Elamur Research⌕
☰
home / papers / 2025 / On Using Arabic Language Dialects in Recommendation Systems

On Using Arabic Language Dialects in Recommendation Systems

Paper · 2025 · Abdulla Alshabanah and Murali Annavaram

Summary

This short paper asks whether the dialect a person writes in can help a recommender predict what they will engage with. The authors target Arabic, where regional varieties differ sharply from Modern Standard Arabic (MSA) and from each other. They take off-the-shelf BERT-based tools: single-label and multi-label Arabic Dialect Identification (ADI) classifiers and an Arabic Level of Dialectness (ALDi) regressor. They apply these to reviews in two public Arabic review collections, BRAD and LABR, and plug the output into a Two-Tower Neural Network (TTNN) trained on implicit feedback. They test two strategies:

  • Explicit: dialect labels become a user feature. This mostly hurts, except for one variant that keeps the feature identical across positive and negative examples.
  • Implicit: items the user reviewed are sampled with probability proportional to the review's ALDi score. Their embeddings are pooled and appended to the user vector.

The implicit strategy gives sizable gains in HR@10 and NDCG@10 on both datasets. The work matters as an early demonstration that sociolinguistic variation in low-resource dialects can serve as a useful modeling signal outside core NLP tasks.

Key points

  • Tools: Dialect identification relies on BERT-based ADI classifiers in both single-label and multi-label versions (Keleg and Magdy, 2023), and dialectness is estimated with a BERT-based ALDi model (Keleg et al., 2023). Any review with an ALDi score below 0.1 is labeled MSA.
  • Explicit variants that change the feature between positive and negative interactions perform badly.
    • On BRAD, HR@10 drops by up to 70% (Single-Label and Multi-Label).
    • A sentence-transformer review baseline (Ar-Review) also drops, by 59% on BRAD and 46% on LABR.
  • Multi-Label++ is the only explicit variant that helps. It assigns each user's two most frequent dialects to every interaction. It yields modest gains: +3% HR@10 on BRAD and +6% on LABR.
  • Implicit sampling gives the largest improvements.
    • Implicit uniform reaches +27% HR@10 and +30% NDCG@10 on BRAD.
    • Implicit weighted, which uses scaled dot-product attention over the sampled items, reaches +19% HR@10 on LABR.
  • Ablations and subgroup checks support the implicit approach.
    • Replacing ALDi-proportional sampling with random sampling lowers results.
    • Users who write mostly in MSA (average ALDi < 0.1) still improve, so there is no bias against them.

Interesting ideas

  • I1. Instead of a hard dialect label, the paper uses a continuous "how non-standard is this text" score, which captures mixed registers better than a single class.
  • I2. When a feature is computed differently for positive and negative examples, the model learns the mismatch itself rather than the preference, which wrecked several variants here.
  • I3. Weighting which examples get sampled by an informativeness score, and checking it against random sampling as an ablation, cleanly isolates the effect of the score.
  • I4. Multi-label dialect tagging accepts that one sentence can be valid in several varieties at once, avoiding forced single choices.
  • I5. The authors split users by the very score the method relies on, to confirm that the low-score group is not harmed.

For Mova

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

  • I1 (tokens): a per-sentence informality/non-standardness score could route text to stricter or more lenient normalization and morphology, and explain why a lemma choice was made.
  • I2 (learn): when building negative controls, compute every feature through exactly the same pipeline as for positives, or the evaluation may reward an artifact.
  • I3 (learn): sample silver sentences in proportion to witness-parser disagreement or rarity, and always report the uniform-sampling baseline alongside.
  • I4 (tree): during UD dialect conversion, let a sentence or arc be marked as compatible with several annotation conventions instead of forcing one, and keep the alternatives explainable.
  • I5 (answer): after adding any score-driven heuristic, report accuracy separately for low- and high-score items (e.g., plain vs figurative fables) to catch silent regressions.

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

🎓University of Southern CaliforniaLos Angeles · United States162 works · 35,836 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksInformation Systems · 1,437 works

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

Cite

Cite as Abdulla Alshabanah and Murali Annavaram. 2025. On Using Arabic Language Dialects in Recommendation Systems. Findings of the Association for Computational Linguistics: NAACL 2025.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.findings-naacl.115

@inproceedings{alshabanah2025using,
  title = {{On Using Arabic Language Dialects in Recommendation Systems}},
  author = {Abdulla Alshabanah and Murali Annavaram},
  booktitle = {Findings of the Association for Computational Linguistics: NAACL 2025},
  year = {2025},
  doi = {10.18653/v1/2025.findings-naacl.115},
  url = {https://doi.org/10.18653/v1/2025.findings-naacl.115}
}

