Elamur Research⌕
☰
home / papers / 2022 / AraDepSu: Detecting Depression and Suicidal Ideation in Arabic Tweets Using Transformers

AraDepSu: Detecting Depression and Suicidal Ideation in Arabic Tweets Using Transformers

Paper · 2022 · Mariam Hassib, Nancy Hossam, Jolie Sameh et al.

Summary

The paper addresses the lack of Arabic resources for spotting signs of depression and suicidal thoughts on social media. Stigma around mental illness in Arab societies makes early diagnosis hard, but many people describe their feelings openly on Twitter. The authors built AraDepSu, a three-class tweet corpus with the labels depression, depression with suicidal ideation, and non-depression. They combined three sources: keyword-driven scraping of tweets from 2016–2022, records from an existing MSA depression dataset, and tweets from four Arabic sentiment corpora that supply neutral examples. Volunteers labeled the data, and the authors then reviewed every label. More than 30 Transformer-based Arabic models were fine-tuned on the corpus. MARBERT, which was pretrained largely on dialectal tweets, performed best, ahead of MSA-oriented models such as AraBERT. The authors attribute this to the strongly dialectal nature of the data. The work provides a rare Arabic benchmark for mental-health text classification and shows that matching the pretraining domain to the target text matters.

Key points

  • AraDepSu has 20,213 tweets: 5,472 depression, 2,167 suicidal ideation and 12,574 non-depression. The texts mix MSA and dialects, with Egyptian Arabic as the main focus.
  • The data sources are:
    • over 10k keyword-scraped tweets;
    • 1,230 records from the MSA mood-changing and depression dataset (Maghraby and Ali, 2022);
    • 9,133 examples from ArTwitter (3,543), TEAD (2,000), BRAD (2,000) and ASTD (1,590).
  • Preprocessing removed hyperlinks, empty columns and duplicate records.
  • Each tweet was labeled by one annotator. The authors then checked every sample and their decision prevailed in disagreements.
  • MARBERT gave the best scores: 91.20% accuracy, 88.74% macro precision, 88.50% macro recall and 88.75% macro F1.

Interesting ideas

  • I1. The neutral class was built from existing sentiment corpora rather than collected from scratch, so the classifier sees many realistic negatives that are not about the target topic.
  • I2. Most errors came from tweets that mention dying or sadness without describing the writer's own current state, such as a remark about a TV character's death or a generic statement.
  • I3. The dialect-heavy MARBERT beat MSA-pretrained models because its pretraining text resembled the target genre.
  • I4. Labeling was cheap (one annotator per item), but the authors reviewed every sample with final authority, which trades inter-annotator redundancy for consistent adjudication.
  • I5. A qualitative table puts two models' predictions next to the gold label for tricky test items, which exposes where each model's assumptions break.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Evaluation with negative controls could draw "off-topic" texts from existing corpora, checking that frame or state detectors stay silent on everyday text.
  • I2 (frames): Frames should record who actually experiences a state (speaker, fictional character, generic person), so a negative emotion word is not automatically attributed to the narrator.
  • I3 (tree): When selecting or pretraining the gparse encoder, match its training text to the target genre (tales, school texts, informal prose) rather than defaulting to the most formal corpus.
  • I4 (learn): Silver data could use a single fast labeler, such as an LLM teacher tier, followed by a systematic reviewer pass that overrides it, with the overrides logged as explainable corrections.
  • I5 (answer): Side-by-side outputs from competing readers on hand-picked hard questions would make failure modes visible beyond aggregate test scores.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Alexandria UniversityAlexandria · Egypt39 works · 287 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksPsychology · 1,552 worksExperimental and Cognitive Psychology · 870 worksSocial Psychology · 306 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Mariam Hassib, Nancy Hossam, Jolie Sameh and Marwan Torki. 2022. AraDepSu: Detecting Depression and Suicidal Ideation in Arabic Tweets Using Transformers. Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2022.wanlp-1.28

@inproceedings{hassib2022aradepsu,
  title = {{AraDepSu: Detecting Depression and Suicidal Ideation in Arabic Tweets Using Transformers}},
  author = {Mariam Hassib, Nancy Hossam, Jolie Sameh and Marwan Torki},
  booktitle = {Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)},
  year = {2022},
  doi = {10.18653/v1/2022.wanlp-1.28},
  url = {https://doi.org/10.18653/v1/2022.wanlp-1.28}
}

