---
title: "Arabic Tweet Act — Alshehri 2024"
description: "Most speech act work targets English or Modern Standard Arabic, while Arabic social media is mostly dialectal and has no fixed spelling or grammar."
authors: ["Khadejaa Alshehri", "Areej Alhothali", "Nahed Alowidi"]
year: 2024
venue: "arXiv:2401.17373"
source: "https://arxiv.org/abs/2401.17373"
doi: "10.48550/arxiv.2401.17373"
arxiv: "2401.17373"
licence: "CC BY 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/alshehri-2024-arabic-tweet-act-weighted-ensemble/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/alshehri-2024-arabic-tweet-act-weighted-ensemble/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2024](https://md.research.elamur.ai/free/en/papers/2024/page/5/) / Arabic Tweet Act: A Weighted Ensemble Pre-Trained Transformer Model for Classifying Arabic Speech Acts on Twitter

# Arabic Tweet Act: A Weighted Ensemble Pre-Trained Transformer Model for Classifying Arabic Speech Acts on Twitter

**Authors:** Khadejaa Alshehri, [Areej Alhothali](https://md.research.elamur.ai/free/en/authors/areej-alhothali/), Nahed Alowidi · **Institutions:** [King Abdulaziz University](https://md.research.elamur.ai/free/en/universities/king-abdulaziz-university/) · **Year:** 2024 · **Venue:** arXiv:2401.17373 · **Licence:** [CC BY 4.0 (arXiv)](http://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2401.17373) · [PDF](https://arxiv.org/pdf/2401.17373) · [DOI 10.48550/arxiv.2401.17373](https://doi.org/10.48550/arxiv.2401.17373) · [arXiv 2401.17373](https://arxiv.org/abs/2401.17373) · [OpenAlex](https://openalex.org/W7118787445)

**Citations:** 1 citations (OpenAlex, 2026-10-14)

## Summary
Most speech act work targets English or Modern Standard Arabic, while Arabic social media is mostly dialectal and has no fixed spelling or grammar. The authors tackle dialectal Arabic tweet act classification. They built a new dataset by labeling 22,352 tweets drawn from the ASAD sentiment corpus, chosen without topic restriction, using a Twitter-specific six-class taxonomy derived from Searle. They then compared several Arabic BERT variants and a BiLSTM baseline on this dataset and on the older ArSAS corpus. On top of the single models they placed a weighted ensemble that merges their predictions. To address class imbalance, they used transformer-based data augmentation to equalize category sizes, and they applied LIME to inspect the errors. araBERTv2-Twitter was the strongest single model on ASAD. The ensemble gave a small further gain. Augmentation raised macro-F1 noticeably on both datasets. The work matters because it supplies a public, topic-neutral resource and a reproducible baseline for intent recognition in Arabic dialects.

## Key points
- New dataset: 22,352 dialectal tweets from ASAD, labeled as Assertion, Expression, Request, Question, Recommendation, or Miscellaneous (the taxonomy of Vosoughi et al., adapted from Searle). It is released on GitHub.
- On ASAD, the best single model was araBERTv2-Twitter, with macro-F1 0.73 and accuracy 0.84. The weighted BERT ensemble reached 0.74 macro-F1 and 0.85 accuracy.
- On ArSAS, camelbert-mix was the best model at 0.58 macro-F1. Transformer-based augmentation lifted macro-F1 to 0.80 on ASAD and 0.69 on ArSAS.
- Transformer models generally beat the BiLSTM baseline on both corpora.
- On ArSAS, most errors fell between Assertion and Expression. The authors attribute this to imbalance and to the corpus being built around 20 often controversial topics.

## Interesting ideas
- Combining several pretrained classifiers with per-model weights captures complementary strengths, and it costs little beyond running the existing models. — For Mova (learn): Witness parsers that produce silver data could be weighted by measured reliability instead of counted by simple vote.
- Using a generative transformer to create extra examples of rare classes until all classes are balanced gave a much larger gain than switching architectures. — For Mova (learn): Rare speech-act or frame classes in training data could be topped up with teacher-generated, then verified, examples before training.
- The paper uses a compact, platform-adapted speech act inventory of six classes that includes an explicit Miscellaneous bucket, rather than a fine-grained rhetorical taxonomy. — For Mova (frames): The speech act layer in prag could begin with this coarse, auditable set and leave everything else to a catch-all, which keeps decisions explainable.
- The authors run per-token attribution with LIME on misclassified items to see which words drove a wrong label, such as courtesy phrases pulling assertions toward Expression. — For Mova (frames): Analogous traces could show which tree nodes or cue words triggered a speech act decision, which would feed self-induced repair rules.
- They sampled the corpus without topic filtering, to avoid the topic and hashtag skew of the earlier dataset, and they reported macro-F1 next to accuracy. — For Mova (learn): Evaluation sets for pragmatics should be topic-neutral and scored with macro-averaged metrics so that majority classes cannot hide failures on minority classes.

