---
title: "Machine learning and emoji prediction: How much accuracy can MARBERT achieve? — Shormani 2026"
description: "The paper asks whether the emoji a user attaches to an Arabic tweet can be guessed from the text… 0.84 precision. 40 references, cited by 1 paper here."
authors: ["Mohammed Q. Shormani", "Ibrahim Abdulmalik Hassan Muneef Y. Alshawsh"]
year: 2026
venue: "arXiv:2604.21108"
source: "https://arxiv.org/abs/2604.21108"
doi: "10.48550/arxiv.2604.21108"
arxiv: "2604.21108"
licence: "CC BY-SA 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by-sa/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/shormani-2026-machine-learning-emoji-prediction-much/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/shormani-2026-machine-learning-emoji-prediction-much/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/) / Machine learning and emoji prediction: How much accuracy can MARBERT achieve?

# Machine learning and emoji prediction: How much accuracy can MARBERT achieve?

**Authors:** [Mohammed Q. Shormani](https://md.research.elamur.ai/free/en/authors/mohammed-q-shormani/) and Ibrahim Abdulmalik Hassan Muneef Y. Alshawsh · **Year:** 2026 · **Venue:** arXiv:2604.21108 · **Licence:** [CC BY-SA 4.0 (arXiv)](http://creativecommons.org/licenses/by-sa/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2604.21108) · [PDF](https://arxiv.org/pdf/2604.21108) · [DOI 10.48550/arxiv.2604.21108](https://doi.org/10.48550/arxiv.2604.21108) · [arXiv 2604.21108](https://arxiv.org/abs/2604.21108) · [OpenAlex](https://openalex.org/W7155570388)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/shormani-2026-machine-learning-emoji-prediction-much/cited-by/) in our library, of the 140,884 papers whose bibliographies we read

## Summary
The paper asks whether the emoji a user attaches to an Arabic tweet can be guessed from the text alone, focusing on colloquial Arabic (CA) rather than Standard Arabic. The authors scraped 11,379 CA tweets from X.com with Python and kept 8695 after cleaning. They grouped the emojis into 14 categories, encoded those categories as numeric labels, and fine-tuned MARBERT, a BERT-style model pretrained on dialectal social-media Arabic, as a multi-class classifier. As an interpretable baseline, they used a preprocessing pipeline with TF-IDF features and Logistic Regression, which shows which words point toward which emoji class. Evaluation used per-class precision, recall and F1. MARBERT reached 0.75 overall accuracy, with macro-averaged scores clearly lower. Most errors fell between emotionally neighbouring categories. The authors read the results as evidence that emoji choice is partly predictable from dialectal text. They also argue that models for low-resource, multi-dialect languages still need improvement.

## Key points
- Data: 11,379 CA tweets were collected and 8695 kept. The conclusion describes a split of 4000 for training and 4695 for testing. Labels are 14 emoji classes, not individual emojis.
- Overall accuracy was 0.75. Macro precision, recall and F1 were 0.58, 0.74 and 0.64. The best single class reached 0.84 precision, 0.87 recall and 0.86 F1.
- The paper also reports 3203 of 4695 test tweets (about 68%) as a separate "encouraging" result. This sits somewhat awkwardly next to the 75% accuracy figure.
- Training loss fell from 1.61 at step 50 to 0.01 at step 400. Validation loss went from 0.133 after epoch 1 to 0.039 after epoch 2.
- Error analysis found that confusions cluster among semantically close classes, for example love versus exaggerated humour, or stress versus weariness.
- Limitations named by the authors:
  - category labels were assigned by a model rather than human annotators;
  - the dataset is small;
  - data comes from a single platform.

