Work

Combining profile features for offensiveness detection on Spanish social media

Year: 2025

Type: article

Abstract: The presence of offensive comments on social media has become a major issue in the use of the Internet. Natural Language Processing area has been studying methods and tools for the automatic detection of such content in social networks. The MeOffendEs task in the IberLEF evaluation campaign opened this problem to the research community for the Spanish language. This paper studies two methods for integrating certain contextual information with several transformer-based models. Our findings suggest that the contextual information provided by organizers does not contribute to a better prediction power. Instead, hyper-parameter search is a very important step in the general learning process, leading to systems outperforming the state-of-the-art in Spanish offensiveness detection. • Two methods for embedding contextual data to identify offensive content are tested. • Contextual information has no significant effect compared to Hyper-parameter search. • Hyper-parameter tuning beats state-of-the-art results in Spanish offensive detection.

Source: Expert Systems with Applications

DOI: 10.1016/j.eswa.2025.126705

Authors: M. Estrella Vallecillo-Rodríguez, Flor Miriam Plaza-del-Arco, Arturo Montejo-Ráez

Institutions: Universidad de Jaén, Bocconi University

Language: English

Topic: Hate Speech and Cyberbullying Detection

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $3,500

Funders: Ministerio de Sanidad, Consumo y Bienestar Social, Consejería de Economía, Conocimiento, Empresas y Universidad, Junta de Andalucía, Federación Española de Enfermedades Raras, Spanish National Plan for Scientific and Technical Research and Innovation