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

Fields: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/)

Topics: [Speech act theory](https://md.research.elamur.ai/free/en/topics/speech-act-theory/) · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Cite

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{alshehri2024arabic,
  title = {{Arabic Tweet Act: A Weighted Ensemble Pre-Trained Transformer Model for Classifying Arabic Speech Acts on Twitter}},
  author = {Alshehri, Khadejaa and Alhothali, Areej and Alowidi, Nahed},
  year = {2024},
  doi = {10.48550/arxiv.2401.17373},
  eprint = {2401.17373},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2401.17373}
}
```

- [← previous Arabic Text Sentiment Analysis: Reinforcing Human-Performed Surveys with Wider Topic Analysis](https://md.research.elamur.ai/free/en/papers/almurqren-2024-arabic-text-sentiment-analysis-reinforcing/)
- [next → AraSpider: Democratizing Arabic-to-SQL](https://md.research.elamur.ai/free/en/papers/heakl-2024-araspider-democratizing-arabic-sql/)
- [↑ back to papers of 2024](https://md.research.elamur.ai/free/en/papers/2024/page/5/)

## References

10 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. 1. 2. Austin, J.L. How to do things with words; Vol. 88, Oxford university press, 1975. Vosoughi, S.; Roy, D. Tweet acts: A speech act classifier for twitter. arXiv preprint arXiv:1605.05156 2016. 3. Kozlowski, D.; Lannelongue, E.; Saudemont, F.; Benamara, F.; Mari, A.; Moriceau, V.; Boumadane, A. A three-level classification of French tweets in ecological crises. [↗](https://arxiv.org/abs/1605.05156)
2. Information Processing & Management 2020, 57, 102284. 4. Jahanbakhsh-Nagadeh, Z.; Feizi-Derakhshi, M.R.; Sharifi, A. A Speech Act Classifier for Persian Texts and its Application in Identify Speech Act of Rumors. arXiv preprint arXiv:1901.03904 2019. 5. Farghaly, A.; Shaalan, K. Arabic natural language processing: Challenges and solutions. ACM Transactions on Asian Language Information Processing (TALIP) 2009, 8, 1–22. 6. Habash, N.Y. Introduction to Arabic natural language processing; Vol. 3, Morgan & Claypool Publishers, 2010; pp. 1–187. 7. Dbabis, S.B.; Mallek, F.; Ghorbel, H.; Belguith, L. Dialogue acts annotation scheme within arabic discussions. In Proceedings of the Proceedings of SemDial 2012 (SeineDial): The 16th Workshop on the Semantics and Pragmatics of Dialogue, 2012, p. 88. 8. Elmadany, A.A.; Abdou, S.M.; Gheith, M. Arabic Inquiry-Answer Dialogue Acts Annotation Schema. arXiv preprint arXiv:1505.04197 2015. 9. Elmadany, A.A.; Abdou, S.M.; Gheith, M. JANA: An Arabic human-human dialogues corpus. In Proceedings of the 2015 IEEE 2nd International Conference on Recent Trends in Information Systems (ReTIS). [↗](https://arxiv.org/abs/1901.03904)
3. IEEE, 2015, pp. 347–352. 10. Ahed, M.; Hammo, B.H.; Abushariah, M.A. An Enhanced Twitter Corpus for the Classification of Arabic Speech Acts 2020. 11. Alharbi, B.; Alamro, H.; Alshehri, M.; Khayyat, Z.; Kalkatawi, M.; Jaber, I.I.; Zhang, X. ASAD: A twitter-based benchmark arabic sentiment analysis dataset. arXiv preprint arXiv:2011.00578 2020. 12. Elmadany, A.; Mubarak, H.; Magdy, W. Arsas: An arabic speech-act and sentiment corpus of tweets. OSACT 2018, 3, 20. 13. Searle, J.R. A taxonomy of illocutionary acts 1975. 14. Vosoughi, S. Automatic detection and verification of rumors on Twitter. PhD thesis, Massachusetts Institute of Technology, 2015. 