## Interesting ideas
- Treat emojis as free, user-supplied affect labels that turn raw social text into a supervised dataset without annotation
- Collapse many fine-grained labels into a smaller set of semantic classes so sparse items still get usable training signal
- Run an interpretable TF-IDF plus Logistic Regression baseline beside the neural model to see which lexical cues drive each label
- Report macro scores alongside accuracy and inspect where errors land; here most mistakes stayed inside neighbouring categories
- Read mispredictions as failures to capture pragmatic force such as intensity, tone or hyperbole rather than lexical meaning

## For Mova

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

- **I1** (frames): emoji-bearing sentences could serve as weak silver signal for checking prag's speech-act, sentiment and irony judgments against what the writer actually signalled.
- **I2** (world): verb classes and ontology nodes with few examples could be backed off to a parent class during learning, with the fine label kept for explanation.
- **I3** (learn): a transparent cue-weight baseline next to gparse or prag classifiers would expose when the strong model is merely exploiting surface words, acting as a cheap negative control.
- **I4** (learn): evaluation could separate near-miss errors (adjacent frame or role) from gross ones, so repair rules target real failures rather than inherent label overlap.
- **I5** (frames): hyperbolic idioms like "killing me" or "I'll die for you" are a concrete test set for prag's irony and exaggeration handling, checking that literal frames are not taken at face value.

*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: [🐪 Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Institutions

- [Ibb University](https://md.research.elamur.ai/free/en/universities/ibb-university/) — University, Ibb, Yemen · 4 works · 6 citations

## Cite

Cite as: Mohammed Q. Shormani and Ibrahim Abdulmalik Hassan Muneef Y. Alshawsh. 2026. Machine learning and emoji prediction: How much accuracy can MARBERT achieve?. arXiv:2604.21108.

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

```bibtex
@misc{shormani2026machine,
  title = {{Machine learning and emoji prediction: How much accuracy can MARBERT achieve?}},
  author = {Mohammed Q. Shormani and Ibrahim Abdulmalik Hassan Muneef Y. Alshawsh},
  year = {2026},
  doi = {10.48550/arxiv.2604.21108},
  eprint = {2604.21108},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2604.21108}
}
```

- [← previous KZ-SafetyPrompts: A Kazakh Safety Evaluation Prompt Dataset for Large Language Models](https://md.research.elamur.ai/free/en/papers/zaghouani-2026-kz-safetyprompts-kazakh-safety-evaluation/)
- [next → MENASpeechBank: A Reference Voice Bank with Persona-Conditioned Multi-Turn Conversations for AudioLLMs](https://md.research.elamur.ai/free/en/papers/ali-2026-menaspeechbank-reference-voice-bank-persona/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/)

