Revictimization
The responsibility of the communication platforms should be to objectively inform the public without attempting to influence their opinions, but so far, the press has been incapable of doing this. “Medios Revictimizantes” (2021) analysis shows that out of 1,300 reports of violence against women published by more than 20 media platforms from Argentina, 897 revictimized the sufferer women. Such approaches present dangerous interpretive frameworks that justify violent acts, perpetrate a misogynistic and discriminative culture, or even jeopardize judicial investigations.
Mexican media has also participated in revictimization cases by stating personal opinions in news articles of occurred femicides such as Lesvy Berlín, Debhani Escobar or Ingrid Escamilla cases, just to name a few. Some example of misogynistic opinions in news articles described at “Fue Su Culpa: Por Qué Es Vital Evitar La Revictimización En Los Casos de Feminicidio” (2022) are “It was her fault,” “Her parents should have picked her up,” “Surely she set herself on fire,” “Look how she was dressed up that day,” “She might have fallen and bumped her head” .Through this comments, the media portrays women victims of violence as stigmatized, guilty, untrustworthy, or sexualized. This method of data governance creates a sense of revictimization in the survivors’ or witnesses’ families and fosters another form of violence against women: media violence.
Existing approaches
Automatic detection of misogyny is a complex Sentiment Analysis task. It is difficult to distinguish it from sexist or aggressive language towards women, given that it can be subtle and dependent on the social context where it takes place.
First analyses of this problem were focused on English texts Pamungkas and Patti (2020), and the small amount of Spanish texts research needs to distinguish the text origin between Spanish-spoken countries García-Díaz (2021). Therefore, cultural and linguistic differences can complicate the classification. Recently, multiple approaches such as Molina Villegas (2022) , Aldana-Bobadilla (2021) and Vera Lagos (2021) have generated data sets of Spanish texts from México labeled as misogynistic or not. However most of the datasets are based on tweets, reducing the misogynistic detection to short texts since tweets can only hold 280 characters.
Finally, Sentiment Analysis detection of news articles has been performed in the past Chen (2018), focusing on “negative” or “positive” sentiment. However, no detection of misogyny in news articles has been performed in any language.
Proposal
The objective of this work is to train multiple supervised models to classify misogynistic and non-misogynistic texts obtained from news articles from Mexico in Spanish and analyzing their results. Which will be evaluated in terms of accuracy, precision, and a deep understanding of how the classifier detects both subtle and apparent misogyny.
Acknowledgment
We want to acknowledge Prof. Trevor Adrianesse for his advice and support on the design and development of this project.
We also want to express our gratitute towards PhD. Alejandro Molina Villegas and PhD. Edwyn Aldana Bobadilla for sharing their data and expertise among the automatic detection of misogyny in Spanish texts.