Elamur Research⌕
☰
home / papers / 2026 / UPR at SemEval-2026 Task 9: Polarization Detection in Urdu with Language-Specific Transformer and Data Augmentation

UPR at SemEval-2026 Task 9: Polarization Detection in Urdu with Language-Specific Transformer and Data Augmentation

Paper · 2026 · Alishba Wazir, Muhammad Asad Khan, Junaid Rashid et al.

Summary

This paper describes the University of Poonch Rawalakot (UPR) system for SemEval-2026 Task 9, Subtask 1. The task is to decide whether an Urdu social media post is polarized or not. Urdu has little labeled data, the posts are noisy and informal, and the task data leans toward the polarized class. The authors fine-tune the monolingual Urdu-BERT checkpoint eshaaftab900/urdu-bert-base-2, adding a linear classifier on the [CLS] representation.

Before training, they normalize text with UrduHack, strip URLs and non-Urdu characters, and drop near-duplicates using TF-IDF cosine similarity. They enlarge the minority class through synonym substitution from a hand-built dictionary, local word reordering, and random deletion.

Urdu-BERT reaches a Macro-F1 of 0.989, ahead of mBERT (0.979), XLM-RoBERTa (0.780), and TF-IDF classifiers (at most 0.674). These are validation scores. The 10,379 augmented posts appear to be split into the training and validation sets, so the scores may be inflated. Results on the official evaluation set are not reported separately. The work nonetheless suggests that a monolingual model is a sensible baseline for low-resource polarization detection.

Key points

  • Data: the SemEval-2026 Task 9 Urdu set had 3,561 posts after cleaning (2,475 polarized, 1,086 non-polarized, ratio about 2.28:1). Deduplication at a cosine threshold of ≥0.95 removed only 2 posts.
  • After augmentation there were 10,379 posts (5,611 polarized and 4,768 non-polarized), split into 8,822 training and 1,557 validation posts. A separate 177-post evaluation set was held out.
  • Setup: BERT-base architecture (12 layers, 12 heads, hidden size 768) with WordPiece tokens capped at 256. Training used AdamW at a learning rate of 2e-5, batch size 16, and up to 20 epochs with early stopping (patience 3). The best Macro-F1 came at epoch 13.
  • Baselines: Naive Bayes, Linear SVM, RBF SVM, and Logistic Regression on TF-IDF features, plus the multilingual transformers XLM-RoBERTa and mBERT.
  • Leakage check: train/test vocabulary Jaccard similarity was 0.232, and 71% of test tokens appear in training. The authors acknowledge that augmentation artifacts and the use of a single architecture are limitations.

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

🎓Sejong UniversitySeoul · South Korea2 works · 1 citation🎓University of Poonch RawalakotRawlakot · Pakistan2 works · 1 citation

 Fields, topics and concepts

📚 Shelves 4The 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 works

Cite

Cite as Alishba Wazir, Muhammad Asad Khan, Junaid Rashid, Shamaila Hayat and Samira Kanwal. 2026. UPR at SemEval-2026 Task 9: Polarization Detection in Urdu with Language-Specific Transformer and Data Augmentation. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.332

@inproceedings{wazir2026upr,
  title = {{UPR at SemEval-2026 Task 9: Polarization Detection in Urdu with Language-Specific Transformer and Data Augmentation}},
  author = {Alishba Wazir, Muhammad Asad Khan, Junaid Rashid, Shamaila Hayat and Samira Kanwal},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.332},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.332}
}