## References

40 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Abdul-Mageed, M., & Elmadany, A. (2021). ARBERT & MARBERT: Deep bidirectional transformers for Arabic. In 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) (pp. 7088-7105). [our card](https://md.research.elamur.ai/free/en/papers/abdul-mageed-2021-arbert-marbert-deep-bidirectional-transformers/)
2. Abdul-Mageed, M., Elmadany, M. & Nagoudi, E. M. B. (2022). CAMeLBERT: A Collection of Pre-trained Models for Arabic NLP. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics (pp. 628–640). [find](https://scholar.google.com/scholar?q=Abdul%2DMageed%2C+M%2E%2C+Elmadany%2C+M%2E+%26+Nagoudi%2C+E%2E+M%2E+B%2E+%282022%29%2E+CAMeLBERT%3A+A+Collection+of+Pre%2Dtrained+Models+for+Arabic+NLP%2E+Proceedings+of+the+16th+Conference+of+the+European+Chapter+of+the+Association+f)
3. Aboutaib, A., El Allaoui, A., Zeroual, I. (2025). A Comparative analysis of transformer models for the prediction of Arabic punctuation. In Farhaoui, Y., Herawan, T., Lucky Imoize, A., Allaoui, A.E. (eds) Intersection of artificial intelligence, data science, and cutting-edge technologies: From concepts to applications in smart environment. ICAISE 2024. Lecture Notes in Networks and Systems, vol 1397. Springer, Cham. https://doi.org/10.1007/978-3-031-90921-4_96 [↗](https://doi.org/10.1007/978-3-031-90921-4_96)
4. Ahamad, R., & Mishra, K. N. (2025). Exploring sentiment analysis in handwritten and E-text documents using advanced machine learning techniques: a novel approach. Journal of Big Data, 12(1), 11. [find](https://scholar.google.com/scholar?q=Ahamad%2C+R%2E%2C+%26+Mishra%2C+K%2E+N%2E+%282025%29%2E+Exploring+sentiment+analysis+in+handwritten+and+E%2Dtext+documents+using+advanced+machine+learning+techniques%3A+a+novel+approach%2E+Journal+of+Big+Data%2C+12%281%29%2C+11%2E)
5. Antoun, W. Baly, F. & Hajj, H. (2020). AraGPT2: Pre-trained Transformer for Arabic Language Generation. arXiv preprint arXiv:2012.15520, 2020. [↗](https://arxiv.org/abs/2012.15520)
6. Barbieri, F., Ballesteros, M., & Saggion, H. (2017). Are emojis predictable? In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers (pp. 105-111). [find](https://scholar.google.com/scholar?q=Barbieri%2C+F%2E%2C+Ballesteros%2C+M%2E%2C+%26+Saggion%2C+H%2E+%282017%29%2E+Are+emojis+predictable%3F+In+Proceedings+of+the+15th+Conference+of+the+European+Chapter+of+the+Association+for+Computational+Linguistics%3A+Volume+2%2C+S)
7. Bassiouney, R. (2009). Arabic sociolinguistics. Edinburgh University Press. [find](https://scholar.google.com/scholar?q=Bassiouney%2C+R%2E+%282009%29%2E+Arabic+sociolinguistics%2E+Edinburgh+University+Press%2E)
8. Chomsky, N. (1956). Three models for the description of language. IRE Transactions on Information Theory, 2(3), 113–124. https://doi.org/10.1109/TIT.1956.1056813 [↗](https://doi.org/10.1109/tit.1956.1056813)
9. Chomsky, N. (1957). Syntactic structures. The Hague: Mouton. [find](https://scholar.google.com/scholar?q=Chomsky%2C+N%2E+%281957%29%2E+Syntactic+structures%2E+The+Hague%3A+Mouton%2E)
10. Chomsky, N. (2013). Problems of projection. Lingua, 130, 33-49. https://doi.org/10.1016/j.lingua.2012.12.003 Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., ... & Stoyanov, V. (2020, July). Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th annual meeting of the association for computational linguistics (pp. 8440-8451). [↗](https://doi.org/10.1016/j.lingua.2012.12.003)
