Unveiling disguised toxicity: A novel pre-processing module for enhanced content moderation.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Cybersecurity

Background:

• Online communications contain disguised toxic content designed to evade detection.
• Existing toxic text detection tools struggle with intentionally altered or obfuscated language.

Purpose of the Study:

• To introduce "Specialis Revelio," a novel text pre-processing module.
• To enhance the accuracy of toxic content detection systems.
• To address the challenge of detecting disguised toxic text.

Main Methods:

• Developed "Specialis Revelio" integrating conventional and novel pre-processing techniques.
• Combined lowercasing, stemming, adversarial example handling, and typo correction.
• Designed for seamless integration with existing toxic text classifiers.

Main Results:

• Demonstrated significant improvement in detecting disguised toxic content.
• Validated effectiveness through comparative analysis against popular APIs.
• Showcased enhanced accuracy in identifying various toxic text indicators.

Conclusions:

• "Specialis Revelio" effectively augments toxic text classifiers.
• The module successfully reveals concealed toxicity through advanced pre-processing.
• This approach offers a more robust solution for online content moderation.