[Paper Review] Analyzing the Targets of Hate in Online Social Media

Leandro Silva, Mainack Mondal+3 more|arXiv (Cornell University)|Mar 24, 2016

Hate Speech and Cyberbullying DetectionComputer Science7 references105 citations

TL;DR

The paper analyzes hate speech targets on Whisper and Twitter using a sentence-structure detector, creating large hate speech datasets and categorizing targets into nine hate categories.

ABSTRACT

Social media systems allow Internet users a congenial platform to freely express their thoughts and opinions. Although this property represents incredible and unique communication opportunities, it also brings along important challenges. Online hate speech is an archetypal example of such challenges. Despite its magnitude and scale, there is a significant gap in understanding the nature of hate speech on social media. In this paper, we provide the first of a kind systematic large scale measurement study of the main targets of hate speech in online social media. To do that, we gather traces from two social media systems: Whisper and Twitter. We then develop and validate a methodology to identify hate speech on both these systems. Our results identify online hate speech forms and offer a broader understanding of the phenomenon, providing directions for prevention and detection approaches.

Motivation & Objective

• Motivate the study by addressing the lack of big-picture understanding of online hate speech in popular social media.
• Develop a scalable method to identify hate speech in social media posts based on sentence structure.
• Construct and validate hate speech datasets for Whisper and Twitter.
• Characterize hate targets to reveal prevalent forms and patterns of online hate.

Proposed method

• Define hate speech as offense motivated by bias against a group characteristic.
• Use a sentence-structure pattern to detect hate posts of the form I <intensity> <userintent> <hatetarget>.
• Target hate analysis with templates like <one word> people and Hatebase word lists (offensivity >50).
• Crawl Hatebase for hate words and filter for high offensivity words (116 words meeting >50%).
• Manually categorize hate targets into nine categories (Race, Behavior, Physical, Sexual orientation, Class, Gender, Ethnicity, Disability, Religion) plus Other.

Experimental results

Research questions

• RQ1 What are the main targets of hate speech in Twitter and Whisper posts?
• RQ2 How prevalent are different hate categories across the two platforms?
• RQ3 Can a pattern-based approach reliably identify hate speech for constructing large-scale datasets?
• RQ4 Do hate targets differ between Whisper (anonymous) and Twitter (public) in terms of category distribution?

Key findings

• The method identified 20,305 tweets and 7,604 whispers containing hate speech.
• Top hate targets include Race, Behavior, and Physical attributes across both platforms.
• Whisper shows a lower share of Race-related hate and a higher share of Non-race categories in some parts, possibly due to prior filtering.
• The nine defined hate categories cover the majority of targets (most targets in both datasets fall into Race, Behavior, Physical).