11. Dongo, I., Cardinale, Y., Aguilera, A., Martinez, F., Quintero, Y., Robayo, G., & Cabeza, D. (2021). A qualitative and quantitative comparison between Web scraping and API methods for Twitter credibility analysis. International Journal of Web Information Systems, 17(6), 580-606. [find](https://scholar.google.com/scholar?q=Dongo%2C+I%2E%2C+Cardinale%2C+Y%2E%2C+Aguilera%2C+A%2E%2C+Martinez%2C+F%2E%2C+Quintero%2C+Y%2E%2C+Robayo%2C+G%2E%2C+%26+Cabeza%2C+D%2E+%282021%29%2E+A+qualitative+and+quantitative+comparison+between+Web+scraping+and+API+methods+for+Twitter+credibil)
12. Drous, H. A., Barakat, H., & Atta, A. B. (2025). Finetuning MARBERT for Sentiment Analysis in Jordanian Arabic Dialects Using a Synthetic Dialectal Corpus. Journal of Computer Science and Technology Studies, 7(11), 373-386. [find](https://scholar.google.com/scholar?q=Drous%2C+H%2E+A%2E%2C+Barakat%2C+H%2E%2C+%26+Atta%2C+A%2E+B%2E+%282025%29%2E+Finetuning+MARBERT+for+Sentiment+Analysis+in+Jordanian+Arabic+Dialects+Using+a+Synthetic+Dialectal+Corpus%2E+Journal+of+Computer+Science+and+Technology+S)
13. Goyal, P., Pandey, S., & Jain, K. (2018). Deep learning for natural language processing. New York: Apress, 1624-1633. [find](https://scholar.google.com/scholar?q=Goyal%2C+P%2E%2C+Pandey%2C+S%2E%2C+%26+Jain%2C+K%2E+%282018%29%2E+Deep+learning+for+natural+language+processing%2E+New+York%3A+Apress%2C+1624%2D1633%2E)
14. Hajba, G. L. (2018). Website Scraping with Python. Apress, Springer. [find](https://scholar.google.com/scholar?q=Hajba%2C+G%2E+L%2E+%282018%29%2E+Website+Scraping+with+Python%2E+Apress%2C+Springer%2E)
15. Hernandez-Suarez, A., Sanchez-Perez, G., Toscano-Medina, K., Martinez-Hernandez, V., Sanchez, V., & Perez-Meana, H. (2018). A web scraping methodology for bypassing twitter API restrictions. arXiv preprint arXiv:1803.09875 Holes, C. (2004). Modern Arabic: Structures, functions, and varieties. Georgetown University Press. [↗](https://arxiv.org/abs/1803.09875)
16. Kanani, P., & Padole, M. (2019). Deep learning to detect skin cancer using google colab. International Journal of Engineering and Advanced Technology Regular Issue, 8(6), 2176-2183. [find](https://scholar.google.com/scholar?q=Kanani%2C+P%2E%2C+%26+Padole%2C+M%2E+%282019%29%2E+Deep+learning+to+detect+skin+cancer+using+google+colab%2E+International+Journal+of+Engineering+and+Advanced+Technology+Regular+Issue%2C+8%286%29%2C+2176%2D2183%2E)
17. Khder, M. A. (2021). Web scraping or web crawling: State of art, techniques, approaches and application. International Journal of Advances in Soft Computing & Its Applications, 13(3). [find](https://scholar.google.com/scholar?q=Khder%2C+M%2E+A%2E+%282021%29%2E+Web+scraping+or+web+crawling%3A+State+of+art%2C+techniques%2C+approaches+and+application%2E+International+Journal+of+Advances+in+Soft+Computing+%26+Its+Applications%2C+13%283%29%2E)
18. Levy, J. J., & O’Malley, A. J. (2020). Don’t dismiss logistic regression: the case for sensible extraction of interactions in the era of machine learning. BMC Medical Research Methodology, 20(1), 171. [find](https://scholar.google.com/scholar?q=Levy%2C+J%2E+J%2E%2C+%26+O%E2%80%99Malley%2C+A%2E+J%2E+%282020%29%2E+Don%E2%80%99t+dismiss+logistic+regression%3A+the+case+for+sensible+extraction+of+interactions+in+the+era+of+machine+learning%2E+BMC+Medical+Research+Methodology%2C+20%281%29%2C+171%2E)
