Punjani, Sara Abdollahi, Simon Gottschalk, Eleni Tsalapati, Elena Demidova, Manolis Koubarakis
Abstract: Most studies on semantic question answering (QA) are predominantly focused on encyclopedic knowledge graphs like DBpedia and Wikidata. These studies cover, if at all, the spatial and temporal characteristics of geospatial entities in isolation, not addressing them simultaneously. In this paper, we introduce a pipeline for creating question answering datasets for evaluating the reasoning capabilities of QA models in the context of geographic changes over time. This pipeline generates questions, GeoSPARQL queries, and corresponding answers by leveraging subgraph and query template extraction techniques. We exemplify this pipeline with the creation of the GeoChangesQA dataset with questions over a knowledge graph of US counties and states and their changes from 1629 to 2000. By evaluating GeoChangesQA using a Transformer-based model, we demonstrate that historical geospatial questions pose a substantial challenge for semantic question answering.
Organisationseinheit(en): Forschungszentrum L3S
Externe Organisation(en): University of Athens
Université Jean Monnet Saint-Étienne
Deutsche Akademie der Technikwissenschaften (acatech)
Athens Technology Center S.A.
Rheinische Friedrich-Wilhelms-Universität Bonn
Lamarr-Institut für Maschinelles Lernen und Künstliche Intelligenz
Archimedes/Athena RC
Typ: Aufsatz in Konferenzband
Seiten: 471-489
Anzahl der Seiten: 19
Publikationsdatum: 2025
Publikationsstatus: Veröffentlicht
Peer-reviewed: Ja
ASJC Scopus Sachgebiete: Theoretische Informatik, Allgemeine Computerwissenschaft
Elektronische Version(en): https://doi.org/10.1007/978-3-031-77792-9_28 (Zugang: Geschlossen)

BibTeX

@inproceedings{f6f9ab36ae914310a249e767f914dd43,
title = "Generating a Question Answering Dataset About Geographic Changes in a Knowledge Graph",
abstract = "Most studies on semantic question answering (QA) are predominantly focused on encyclopedic knowledge graphs like DBpedia and Wikidata. These studies cover, if at all, the spatial and temporal characteristics of geospatial entities in isolation, not addressing them simultaneously. In this paper, we introduce a pipeline for creating question answering datasets for evaluating the reasoning capabilities of QA models in the context of geographic changes over time. This pipeline generates questions, GeoSPARQL queries, and corresponding answers by leveraging subgraph and query template extraction techniques. We exemplify this pipeline with the creation of the GeoChangesQA dataset with questions over a knowledge graph of US counties and states and their changes from 1629 to 2000. By evaluating GeoChangesQA using a Transformer-based model, we demonstrate that historical geospatial questions pose a substantial challenge for semantic question answering.",
author = "Michalis Mitsios and Dharmen Punjani and Sara Abdollahi and Simon Gottschalk and Eleni Tsalapati and Elena Demidova and Manolis Koubarakis",
note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.; 24th International Conference on Knowledge Engineering and Knowledge Management, EKAW 2024 ; Conference date: 26-11-2024 Through 28-11-2024",
year = "2025",
doi = "10.1007/978-3-031-77792-9_28",
language = "English",
isbn = "9783031777912",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Science and Business Media Deutschland GmbH",
pages = "471--489",
editor = "Mehwish Alam and Marco Rospocher and {van Erp}, Marieke and Laura Hollink and Gesese, {Genet Asefa}",
booktitle = "Knowledge Engineering and Knowledge Management",
address = "Germany",
}