📄 From Posts to Pressure: An Arabic Dataset about Stress and Mental-Health Monitoring

Summary

Arabic mental-health NLP has concentrated mostly on depression and suicidal ideation, usually with data from X. Psychological stress has received little attention, and Facebook, one of the most popular platforms in the Arab world, has been largely ignored. The authors address this by building AraStress, which they describe as the first large-scale Arabic corpus focused on stress. It contains 175,862 public posts collected through CrowdTangle from mental-health and wellness pages using stress-related keywords. The posts cover the pandemic and the years after it, and each one comes with detailed engagement, temporal, geographic and page-type metadata. The analysis combines normalized reaction ratios, two small lexicons (religious and therapeutic), and a proportional comparison against a separate Arabic Trauma Corpus. Responses to stress posts are mostly warm or humorous, with very little anger. Stress tends to be told as personal narrative that blends faith-based and clinical vocabulary, and this mix varies by country. Trauma posts draw noticeably more sadness. The resource supports culturally sensitive stress-detection models for Arabic speakers.

Key points

• AraStress: 175,862 public Arabic Facebook posts collected between January 2020 and May 2024 via CrowdTangle. Each post has 52 metadata fields, including separate counts for each reaction type, comments, shares, timestamps, Page Admin Top Country, and page category. Access is for research only, through an anonymized release or on request.
• Reactions to stress posts: Love 60.7%, Haha 17.7%, Sad 11.6%, Care 6.3%, Angry 1.8%. The authors read this as a mix of empathy and humor used for coping. Virality was defined as engagement above the 90th percentile.
• Comparison corpus: the Arabic Trauma Corpus has 16,529 Facebook posts (May 2014 to May 2024) with 551,904 reactions. Its reactions are Love 51.4%, Sad 22.5% and Haha 19.7%, so sadness is roughly twice as common as in the stress data.
• Lexical framing: "God" accounts for about 75% of religious-lexicon hits (49,761 occurrences). "Treatment/therapy" is the leading therapeutic term (23,875 occurrences, about 34%). The ratio of religious to therapeutic vocabulary differs across countries.
• Limitations: country labels come from page administrators, not from the individual users who wrote the posts. Reactions reflect user behavior rather than clinical assessment. The lexicon-based method cannot tell supportive humor from sarcasm.