Aggression and Hate in Spanish Text Messages, Identification Using a Pre-Trained Transformer Model. (#1077)

Authors

Espin-Riofrio, César

Rodríguez Soria, Helen

San Martín Torres, Josué

Mendoza Morán, Verónica

Cruz Chóez, Angélica

Montejo-Ráez, Arturo

Abstract

Nowadays, social networks have given rise to the free expression of opinions and thoughts in real time, however, this can also lead to negative interactions, such as bullying, discrimination and other aggressive and hateful behaviour. To address this issue, different Natural Language Processing (NLP) methods and techniques exist. In this paper, a quasi-experimental investigation was carried out using the pre-trained Pysentimiento Transformer model to detect the presence of aggression and hate in Spanish text messages on the social network Twitter. Data extraction and processing tools were used to ensure the quality of the data before it was run through the model. A web interface was also designed to present the information obtained through graphs and tables, allowing a clear assessment of the detection of aggressive and hateful content in text messages through various analysis criteria. It is shown that it is possible to detect aggression and hatred in text messages using a Transformer model pre-trained for the task, and use it to create systems or applications that detect and quantify these symptoms in messages written by people.