@inproceedings{santos-etal-2026-evaluating,
title = "Evaluating {LLM}-based Triple Extraction for Knowledge Graph Fact-Checking in {P}ortuguese",
author = "Santos, Roney Lira de Sales and
Santos, Lucas dos and
Souza, Jo{\~a}o Pedro Holanda",
editor = "Barbosa, Bryan Khelven da Silva and
Paes, Aline and
Felippo, Ariani Di",
booktitle = "Proceedings of the 17th {B}razilian Symposium in Information and Human Language Technology",
month = oct,
year = "2026",
address = "Cuiab{\'a}, Mato Grosso, Brazil",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.stil-1.30/",
doi = "10.5753/stil.2026.26618",
pages = "361--374",
abstract = "Automatic fake news detection in Portuguese is still often treated as a text classification task, without explicitly representing the factual veracity of the claims. In this work, we evaluate LLM-based triple extraction in a knowledge graph (KG)-based fact-checking system. We replace the Open Information Extraction component of a previous approach with triples generated by Sabi{\'a} 4, while keeping the remaining pipeline unchanged. The graph is built only from triples extracted from true news articles and is used as factual support to verify new instances. The experiments use true and fake news articles across four evaluation settings. In the closed setting with the complete KG, LLM-based extraction achieves an F1 score of 0.9992, outperforming the OIE-based configuration. However, in more restrictive evaluation settings, expressive triples require entity normalization, relation standardization, and semantic consolidation to provide factual support beyond direct evidence."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="santos-etal-2026-evaluating">
<titleInfo>
<title>Evaluating LLM-based Triple Extraction for Knowledge Graph Fact-Checking in Portuguese</title>
</titleInfo>
<name type="personal">
<namePart type="given">Roney</namePart>
<namePart type="given">Lira</namePart>
<namePart type="given">de</namePart>
<namePart type="given">Sales</namePart>
<namePart type="family">Santos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lucas</namePart>
<namePart type="given">dos</namePart>
<namePart type="family">Santos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">João</namePart>
<namePart type="given">Pedro</namePart>
<namePart type="given">Holanda</namePart>
<namePart type="family">Souza</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 17th Brazilian Symposium in Information and Human Language Technology</title>
</titleInfo>
<name type="personal">
<namePart type="given">Bryan</namePart>
<namePart type="given">Khelven</namePart>
<namePart type="given">da</namePart>
<namePart type="given">Silva</namePart>
<namePart type="family">Barbosa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aline</namePart>
<namePart type="family">Paes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ariani</namePart>
<namePart type="given">Di</namePart>
<namePart type="family">Felippo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Cuiabá, Mato Grosso, Brazil</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Automatic fake news detection in Portuguese is still often treated as a text classification task, without explicitly representing the factual veracity of the claims. In this work, we evaluate LLM-based triple extraction in a knowledge graph (KG)-based fact-checking system. We replace the Open Information Extraction component of a previous approach with triples generated by Sabiá 4, while keeping the remaining pipeline unchanged. The graph is built only from triples extracted from true news articles and is used as factual support to verify new instances. The experiments use true and fake news articles across four evaluation settings. In the closed setting with the complete KG, LLM-based extraction achieves an F1 score of 0.9992, outperforming the OIE-based configuration. However, in more restrictive evaluation settings, expressive triples require entity normalization, relation standardization, and semantic consolidation to provide factual support beyond direct evidence.</abstract>
<identifier type="citekey">santos-etal-2026-evaluating</identifier>
<identifier type="doi">10.5753/stil.2026.26618</identifier>
<location>
<url>https://aclanthology.org/2026.stil-1.30/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>361</start>
<end>374</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Evaluating LLM-based Triple Extraction for Knowledge Graph Fact-Checking in Portuguese
%A Santos, Roney Lira de Sales
%A Santos, Lucas dos
%A Souza, João Pedro Holanda
%Y Barbosa, Bryan Khelven da Silva
%Y Paes, Aline
%Y Felippo, Ariani Di
%S Proceedings of the 17th Brazilian Symposium in Information and Human Language Technology
%D 2026
%8 October
%I Association for Computational Linguistics
%C Cuiabá, Mato Grosso, Brazil
%F santos-etal-2026-evaluating
%X Automatic fake news detection in Portuguese is still often treated as a text classification task, without explicitly representing the factual veracity of the claims. In this work, we evaluate LLM-based triple extraction in a knowledge graph (KG)-based fact-checking system. We replace the Open Information Extraction component of a previous approach with triples generated by Sabiá 4, while keeping the remaining pipeline unchanged. The graph is built only from triples extracted from true news articles and is used as factual support to verify new instances. The experiments use true and fake news articles across four evaluation settings. In the closed setting with the complete KG, LLM-based extraction achieves an F1 score of 0.9992, outperforming the OIE-based configuration. However, in more restrictive evaluation settings, expressive triples require entity normalization, relation standardization, and semantic consolidation to provide factual support beyond direct evidence.
%R 10.5753/stil.2026.26618
%U https://aclanthology.org/2026.stil-1.30/
%U https://doi.org/10.5753/stil.2026.26618
%P 361-374
Markdown (Informal)
[Evaluating LLM-based Triple Extraction for Knowledge Graph Fact-Checking in Portuguese](https://aclanthology.org/2026.stil-1.30/) (Santos et al., STIL 2026)
ACL