Hate Speech Detection Against the Mexican Spanish LGBTQ+ Community Using BERT-based Transformers

@Article {
title = {Hate Speech Detection Against the Mexican Spanish LGBTQ+ Community Using BERT-based Transformers},
keywds = {
Deep Learning; Hate Speech; NLP; Text Classification
}
abstract = {

In this paper we present our approach to the HOMO-MEX task: Hate speech detection in Online Messages directed tOwards the MEXican spanish speaking LGBTQ+ population. We present our results for both Track 1: Hate speech detection track, in which the aim is to indicate whether a set of tweets exhibit LGBT+phobic content or not, and Track 2: Fine-grained hate speech detection track (Multi-labeled), in which the tweets labeled as LGBT+phobic need to be classified according to the type of LGBT+phobia they show. We utilized both classical machine learning and Transformer-based deep learning models focused on BERT-like architectures to tackle both tracks. The model that achieved the best results in terms of F1-Score (0.84 in Track 1) and macro-average F1-Score (0.68 in Track 2) was robertuito-base-uncased. With this model our team reached the 2nd position in both tracks.

}
date = {2023-09-26},
}