License: 
Creative Commons Attribution 4.0 International license (CC BY 4.0)
When quoting this document, please refer to the following
DOI: 10.4230/OASIcs.SLATE.2023.1
URN: urn:nbn:de:0030-drops-185155
URL: http://dagstuhl.sunsite.rwth-aachen.de/volltexte/2023/18515/
Faria, Bruno ;
Perdigão, Dylan ;
Gonçalo Oliveira, Hugo
Question Answering over Linked Data with GPT-3
Abstract
This paper explores GPT-3 for answering natural language questions over Linked Data. Different engines of the model and different approaches are adopted for answering questions in the QALD-9 dataset, namely: zero and few-shot SPARQL generation, as well as fine-tuning in the training portion of the dataset. Answers retrieved by the generated queries and answers generated directly by the model are also compared. Overall results are generally poor, but several insights are provided on using GPT-3 for the proposed task.
BibTeX - Entry
@InProceedings{faria_et_al:OASIcs.SLATE.2023.1,
author = {Faria, Bruno and Perdig\~{a}o, Dylan and Gon\c{c}alo Oliveira, Hugo},
title = {{Question Answering over Linked Data with GPT-3}},
booktitle = {12th Symposium on Languages, Applications and Technologies (SLATE 2023)},
pages = {1:1--1:15},
series = {Open Access Series in Informatics (OASIcs)},
ISBN = {978-3-95977-291-4},
ISSN = {2190-6807},
year = {2023},
volume = {113},
editor = {Sim\~{o}es, Alberto and Ber\'{o}n, Mario Marcelo and Portela, Filipe},
publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
address = {Dagstuhl, Germany},
URL = {https://drops.dagstuhl.de/opus/volltexte/2023/18515},
URN = {urn:nbn:de:0030-drops-185155},
doi = {10.4230/OASIcs.SLATE.2023.1},
annote = {Keywords: SPARQL Generation, Prompt Engineering, Few-Shot Learning, Question Answering, GPT-3}
}