19. Mahardhika, I. E. P., Prayitno, H. J., Indri, I., & Fitriyan, M. R. (2025). Visual and contextual learning for deep learning education: A unified tool for theory–practice integration. Journal of Deep Learning, 69-80. [find](https://scholar.google.com/scholar?q=Mahardhika%2C+I%2E+E%2E+P%2E%2C+Prayitno%2C+H%2E+J%2E%2C+Indri%2C+I%2E%2C+%26+Fitriyan%2C+M%2E+R%2E+%282025%29%2E+Visual+and+contextual+learning+for+deep+learning+education%3A+A+unified+tool+for+theory%E2%80%93practice+integration%2E+Journal+of+Deep+)
20. McShane, M., Nirenburg, S. (2021). Linguistics for the age of AI. MIT Press. [find](https://scholar.google.com/scholar?q=McShane%2C+M%2E%2C+Nirenburg%2C+S%2E+%282021%29%2E+Linguistics+for+the+age+of+AI%2E+MIT+Press%2E)
21. Murphy, K. P. (2012). Machine Learning: A probabilistic perspective. MIT Press. [find](https://scholar.google.com/scholar?q=Murphy%2C+K%2E+P%2E+%282012%29%2E+Machine+Learning%3A+A+probabilistic+perspective%2E+MIT+Press%2E)
22. Naithani, K., & Raiwani, Y. P. (2023). Realization of natural language processing and machine learning approaches for text‐based sentiment analysis. Expert Systems, 40(5), e13114. [find](https://scholar.google.com/scholar?q=Naithani%2C+K%2E%2C+%26+Raiwani%2C+Y%2E+P%2E+%282023%29%2E+Realization+of+natural+language+processing+and+machine+learning+approaches+for+text%E2%80%90based+sentiment+analysis%2E+Expert+Systems%2C+40%285%29%2C+e13114%2E)
23. Nusrat, M. O., Habib, Z., Alam, M., & Jamal, S. A. (2023). Emoji prediction in tweets using BERT. arXiv preprint arXiv:2307.02054. [↗](https://arxiv.org/abs/2307.02054)
24. Samuel, A. L. (1953). Computing Bit by Bit or digital computers made easy. Proceedings of the IRE 41 (10): 1223. https://doi.org/doi:10.1109/JRPROC.1953.274271 [↗](https://doi.org/10.1109/jrproc.1953.274271)
25. Sarker, I. H. (2021). Machine learning: Algorithms, real-world applications and research directions. SN Computer Science, 2(3), 1-21. [find](https://scholar.google.com/scholar?q=Sarker%2C+I%2E+H%2E+%282021%29%2E+Machine+learning%3A+Algorithms%2C+real%2Dworld+applications+and+research+directions%2E+SN+Computer+Science%2C+2%283%29%2C+1%2D21%2E)
26. Sohail, S.S., Farhat, F., Himeur, Y., Nadeem, M., Madsen, D., Singh, Y., Atalla, S. & Mansoor, W. (2023). Decoding ChatGPT: A taxonomy of existing research, current challenges, and possible future directions. J King Saud Univer Comput Inform Sci. https://doi.org/10.1016/j.jksuci.2023.101675 [↗](https://doi.org/10.1016/j.jksuci.2023.101675)
27. Shormani, M. Q. (2014). Collocability difficulty: A UG-based model for stable acquisition. International Journal of Literature, Language and Linguistics, 4, 54–64. [find](https://scholar.google.com/scholar?q=Shormani%2C+M%2E+Q%2E+%282014%29%2E+Collocability+difficulty%3A+A+UG%2Dbased+model+for+stable+acquisition%2E+International+Journal+of+Literature%2C+Language+and+Linguistics%2C+4%2C+54%E2%80%9364%2E)
28. Shormani, M. Q. (2024). Generative linguistics contribution to artificial intelligence: Where this contribution lies? arXiv. https://doi.org/10.48550/arXiv.2410.20221 [↗](https://doi.org/10.48550/arxiv.2410.20221)
29. Shormani, M. Q. (2025a). Generative approaches to language acquisition: Issues from Arab context. Lincom Europa Press. [find](https://scholar.google.com/scholar?q=Shormani%2C+M%2E+Q%2E+%282025a%29%2E+Generative+approaches+to+language+acquisition%3A+Issues+from+Arab+context%2E+Lincom+Europa+Press%2E)
