ArabDiscrim: A Decade-Long Arabic Facebook Corpus on Racism and Discrimination

2026-05-22 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

ArabDiscrim is a newly released lexical resource and corpus comprising 293,000 public Arabic Facebook posts collected between 2014 and 2024, specifically focusing on discussions of racism and discrimination. Unlike existing datasets primarily derived from Twitter, ArabDiscrim uniquely incorporates platform-native engagement signals such as reactions, shares, comments, and page metadata, enabling a joint analysis of language and audience response. The resource features 200 curated terms—100 related to racism and 100 to discrimination—each with morphological regex families covering over 13 inflections per lemma. It also delineates 20 distinct discrimination axes and provides explicit attribution patterns. Released under a restricted research-use license, ArabDiscrim is designed to support weak supervision, axis-aware sampling, and platform ecology research, establishing a crucial foundation for fairness-oriented, platform-aware Arabic Natural Language Processing.

Key takeaway

For NLP Engineers developing fairness-aware systems for Arabic, ArabDiscrim offers a critical resource to improve model robustness. You should integrate this corpus to train and evaluate models for racism and discrimination detection, leveraging its unique platform engagement signals and 20 discrimination axes. This approach will enhance your system's ecological validity and ethical compliance, moving beyond Twitter-centric biases.

Key insights

ArabDiscrim bridges lexical depth and ecological validity for fairness-oriented, platform-aware Arabic NLP.

Principles

• Platform-native engagement signals enrich language analysis.
• Morphological regex families enhance lexical resource coverage.
• Ethical compliance guides corpus development and release.

Method

A corpus of 293K Arabic Facebook posts (2014-2024) was built, integrating engagement signals, 200 curated terms with morphological regex, and 20 discrimination axes.

In practice

• Employ ArabDiscrim for weak supervision.
• Conduct axis-aware sampling for fairness models.
• Analyze platform ecology via language and audience response.

Topics

• ArabDiscrim
• Arabic NLP
• Racism Detection
• Discrimination Analysis
• Social Media Data
• Corpus Development

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Ethicist

Related on AIssential

• ArabDiscrim: A Decade-Long Arabic Facebook Corpus on Racism and Discrimination — ArabDiscrim offers a unique, decade-long Arabic Facebook corpus for racism and discrimination research, integrating platform engagement signals.
• IndoBias: A Dual Track Culturally Grounded Benchmark for LLMs Bias Evaluation in Indonesian Languages — Culturally-grounded benchmarks are crucial for evaluating LLM bias in diverse, multilingual contexts like Indonesia.
• Contaminated Collaboration: Measuring Gender Bias Transfer in LLM-Assisted Student Writing — Gender bias from LLM writing assistants transfers to human-produced text, particularly suppressing agency in female-targeted writing.
• Investigating Counterfactual Unfairness in LLMs towards Identities through Humor — LLMs exhibit counterfactual unfairness, applying stricter safety policies and attributing more malicious intent based on speaker-target identity.
• Audience Engagement with Arabic Women's Social Empowerment and Wellbeing: A Decadal Corpus — The corpus offers a decadal view of Arabic audience engagement with women's empowerment and wellbeing via Facebook posts and interactions.
• RacismoBR: A Manually Annotated Dataset for Racist Discourse Detection in Brazilian Portuguese — Culturally grounded annotation is more critical than model architecture for effective racism detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.