Arabic Mental Health Dataset

Questionnaire Website: We developed a dedicated website to collect mental state descriptions, targeting 1,000–2,000 responses from Arabic-speaking regions (e.g., Egypt, Lebanon, Saudi Arabia). The questionnaire captured symptoms for anxiety, OCD, depression, and suicidality. Respondents used MSA or dialects. To maximize reach, we implemented a Facebook marketing plan, targeting diverse demographics, and collaborated with psychologists to promote the questionnaire and ensure credibility. Data was labeled based on certified psychological test results (e.g., PHQ-9 for depression, GAD-7 for anxiety, Y-BOCS for OCD, C-SSRS for suicidality).

Web Scraped Data:

• Altibbi (altibbi.com): A digital health platform with doctor-verified Q&A. We scraped over 5,000 items per category (anxiety, OCD, depression, suicidality), selecting high-quality entries per condition based on doctor-provided labels aligned with DSM-5.
• Islamweb (islamweb.net): Offers Islamic fatwas and articles, including mental health topics. We scraped over 5,000 items per category, selecting entries labeled by medical professionals or religious scholars referencing health expertise, aligned with DSM-5.
• Additional Sources: CairoDep, CairoMent, ASKfm, Hellooha for diverse self-describing posts, contributing to a total of 30,000 items.

English Dataset for Suicidality: Due to limited Arabic suicidality data, we used the Suicide and Depression Detection dataset (Kaggle), containing 232,074 Reddit posts labeled for suicide, depression, and non-suicide ("normal"). We translated 10,000 suicide-related posts to Arabic, and selected items validated by native speakers for cultural accuracy.

Data Labeling and Conversion

Labeling:

• Questionnaire: Labeled using certified test results (e.g., PHQ-9, GAD-7, Y-BOCS, C-SSRS for suicidality). "Normal" cases were identified by symptom-level analysis (e.g., mild stress not meeting diagnostic thresholds), ensuring granular differentiation. Manual labeling by mental health experts (psychiatrist, psychologist) achieved 99% inter-rater agreement.
• Altibbi and Islamweb: Labeled based on doctor-verified answers or expert responses, aligned with DSM-5 criteria. We selected items per condition and "normal"/"other" cases, ensuring balance.
• Translated Data: Labeled as in the original English dataset, validated for suicidality and "normal" cases.