Analyzing and learning the language for different types of harassment.

Area of Science:

• Computational Linguistics
• Social Media Analysis
• Natural Language Processing

Background:

• Harassment in user-generated content poses a significant challenge, necessitating advanced automatic detection methods.
• Existing harassment classification lacks contextual understanding, limiting accurate identification and analysis.
• Distinguishing types of harassment is crucial for developing effective mitigation strategies.

Purpose of the Study:

• To introduce and define five contextual types of harassment: sexual, racial, appearance-related, intellectual, and political.
• To analyze the linguistic characteristics and distribution of these harassment types on Twitter.
• To develop and evaluate type-aware classifiers for automated, accurate harassment detection.

Main Methods:

• Utilized a manually annotated Twitter corpus to categorize harassment into five contextual types.
• Conducted extensive linguistic analysis and uni-gram statistical distribution studies.
• Built and tested type-aware machine learning classifiers for harassment identification.

Main Results:

• Demonstrated that type-aware classifiers achieve competitive accuracy in detecting and analyzing social media harassment.
• Identified significant observations regarding the effectiveness of type-dependent features in harassment classification.
• Provided insights into the linguistic nuances and distribution patterns of different harassment types.

Conclusions:

• Contextual type classification is essential for a more nuanced understanding and detection of online harassment.
• Type-aware classifiers offer a robust approach to improving the accuracy of harassment identification on social media platforms.
• Further research into type-dependent features can enhance the performance of automated harassment detection systems.