Us and them: identifying cyber hate on Twitter across multiple protected characteristics

Area of Science:

• Computational Linguistics
• Social Computing
• Cybersecurity

Background:

• Online platforms facilitate the spread of hateful content, causing individual harm and societal disruption.
• Existing legislation and platform policies struggle to effectively police and prosecute cyber hate due to detection challenges.
• Automatic detection of cyber hate is crucial for mitigating its negative impacts.

Purpose of the Study:

• To develop and evaluate machine learning models for automatically detecting cyber hate on Twitter.
• To improve the accuracy of cyber hate classification by incorporating linguistic features beyond simple keyword matching.
• To address the challenge of classifying intersectional cyber hate, which targets multiple protected characteristics simultaneously.

Main Methods:

• Utilized text parsing to extract typed dependencies, capturing syntactic and grammatical relationships.
• Developed individual classification models for various protected characteristics (race, disability, sexual orientation).
• Constructed a data-driven blended model to handle cases of intersectional cyber hate.

Main Results:

• Typed dependencies significantly improved cyber hate classification accuracy compared to Bag of Words and known hateful terms.
• The developed models demonstrated effectiveness in identifying cyber hate across different protected characteristics.
• The blended model showed promise in classifying complex cases of intersectional cyber hate.

Conclusions:

• Text parsing and typed dependencies are valuable features for enhancing automated cyber hate detection systems.
• The study contributes to more robust methods for identifying and combating online hate speech, particularly intersectional forms.
• Further research into intersectionality in hate crime detection can improve online safety and support legal prosecution.