Skip to main content

HEART-NLP-UJA: PLN CENTRADO EN EL SER HUMANO Y BASADA EN EL CONOCIMIENTO PARA GARANTIZAR LA CALIDAD, LA INTEGRIDAD Y LA TRANSPARENCIA EN UN ECOSISTEMA DIGITAL CONFIABLE

PID2024-156263OB-C21

•
Funding agency name Agencia Estatal de Investigación
Funding agency acronym AEI
Program Programa Estatal para la Investigación y el Desarrollo Experimental
Subprogram Subprograma Estatal de Generación de Conocimiento Científico-Técnico y Desarrollo Experimental
Call Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Call year 2024
State Plan 2024-2027 Plan Estatal de Investigación Científica y Técnica y de Innovación
Beneficiary centre UNIVERSIDAD DE JAEN

Resultados relacionados

Resultados totales (Incluyendo duplicados): 1
Encontrada(s) 1 página(s)

RUA. Repositorio Institucional de la Universidad de Alicante
Publication Otros (Other)

Nuanced Toxicity Detection in Spanish: A New Corpus and Benchmark Study

  • Mármol-Romero, Alba María
  • Sepúlveda-Torres, Robiert
  • Saquete Boró, Estela
  • Martín Valdivia, María Teresa
  • Ureña López, Luis Alfonso
  • Sepúlveda-Torres, Robiert|||/items/60f24ee2-fe45-4b5c-a70c-c025ed1b5e4a
  • Saquete Boró, Estela|||/items/452393f2-fb8d-4840-b797-4b65a4bc9a87
The rise of toxic content on digital platforms has intensified the demand for automatic moderation tools. While English has benefited from large-scale annotated corpora, Spanish remains under-resourced, particularly for nuanced cases of toxicity such as irony, sarcasm, or indirect aggression. We present an extended version of the NECOS-TOX corpus, comprising 4,011 Spanish comments collected from 16 major news outlets. Each comment is annotated across three levels of toxicity (Non-Toxic, Slightly Toxic, and Toxic), following an iterative annotation protocol that achieved substantial inter-annotator agreement (k = 0.74). To reduce annotation costs while maintaining quality, we employed a human-in-the-loop active learning strategy, with manual correction of model pre-labels. We benchmarked the dataset with traditional machine learning (ML) methods, domain-specific transformers, and instruction-tuned large language models (LLMs). Results show that compact encoder models (e.g., RoBERTa-base-bne, 125M parameters) perform on par with much larger models (e.g., LLaMA-3.1-8B), underscoring the value of in-domain adaptation over raw scale. Our error analysis highlights persistent challenges in distinguishing subtle forms of toxicity, especially sarcasm and implicit insults, and reveals entity-related biases that motivate anonymization strategies. The dataset and trained models are released publicly.