References

42 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Muhammad Abdul-Mageed, Abdelrahim Elmadany, Chiyu Zhang, Houda Bouamor, Nizar Habash, et al. 2023. Nadi 2023: The fourth nuanced arabic dialect identification shared task. In Proceedings of Arabic-NLP 2023, pages 600–613.
  2. Muhammad Abdul-Mageed, Amr Keleg, Abdelrahim Elmadany, Chiyu Zhang, Injy Hamed, Walid Magdy, Houda Bouamor, and Nizar Habash. 2024. Nadi 2024: The fifth nuanced arabic dialect identification shared task. In Proceedings of The Second Arabic Natural Language Processing Conference, pages 709– 728.
  3. Muhammad Abdul-Mageed, Chiyu Zhang, Abdelrahim Elmadany, Houda Bouamor, and Nizar Habash. 2021. Nadi 2021: The second nuanced arabic dialect identification shared task. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 244–259. ↗
  4. Muhammad Abdul-Mageed, Chiyu Zhang, Abdelrahim Elmadany, Houda Bouamor, and Nizar Habash. 2022. Nadi 2022: The third nuanced arabic dialect identification shared task. In Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP), pages 85–97.
  5. Amjad Almahairi, Kyle Kastner, Kyunghyun Cho, and Aaron Courville. 2015. Learning distributed representations from reviews for collaborative filtering. In Proceedings of the 9th ACM Conference on Recommender Systems, pages 147–154. find
  6. Maha Jarallah Althobaiti. 2020. Automatic arabic dialect identification systems for written texts: A survey. International Journal of Computational Linguistics (IJCL), 11(3):61–89. Revised: 31-10-2020; Published: 01-12-2020. ↗
  7. Mohamed Aly and Amir Atiya. 2013. Labr: A large scale arabic book reviews dataset. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 494–498. ↗
  8. Keshav Balasubramanian, Abdulla Alshabanah, Elan Markowitz, Greg Ver Steeg, and Murali Annavaram. 2024. Biased user history synthesis for personalized long-tail item recommendation. In Proceedings of the 18th ACM Conference on Recommender Systems, RecSys ’24, page 189–199, New York, NY, USA. Association for Computing Machinery. find
  9. Reem Bassiouney. 2006. Functions of code switching in Egypt: Evidence from monologues, volume 46. Brill. find
  10. A. Bergman and Mona Diab. 2022. Towards responsible natural language annotation for the varieties of Arabic. In Findings of the Association for Computational Linguistics: ACL 2022, pages 364–371, Dublin, Ireland. Association for Computational Linguistics.
  11. Houda Bouamor, Sabit Hassan, and Nizar Habash. 2019. The MADAR shared task on Arabic fine-grained dialect identification. In Proceedings of the Fourth Arabic Natural Language Processing Workshop, pages 199–207, Florence, Italy. Association for Computational Linguistics.
  12. Rose Catherine and William Cohen. 2017. Transnets: Learning to transform for recommendation. In Proceedings of the eleventh ACM conference on recommender systems, pages 288–296. find
  13. Li Chen, Guanliang Chen, and Feng Wang. 2015. Recommender systems based on user reviews: the state of the art. User Modeling and User-Adapted Interaction, 25:99–154. ↗
  14. Mingyue Cheng, Fajie Yuan, Qi Liu, Shenyang Ge, Zhi Li, Runlong Yu, Defu Lian, Senchao Yuan, and Enhong Chen. 2021. Learning recommender systems with implicit feedback via soft target enhancement. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’21, page 575–584, New York, NY, USA. Association for Computing Machinery. find
  15. Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM Conference on Recommender Systems, New York, NY, USA. find
  16. Ashraf Elnagar and Omar Einea. 2016. Brad 1.0: Book reviews in arabic dataset. In 2016 IEEE/ACS 13th International Conference of Computer Systems and Applications (AICCSA), pages 1–8. IEEE. find
  17. 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
  18. Charles A. Ferguson. 1959. 15(2):325–340. Diglossia. find
  19. WORD, Carlos A. Gomez-Uribe and Neil Hunt. 2016. The netflix recommender system: Algorithms, business value, and innovation. ACM Trans. Manage. Inf. Syst., 6(4). ↗
  20. Asnat Greenstein-Messica and Lior Rokach. 2018. Personal price aware multi-seller recommender system: Evidence from ebay. Know.-Based Syst., 150(C):14–26. ↗
  21. Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. CoRR, abs/1708.05031. ↗
  22. Dan Hendrycks and Kevin Gimpel. 2016. Bridging nonlinearities and stochastic regularizers with gaussian error linear units. CoRR, abs/1606.08415. ↗
