Macht: An Application Based on a Sentiment Analysis Model to the Identification of Messages in Spanish with Gender Violence Content on Twitter
Ivonne Soldevilla, Nahum Flores, Sebastian Tuesta · 2022
During confinement, the number of anonymous complaints of gender violence in social networks has increased, affecting society. This work presents an alternative to detect messages with offensive content to women who have gone through a process of violence automatically. The applied methodology considers the construction of a public dataset with 1042 tweets in Spanish tagged by 48 volunteers. The model considers the fine-tuning process to 3 pre-trained BERT models (SpanBERT, BETO, multilingualBERT), with which 252 experiments were carried out to find the model with the best performance, obtaining an Area Under the Curve of 0.9349 and precision of 0.9043. The research contributes with a new public data labeled in Spanish in 5 age ranges where anyone from anywhere in the world will access the application and test the performance of the model.