15. Zhang, R.; Li, W.; Gao, D.; Ouyang, Y. Automatic twitter topic summarization with speech acts. IEEE transactions on audio, speech, and language processing 2012, 21, 649–658. 16. Zhang, F.; Stromer-Galley, J.; Tanupabrungsun, S.; Hegde, Y.; McCracken, N.; Hemsley, J. Understanding discourse acts: Political campaign messages classification on Facebook and Twitter. In Proceedings of the International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation. [↗](https://arxiv.org/abs/2011.00578) [our card](https://md.research.elamur.ai/free/en/papers/alharbi-2020-asad-twitter-based-benchmark-arabic/)
4. Springer, 2017, pp. 242–247. 17. Khanpour, H.; Guntakandla, N.; Nielsen, R. Dialogue act classification in domain-independent conversations using a deep recur- rent neural network. In Proceedings of the Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, 2016, pp. 2012–2021. 18. Saha, T.; Srivastava, S.; Firdaus, M.; Saha, S.; Ekbal, A.; Bhattacharyya, P. Exploring machine learning and deep learning frameworks for task-oriented dialogue act classification. In Proceedings of the 2019 International Joint Conference on Neural Networks (IJCNN). [find](https://scholar.google.com/scholar?q=Springer%2C+2017%2C+pp%2E+242%E2%80%93247%2E+17%2E+Khanpour%2C+H%2E%3B+Guntakandla%2C+N%2E%3B+Nielsen%2C+R%2E+Dialogue+act+classification+in+domain%2Dindependent+conversations+using+a+deep+recur%2D+rent+neural+network%2E+In+Proceedings+of+t)
5. IEEE, 2019, pp. 1–8. 19. Kim, M.; Kim, H. Dialogue act classification model based on deep neural networks for a natural language interface to databases in korean. In Proceedings of the 2018 IEEE International Conference on Big Data and Smart Computing (BigComp). [find](https://scholar.google.com/scholar?q=IEEE%2C+2019%2C+pp%2E+1%E2%80%938%2E+19%2E+Kim%2C+M%2E%3B+Kim%2C+H%2E+Dialogue+act+classification+model+based+on+deep+neural+networks+for+a+natural+language+interface+to+databases+in+korean%2E+In+Proceedings+of+the+2018+IEEE+Inter)
6. IEEE, 2018, pp. 537–540. 20. Saha, T.; Patra, A.P.; Saha, S.; Bhattacharyya, P. A Transformer based Approach for Identification of Tweet Acts. In Proceedings of the 2020 International Joint Conference on Neural Networks (IJCNN). [find](https://scholar.google.com/scholar?q=IEEE%2C+2018%2C+pp%2E+537%E2%80%93540%2E+20%2E+Saha%2C+T%2E%3B+Patra%2C+A%2EP%2E%3B+Saha%2C+S%2E%3B+Bhattacharyya%2C+P%2E+A+Transformer+based+Approach+for+Identification+of+Tweet+Acts%2E+In+Proceedings+of+the+2020+International+Joint+Conference)
7. IEEE, 2020, pp. 1–8. 21. Zhang, R.; Gao, D.; Li, W. What are tweeters doing: Recognizing speech acts in twitter. In Proceedings of the Workshops at the Twenty-Fifth AAAI Conference on Artificial Intelligence. [find](https://scholar.google.com/scholar?q=IEEE%2C+2020%2C+pp%2E+1%E2%80%938%2E+21%2E+Zhang%2C+R%2E%3B+Gao%2C+D%2E%3B+Li%2C+W%2E+What+are+tweeters+doing%3A+Recognizing+speech+acts+in+twitter%2E+In+Proceedings+of+the+Workshops+at+the+Twenty%2DFifth+AAAI+Conference+on+Artificial+Intel)
8. Citeseer, 2011. 22. Saha, T.; Saha, S.; Bhattacharyya, P. Tweet Act Classification: A Deep Learning based Classifier for Recognizing Speech Acts in Twitter. In Proceedings of the 2019 International Joint Conference on Neural Networks (IJCNN). [find](https://scholar.google.com/scholar?q=Citeseer%2C+2011%2E+22%2E+Saha%2C+T%2E%3B+Saha%2C+S%2E%3B+Bhattacharyya%2C+P%2E+Tweet+Act+Classification%3A+A+Deep+Learning+based+Classifier+for+Recognizing+Speech+Acts+in+Twitter%2E+In+Proceedings+of+the+2019+International+Jo)