References

21 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdelali, Sabit Hassan, Hamdy Mubarak, Kareem Darwish, and Younes Samih. 2021. Pre-Wissam Antoun, Fady Baly, and Hazem Hajj. Arabert: Transformer-based model for arabic language understanding. In LREC 2020 Workshop Language Resources and Evaluation Conference 11–16 May 2020, page 9.
  2. Wissam Antoun, Fady Baly, and Hazem Hajj. 2021a. AraELECTRA: Pre-training text discriminators for Arabic language understanding. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 191–195, Kyiv, Ukraine (Virtual). Association for Computational Linguistics. ↗
  3. Wissam Antoun, Fady Baly, and Hazem Hajj. 2021b. AraGPT2: Pre-trained transformer for Arabic language generation. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 196–207, Kyiv, Ukraine (Virtual). Association for Computational Linguistics. ↗
  4. Nirmal Varghese Babu and E Kanaga. 2022. Sentiment analysis in social media data for depression detection using artificial intelligence: A review. SN Computer Science, 3(1):1–20. ↗
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Unsupervised cross-lingual representation learning at scale. CoRR, abs/1911.02116. ↗
  6. Glen Coppersmith, Mark Dredze, and Craig Harman. 2014. Quantifying mental health signals in twitter. In Proceedings of the workshop on computational linguistics and clinical psychology: From linguistic signal to clinical reality, pages 51–60.
  7. Munmun De Choudhury, Scott Counts, and Eric Horvitz. 2013a. Social media as a measurement tool of depression in populations. In Proceedings of the 5th annual ACM web science conference, pages 47–56. find
  8. Munmun De Choudhury, Michael Gamon, Scott Counts, and Eric Horvitz. 2013b. Predicting depression via social media. In Seventh international AAAI conference on weblogs and social media. find
  9. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805. Ahmed Abdelali Kareem Darwish Nadir Durrani and Hamdy Mubarak. Farasa: A fast and furious segmenter for arabic. ↗
  10. Ashraf Elnagar, Leena Lulu, and Omar Einea. 2018. An annotated huge dataset for standard and colloquial arabic reviews for subjective sentiment analysis. Procedia computer science, 142:182–189. find
  11. Wuwei Lan, Yang Chen, Wei Xu, and Alan Ritter. 2020. An empirical study of pre-trained transformers for arabic information extraction. arXiv preprint arXiv:2004.14519. ↗
  12. Ashwag Maghraby and Hosnia Ali. 2022. Modern standard arabic mood changing and depression dataset. Data in Brief, 41:107999. ↗
  13. Marina Marcus, M Taghi Yasamy, Mark van van Ommeren, Dan Chisholm, and Shekhar Saxena. 2012. Depression: A global public health concern. find
  14. Rita Merhej. 2019. Stigma on mental illness in the arab world: beyond the socio-cultural barriers. International Journal of Human Rights in Healthcare. ↗
  15. Dhiaa A Musleh, Taef A Alkhales, Reem A Almakki, Shahad E Alnajim, Shaden K Almarshad, Rana S Alhasaniah, Sumayh S Aljameel, and Abdullah A Almuqhim. 2022. Twitter arabic sentiment analysis to detect depression using machine learning. CMC-COMPUTERS MATERIALS & CONTINUA, 71(2):3463–3477. ↗
  16. Mahmoud Nabil, Mohamed Aly, and Amir Atiya. 2015. Astd: Arabic sentiment tweets dataset. In Proceedings of the 2015 conference on empirical methods in natural language processing, pages 2515–2519. ↗
  17. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. 2022. AraT5: Text-to-text transformers for Arabic language generation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 628–647, Dublin, Ireland. Association for Computational Linguistics.
  18. Ali Safaya. 2020. Arabic-albert. find
  19. Ali Safaya, Moutasem Abdullatif, and Deniz Yuret. 2020. KUISAIL at SemEval-2020 task 12: BERT-CNN for offensive speech identification in social media. In Proceedings of the Fourteenth Workshop on Semantic Evaluation, pages 2054–2059, Barcelona (online). International Committee for Computational Linguistics.
  20. Pedro Javier Ortiz Suárez, Laurent Romary, and Benoît Sagot. 2020. A monolingual approach to contextualized word embeddings for mid-resource languages. arXiv preprint arXiv:2006.06202. ↗
  21. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, et al. 2019. Huggingface’s transformers: State-ofthe-art natural language processing. arXiv preprint arXiv:1910.03771. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/hassib-2022-aradepsu-detecting-depression-suicidal-ideation/