  23. Barbara Johnstone and Judith Mattson Bean. 1997. Self-expression and linguistic variation. Language in Society, 26(2):221–246. ↗
  24. Amr Keleg, Sharon Goldwater, and Walid Magdy. 2023. ALDi: Quantifying the Arabic level of dialectness of text. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 10597–10611, Singapore. Association for Computational Linguistics.
  25. Amr Keleg and Walid Magdy. 2023. Arabic dialect identification under scrutiny: Limitations of single-label classification. In Proceedings of ArabicNLP 2023, pages 385–398, Singapore (Hybrid). Association for Computational Linguistics.
  26. Amr Keleg, Walid Magdy, and Sharon Goldwater. 2024. Estimating the level of dialectness predicts inter-annotator agreement in multi-dialect Arabic datasets. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 766–777, Bangkok, Thailand. Association for Computational Linguistics.
  27. Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. ↗
  28. G. Linden, B. Smith, and J. York. 2003. Amazon.com recommendations: item-to-item collaborative filtering. IEEE Internet Computing, 7(1):76–80. ↗
  29. Omer Nacar. 2024. Arabic matryoshka embeddings model - arabic minilm l12 v2 all nli triplet. find
  30. Maxim Naumov, Dheevatsa Mudigere, Hao-Jun Michael Shi, Jianyu Huang, Narayanan Sundaraman, Jongsoo Park, Xiaodong Wang, Udit Gupta, Carole-Jean Wu, Alisson G. Azzolini, Dmytro Dzhulgakov, Andrey Mallevich, Ilia Cherniavskii, Yinghai Lu, Raghuraman Krishnamoorthi, Ansha Yu, Volodymyr Kondratenko, Stephanie Pereira, Xianjie Chen, Wenlin Chen, Vijay Rao, Bill Jia, Liang Xiong, and Misha Smelyanskiy. 2019. Deep learning recommendation model for personalization and recommendation systems. CoRR, abs/1906.00091. ↗
  31. Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32. ↗
  32. Gang Qiu, Yanli Guo, and Changjun Song. 2022. Tann: Text-based attention neural network recommendation model. In 2022 IEEE 8th Intl Conference on Big Data Security on Cloud (BigDataSecurity), IEEE Intl Conference on High Performance and Smart Computing,(HPSC) and IEEE Intl Conference on Intelligent Data and Security (IDS), pages 119–124. IEEE. find
  33. Xi Wang, Iadh Ounis, and Craig Macdonald. 2021b. Leveraging review properties for effective recommendation. In Proceedings of the Web Conference 2021, pages 2209–2219. find
  34. Xinyang Yi, Ji Yang, Lichan Hong, Derek Zhiyuan Cheng, Lukasz Heldt, Aditee Ajit Kumthekar, Zhe Zhao, Li Wei, and Ed Chi, editors. 2019. Sampling-Bias-Corrected Neural Modeling for Large Corpus Item Recommendations. find
  35. Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L. Hamilton, and Jure Leskovec. 2018. Graph convolutional neural networks for web-scale recommender systems. CoRR, abs/1806.01973. ↗
  36. Yin Zhang, Derek Zhiyuan Cheng, Tiansheng Yao, Xinyang Yi, Lichan Hong, and Ed H. Chi. 2020. A model of two tales: Dual transfer learning framework for improved long-tail item recommendation. CoRR, abs/2010.15982. ↗
  37. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
  38. Noveen Sachdeva and Julian McAuley. 2020. How useful are reviews for recommendation? a critical review and potential improvements. In proceedings of the 43rd international ACM SIGIR conference on research and development in information retrieval, pages 1845–1848. find
  39. Xuehan Sun, Tianyao Shi, Xiaofeng Gao, Yanrong Kang, and Guihai Chen. 2021. FORM: Follow the Online Regularized Meta-Leader for Cold-Start Recommendation, page 1177–1186. Association for Computing Machinery, New York, NY, USA. find
  40. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. CoRR, abs/1706.03762. ↗
  41. Mengting Wan and Julian McAuley. 2018. Item recommendation on monotonic behavior chains. In Proceedings of the 12th ACM Conference on Recommender Systems, RecSys ’18, page 86–94, New York, NY, USA. Association for Computing Machinery. find
  42. Jianling Wang, Kaize Ding, and James Caverlee. 2021a. Sequential recommendation for cold-start users with meta transitional learning. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’21, page 1783–1787, New York, NY, USA. Association for Computing Machinery. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alshabanah-2025-arabic-language-dialects-recommendation-systems/