9. IEEE, 2019, pp. 1–8. 23. Saha, T.; Jayashree, S.R.; Saha, S.; Bhattacharyya, P. BERT-Caps: A Transformer-Based Capsule Network for Tweet Act Classifica- tion. IEEE Transactions on Computational Social Systems 2020, 7, 1168–1179. 24. Sherkawi, L.; Ghneim, N.; Dakkak, O.A. Arabic speech act recognition techniques. ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP) 2018, 17, 1–12. 25. Algotiml, B.; Elmadany, A.; Magdy, W. Arabic Tweet-Act: Speech Act Recognition for Arabic Asynchronous Conversations. In Proceedings of the Proceedings of the Fourth Arabic Natural Language Processing Workshop, 2019, pp. 183–191. 26. Joukhadar, A.; Ghneim, N.; Rebdawi, G. Impact of Using Bidirectional Encoder Representations from Transformers (BERT) Models for Arabic Dialogue Acts Identification. Ingénierie des Systèmes d Inf. 2021, 26, 469–475. 27. Tavafi, M.; Mehdad, Y.; Joty, S.; Carenini, G.; Ng, R. Dialogue act recognition in synchronous and asynchronous conversations. In Proceedings of the Proceedings of the SIGDIAL 2013 Conference, 2013, pp. 117–121. 28. Ma, E. NLP Augmentation. https://github.com/makcedward/nlpaug, 2019. 29. Devlin, J.; Chang, M.W.; Lee, K.; Toutanova, K. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805 2018. 30. Soliman, A.B.; Eissa, K.; El-Beltagy, S.R. Aravec: A set of arabic word embedding models for use in arabic nlp. [↗](https://arxiv.org/abs/1810.04805) [our card](https://md.research.elamur.ai/free/en/papers/tavafi-2013-dialogue-act-recognition-synchronous-asynchronous/)
10. Procedia Computer Science 2017, 117, 256–265. 31. Antoun, W.; Baly, F.; Hajj, H. AraBERT: Transformer-based Model for Arabic Language Understanding. In Proceedings of the LREC 2020 Workshop Language Resources and Evaluation Conference 11–16 May 2020, p. 9. 32. Inoue, G.; Alhafni, B.; Baimukan, N.; Bouamor, H.; Habash, N. The Interplay of Variant, Size, and Task Type in Arabic Pre-trained Language Models. In Proceedings of the Proceedings of the Sixth Arabic Natural Language Processing Workshop; Association for Computational Linguistics: Kyiv, Ukraine (Online), 2021. 33. Abdul-Mageed, M.; Elmadany, A.; Nagoudi, E.M.B. ARBERT & MARBERT: Deep Bidirectional Transformers for Arabic. In Proceedings of the Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers); Association for Computational Linguistics: Online, 2021; pp. 7088–7105. https://doi.org/10.18653/v1/2021.acl-long.551. 34. Abdelali, A.; Hassan, S.; Mubarak, H.; Darwish, K.; Samih, Y. Pre-Training BERT on Arabic Tweets: Practical Considerations 2021. [arXiv:cs.CL/2102.10684]. 35. Elmadany, A. Automatic Act Classification for Arabic Dialogue Context. PhD thesis, Ph. D. thesis, Thesis, 2016. 36. Zhao, X.; Jiang, J. An empirical comparison of topics in twitter and traditional media. Singapore Management University School of Information Systems Technical paper series. Retrieved November 2011, 10, 2011. 37. Ribeiro, M.T.; Singh, S.; Guestrin, C. " Why should i trust you?" Explaining the predictions of any classifier. In Proceedings of the Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, 2016, pp. 1135–1144. [↗](https://doi.org/10.18653/v1/2021.acl-long.551) [our card](https://md.research.elamur.ai/free/en/papers/ribeiro-2016-why-should-i-trust-explaining/)