Core Concepts
Hate and offensive speech on social media exist on a continuous spectrum with varying degrees of intensity, rather than as discrete binary categories.
Abstract
The study presents an extensive benchmark dataset for Amharic, comprising 8,258 tweets annotated for three distinct tasks: category classification (hate, offensive, normal), identification of hate targets, and rating offensiveness and hatefulness intensities on a 5-point Likert scale.
The key highlights and insights are:
A significant majority of tweets belong to the less offensive and less hate intensity levels, underscoring the need for early interventions by stakeholders.
The prevalence of ethnic and political hatred targets, with significant overlaps, emphasizes the complex relationships within Ethiopia's sociopolitical landscape.
Hate and offensive speech cannot be addressed by a simplistic binary classification, as they manifest as variables across a continuous range of values.
The Afro-XLMR-large model exhibits the best performances, achieving F1-scores of 75.30%, 70.59%, and 29.42% for the category, target, and regression tasks, respectively. The 80.22% correlation coefficient of the Afro-XLMR-large model indicates strong alignments between predicted and actual intensity levels.
The findings demonstrate that hate and offensive speech exist on a spectrum, with varying degrees of intensity, rather than as discrete binary categories. This highlights the need for more nuanced approaches to detect and mitigate the impact of such content on social media.
Stats
"The average offensiveness intensity ranges from 0.2 to 3.0 for mild, 3.0 to 4.0 for moderate, and 4.0 to 5.0 for severe categories."
"The average hatefulness intensity ranges from 0.2 to 3.0 for early warning, 3.0 to 4.0 for dehumanization, and 4.0 to 5.0 for violence and incitement categories."
Quotes
"Hate speech on social media can take various forms, including discriminatory language, threats, harassment, and the incitement of violence against specific individuals or groups of communities."
"Hate speech classification demonstrates a spectrum of continuity, and contemporary studies have recognized this limitation by prompting a shift towards adopting multifaceted methodologies to gain a better understanding of the nature, dimension, and intensity of hate speech."