References

23 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Naeem Ahmed, Rashid Amin, Hamza Aldabbas, Muhammad Saeed, Muhammad Bilal, and Houbing Song. 2024. A novel approach for sentiment analysis of a low resource language using deep learning models. ACM Transactions on Asian and Low-Resource Language Information Processing. find
  2. Muhammad Rehan Ashraf, Muzammal Hussain, M Arfan Jaffar, Waheed Yousuf Ramay, and Muhammad Faheem. 2024. Revolutionizing urdu sentiment analysis: Harnessing the power of xlm-r and gpt-2. IEEE Access, 12:99779–99793. find
  3. Kamran Aziz, Donghong Ji, Bobo Li, Fei Li, and Jun Zhou. 2024. Advancing urdu nlp: Aspect-based sentiment analysis with graph attention networks. In 2024 International Joint Conference on Neural Networks (IJCNN), pages 1–8. IEEE. find
  4. Wei Yen Chong, Bhawani Selvaretnam, and Lay-Ki Soon. 2014. Natural language processing for sentiment analysis: an exploratory analysis on tweets. In 2014 4th international conference on artificial intelligence with applications in engineering and technology, pages 212–217. IEEE. find
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th annual meeting of the association fo… find
  6. Corinna Cortes and Vladimir Vapnik. 1995. Support-vector networks. Machine learning, 20(3):273–297.
  7. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,…
  8. Parush Gera and Tempestt Neal. 2025. Deep learning in stance detection: A survey. ACM Computing Surveys, 58(1):1–37. find
  9. David W Hosmer Jr, Stanley Lemeshow, and Rodney X Sturdivant. 2013. Applied logistic regression. John Wiley & Sons. find
  10. Asad Khattak, Muhammad Zubair Asghar, Anam Saeed, Ibrahim A. Hameed, Syed Asif Hassan, and Shakeel Ahmad. 2021. A survey on sentiment analysis in urdu: A resource-poor language. Egyptian Informatics Journal, 22(1):53–74. find
  11. Saif Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiaodan Zhu, and Colin Cherry. 2016. SemEval-2016 task 6: Detecting stance in tweets. In Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016), pages 31– 41, San Diego, California. Association for Computational Ling…
  12. Khalid Bin Muhammad and SM Aqil Burney. 2023. Innovations in urdu sentiment analysis using machine and deep learning techniques for two-class classification of symmetric datasets. Symmetry, 15(5):1027. find
  13. Uzma Naqvi, Abdul Majid, and Syed Ali Abbas. 2021. Utsa: Urdu text sentiment analysis using deep learning methods. IEEE Access, 9:114085–114094. find
  14. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, "Ozge Alacam, Cengiz Acar"urk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Elena Tutubalina, Aung Kyaw Htet, Xintong Wang, Surendrabikram Thapa, Tanmoy Chak… — SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization
  15. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, Özge Alacam, Cengiz Acartürk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Simona Frenda, Alessandra Teresa Cignarella, Elena Tutubalina, Oleg Rogov, Aung Ky… — POLAR: A Benchmark for Multilingual, Multicultural, and Multi-Event Online Polarization
  16. Lata Pangtey, Anukriti Bhatnagar, Shubhi Bansal, Shahid Shafi Dar, and Nagendra Kumar. 2025. Large language models meet stance detection: A survey of tasks, methods, applications, challenges and future directions. arXiv preprint arXiv:2505.08464. find
  17. AM Muntasir Rahman, Wenpeng Yin, and Guiling Wang. 2023. Data augmentation for text classification with ease. In Proceedings of the 6th International Conference on Natural Language and Speech Processing (ICNLSP 2023), pages 324–332. find
  18. Irina Rish and 1 others. 2001. An empirical study of the naive bayes classifier. In IJCAI 2001 workshop on empirical methods in artificial intelligence, volume 3, pages 41–46. Seattle, USA. find
  19. Mamoona Shabbir and Muhammad Majid. 2024. Sentiment analysis from urdu language-based text using deep learning techniques. In 2024 5th International Conference on Advancements in Computational Sciences (ICACS), pages 1–5. IEEE. find
  20. Neha Singh and Umesh Chandra Jaiswal. 2023. Sentiment analysis based on urdu reviews using hybrid deep learning models. Appl. Comput. Syst., 28(2):258–265. find
  21. Jason Wei and Kai Zou. 2019. Eda: Easy data augmentation techniques for boosting performance on text classification tasks. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP…
  22. Shijie Wu and Mark Dredze. 2019. Beto, bentz, becas: The surprising cross-lingual effectiveness of bert. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 833–844.
  23. Ruike Zhang, Hanxuan Yang, and Wenji Mao. 2023. Cross-lingual cross-target stance detection with dual knowledge distillation framework. In Proceedings of the 2023 conference on empirical methods in natural language processing, pages 10804–10819. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wazir-2026-upr-semeval-2026-task-9/