Open Access

Sergio Esteban Romero, Ricardo Kleinlein, Cristina Luna-Jiménez, Juan Manuel Montero, Fernando Fernández-Martínez

• Sadly, the presence of toxic messages on social networks, whether in the form of stereotypes, sarcasm, mockery, insult, inappropriate language, aggressiveness, intolerance, or typical of hate speech against immigrants and / or women, among others, is relatively frequent. This presence should not be ignored by the scientific community, since it is their responsibility to develop tools and systems that allow their automatic detection and elimination. In this paper, we present an exploratory analysis in which different deep learning (DL) models for the detection of toxic expressions have been evaluated on the DETOXISIberLEF 2021 challenge using the official release of the NewsCom-TOX corpus. Particularly, we compare traditional RNN and state-of-the-art transformer models. Our experiments confirmed that optimum performance can be obtained from transformer models. Specifically, top performance was achieved by fine tuning a BETO model (the pre-trained BERT model for the Spanish languageSadly, the presence of toxic messages on social networks, whether in the form of stereotypes, sarcasm, mockery, insult, inappropriate language, aggressiveness, intolerance, or typical of hate speech against immigrants and / or women, among others, is relatively frequent. This presence should not be ignored by the scientific community, since it is their responsibility to develop tools and systems that allow their automatic detection and elimination. In this paper, we present an exploratory analysis in which different deep learning (DL) models for the detection of toxic expressions have been evaluated on the DETOXISIberLEF 2021 challenge using the official release of the NewsCom-TOX corpus. Particularly, we compare traditional RNN and state-of-the-art transformer models. Our experiments confirmed that optimum performance can be obtained from transformer models. Specifically, top performance was achieved by fine tuning a BETO model (the pre-trained BERT model for the Spanish language from the Universidad de Chile) for the toxicity detection tasks. Another contribution of this analysis is the validation of the proposed method for adding task-specific vocabulary (new tokens) that could help to effectively extend the original vocabulary of the pre-trained models.…