30. Shormani, M. Q. (2025b). What fifty-one years of linguistics and artificial intelligence research tell us about their correlation: A scientometric analysis. Artif Intell Rev 58, 379. https://doi.org/10.1007/s10462-025-11332-5 [↗](https://doi.org/10.1007/s10462-025-11332-5)
31. Shormani, M.Q. & Alenezi, M. A. (2026). Arabic Nominalization as Form of Pragmatic Actions in Digital Discourse: A Corpus-Based Study. Corpus Pragmatics. https://doi.org/10.1007/s41701-026-00237-5 [↗](https://doi.org/10.1007/s41701-026-00237-5)
32. Taye, M. M. (2023). Understanding of machine learning with deep learning: architectures, workflow, applications and future directions. Computers, 12(5), 91. [find](https://scholar.google.com/scholar?q=Taye%2C+M%2E+M%2E+%282023%29%2E+Understanding+of+machine+learning+with+deep+learning%3A+architectures%2C+workflow%2C+applications+and+future+directions%2E+Computers%2C+12%285%29%2C+91%2E)
33. Thomas, D. M., & Mathur, S. (2019). Data analysis by web scraping using python. In 2019 3rd International conference on Electronics, Communication and Aerospace Technology (ICECA) (pp. 450-454). IEEE. [find](https://scholar.google.com/scholar?q=Thomas%2C+D%2E+M%2E%2C+%26+Mathur%2C+S%2E+%282019%29%2E+Data+analysis+by+web+scraping+using+python%2E+In+2019+3rd+International+conference+on+Electronics%2C+Communication+and+Aerospace+Technology+%28ICECA%29+%28pp%2E+450%2D454%29%2E+IEEE%2E)
34. Turing, A. M. (1950). Can Machines think. Mind, 59(236), 433-460. [find](https://scholar.google.com/scholar?q=Turing%2C+A%2E+M%2E+%281950%29%2E+Can+Machines+think%2E+Mind%2C+59%28236%29%2C+433%2D460%2E)
35. Ullah, N., Khan, H., Qazi, A. M., Zeb, S., & Rabi, U. (2025). Emojis as a semiotic system: exploring their influence on linguistic simplification and emotional clarity in digital interactions. Journal of Media Horizons, 6(3), 2358-2379. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30. [our card](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/)
36. Venkit, P., Karishma, Z., Hsu, C. Y., Katiki, R., Huang, K., Wilson, S., & Dudas, P. (2021). A Sourceful Twist: Emoji prediction based on sentiment, hashtags and application source. arXiv preprint arXiv:2103.07833. [↗](https://arxiv.org/abs/2103.07833)
37. Winograd, T. (1971). Procedures as a representation for data in a computer program for understanding natural language. Unpublished PhD Thesis. Cambridge, MA: MIT. [find](https://scholar.google.com/scholar?q=Winograd%2C+T%2E+%281971%29%2E+Procedures+as+a+representation+for+data+in+a+computer+program+for+understanding+natural+language%2E+Unpublished+PhD+Thesis%2E+Cambridge%2C+MA%3A+MIT%2E)
38. Yus, F. (2014). Not All Emoticons Are Created Equal. Linguagem em (Dis)Curso, 14(3), 511–529. http://dx.doi.org/10.1590/1982-4017-140304-0414 [↗](https://doi.org/10.1590/1982-4017-140304-0414)
39. Yus, F. (2025). Emoji Pragmatics. Springer Nature. [find](https://scholar.google.com/scholar?q=Yus%2C+F%2E+%282025%29%2E+Emoji+Pragmatics%2E+Springer+Nature%2E)
40. Zhao, P., Jia, J., An, Y., Liang, J., Xie, L., & Luo, J. (2018). Analyzing and predicting emoji usages in social media. In Companion Proceedings of the Web Conference 2018 (pp. 327-334). [find](https://scholar.google.com/scholar?q=Zhao%2C+P%2E%2C+Jia%2C+J%2E%2C+An%2C+Y%2E%2C+Liang%2C+J%2E%2C+Xie%2C+L%2E%2C+%26+Luo%2C+J%2E+%282018%29%2E+Analyzing+and+predicting+emoji+usages+in+social+media%2E+In+Companion+Proceedings+of+the+Web+Conference+2018+%28pp%2E+327%2D334%29%2E)