Gender biases and hate speech: Promoters and targets in the Argentinean political context
Analysis of Gender Bias and Hate Speech in the Argentinean Political Context
Introduction
Hate speech has emerged as a significant concern due to its detrimental effects on society. It harms individuals, contributes to the radicalization of political stances, and can escalate to physical violence against marginalized groups. International organizations have implemented strategies to combat and eradicate online hate speech. Hate speech is any form of public discourse that promotes, incites, or legitimizes discrimination, dehumanization, and violence based on an individual's or group's identity characteristics, such as religion, ethnicity, nationality, gender, or sexual orientation, among others.
While hate speech has a long history, modern communication technologies such as social media have amplified its reach and impact. Twitter has become a prime platform for the spread and dissemination of hate speech, leading to numerous studies focused on detecting hate speech on the platform. Such discourse can also serve political purposes. Studies have shown how far-right groups employ hate speech as a tactic to advance campaigns against women's rights and the rights of marginalized groups while discrediting public figures. Feminist activists are a frequent target of this type of digital violence.
Moreover, there is a strong correlation between the dissemination of hate speech and the rise of certain right-wing political figures on social media platforms. The establishment of political factions that discredit and target social movements fosters the amplification of hate speech, granting it credibility and social acceptance within some online communities.
Theoretical Framework
This study employs a feminist and critical race theory framework to analyze gender bias and hate speech within the political context of Argentina. Gender bias stems from sexism and is evident in online vitriol as sexist hate speech, a form of hateful rhetoric targeting women because of their gender, which can impact their physical integrity and inclusion. Additionally, hate speech serves as a tool to marginalize and oppress historically disenfranchised groups, including women, people of color, trans people, and lesbian, gay, bisexual, and transgender (LGBT) people.
Furthermore, hate speech emerges as a key element in the rise of right-wing populist governments. It allows these groups to gain visibility and influence public discourse, especially on social networks where polarizing views often yield greater significance. A striking example of this phenomenon is seen in the case of Brazil's former president, Jair Bolsonaro. Text-mining analysis of his Facebook page revealed expressions of partisan political animosity, misogyny, LGBT-phobia, and xenophobia.
Data and Methods
Dataset Construction
This study analyzes tweets related to Argentine political figures to examine their involvement in hate speech as both recipients and senders. Political figures were chosen to represent the main political parties within the Argentine electoral context, with a focus on their prominence in past elections or significant positions held in executive or legislative branches or their representation of a political party within their respective electoral blocs.
The final dataset comprises approximately 259,000 tweets mentioning the political figures and over 44,000 tweets sent by the figures themselves. Note that some figures had fewer original tweets available, potentially impacting analysis pertaining to hate speech reception.
Hate Speech Classification Methods
To determine hate speech within the gathered tweets, we utilized a Python toolkit that leverages pre-trained language models to classify hate speech content across various categories, including targeting women, LGBTQ+, racism, and politics. Each tweet was assigned a score between 0 and 1 for each category, with a value exceeding 0.5 indicating the presence of hate speech in the specific aspect.
Detecting User Political Orientation
To gain insight into the directionality of hate speech, we inferred the political orientation of commenting users based on user descriptions. Twitter bios serve as a proxy for an individual's sense of identity, and when users incorporate political affiliations, it suggests a strong connection between the views and their identity. Regular expression rules were used to classify political orientations into six categories.
Limitations
Certain limitations and potential drawbacks of the proposed methodology should be acknowledged. First, the analyzed social media data is biased toward a certain demographic: Twitter users in Argentina tend to be concentrated in high and upper-middle socioeconomic levels, younger age groups (12 to 29 years old), and larger cities.
In addition, certain challenges of the text classification models and natural language processing (NLP) in general add another layer of complexity. Issues such as difficulties in identifying irony, ambiguous words, slang, and informal language pose challenges, particularly in the context of broader political discourse and on social media platforms. The use of automated hate speech detection algorithms for content moderation also raises ethical concerns regarding the potential censorship of messages that may be misclassified as hateful, as well as imposing a definition of hate speech that might not be shared across users.
Results
Measuring Hate Speech
Analysis of hate speech detection results reveal that the average rate of hate speech reception by political leaders is significantly higher than the rate of hate speech dissemination. This discrepancy is attributed to the anonymity prevalent among social media users, unlike politicians, making them more susceptible to digital violence.
When disaggregating the data, four out of five politicians who receive the highest levels of hate speech are women. Furthermore, eight of the thirteen politicians who receive higher than average hate speech are women, while only five of the twelve politicians who receive less than average hate speech are women.
Concerning the dissemination of hate speech by political leaders, Espert, Grabois, Villarruel, Maslaton, and Ofelia Fernandez stand out as the primary proponents, with only Villarruel receiving less hateful mentions than average.
The Content of Online Digital Violence
Hate speech encompasses various forms of discrimination, and when analyzing the specific content targeting political figures, we find that women are more prone to attacks across almost all categories, except appearance, criminalization, and LGBTIQ topics. Notably, the "politics" category has the highest average percentage, suggesting that hate speech primarily revolves around political narratives.
Most Attacked Politicians
Qualitative analysis informed by automated hate speech detection reveals gender disparities and differences in the nature of online violence. Left-wing women are the most frequent targets of explicit attacks, with hate speech often targeting their political ideology. In contrast, hate speech directed at right-wing male politicians appears to be expressing support for their own hate speech rather than personal attacks.
Most Violent Politicians
Analyzing the content of tweets from politicians who rank high in hate speech reveals that references to minorities or targeted groups are common. While some instances may reflect false positives in hate speech detection, others clearly promote hate speech and calls to violence against marginalized groups and individuals.
The Political Alignment of Users
Political alignment analysis of users who engage in hate speech reveals a correlation between right-wing orientations and the reproduction of hate speech, particularly against women politicians. This aligns with previous research indicating the use of hate speech as a tactic by right-wing groups to target female candidates and undermine their political participation.
Conclusion
Our comprehensive analysis sheds light on the prevalence of hate speech in the context of Argentine politics. It demonstrates the gendered nature of online violence, with women politicians disproportionately targeted. While our models face limitations and may generate false positives in smaller samples, the qualitative analysis informs our interpretation and allows us to gain a deeper understanding of the discursive context within which various forms of hate speech emerge.
Our study underscores the need for strategies to combat hate speech in digital spaces, particularly against vulnerable groups. It also highlights the importance of fostering inclusive and respectful online environments where diverse voices can engage in discussions without resorting to harmful and hateful rhetoric. By bringing these issues to the forefront, we can work towards a more just and equitable online landscape.