---
title: "WOLI at SemEval-2020 Task 12 — paper by Otiefy et al. (2020)"
description: "The paper describes WideBot AI Lab's entry to SemEval-2020 Task 12 (OffensEval), Sub-task A for… 28 references, cited by 3 papers here."
authors: ["Yasser Otiefy", "Ahmed Abdelmalek", "Islam El Hosary"]
year: 2020
venue: "Proceedings of the Fourteenth Workshop on Semantic Evaluation"
source: "https://aclanthology.org/2020.semeval-1.298/"
doi: "10.18653/v1/2020.semeval-1.298"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/otiefy-2020-woli-semeval-2020-task-12/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/otiefy-2020-woli-semeval-2020-task-12/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/8/) / WOLI at SemEval-2020 Task 12: Arabic Offensive Language Identification on Different Twitter Datasets

# WOLI at SemEval-2020 Task 12: Arabic Offensive Language Identification on Different Twitter Datasets

**Authors:** Yasser Otiefy, Ahmed Abdelmalek and Islam El Hosary · **Year:** 2020 · **Venue:** [Proceedings of the Fourteenth Workshop on Semantic Evaluation](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2020](https://md.research.elamur.ai/free/en/venues/semeval/2020/)) · **Publisher:** [International Committee on Computational Linguistics (ICCL)](https://md.research.elamur.ai/free/en/publishers/international-committee-on-computational-linguistics-iccl/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2020.semeval-1.298/) · [PDF](https://aclanthology.org/2020.semeval-1.298.pdf) · [DOI 10.18653/v1/2020.semeval-1.298](https://doi.org/10.18653/v1/2020.semeval-1.298) · [OpenAlex](https://openalex.org/W3086873000)

**Citations:** 13 citations (OpenAlex, 2026-10-14) · [3](https://md.research.elamur.ai/free/en/papers/otiefy-2020-woli-semeval-2020-task-12/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes WideBot AI Lab's entry to SemEval-2020 Task 12 (OffensEval), Sub-task A for Arabic. The task is a binary decision on whether a tweet is offensive (OFF) or not (NOT). The official training data of about 10,000 tweets is skewed, with roughly four NOT tweets for every OFF tweet. To reduce the imbalance, the authors added two external sources: the Levantine L-HSAB corpus, with its labels collapsed to the binary scheme, and only the offensive half of a private WideBot chat dataset.

They compared three systems:
- a logistic-regression bag-of-words baseline;
- a linear SVM over combined word and character n-gram TF-IDF features;
- a neural model called CHBA, which combines CNN, highway, Bi-LSTM and attention layers.

The SVM was the official submission. It scored 86.9% Macro-F1 and placed 10th among 52 teams. After the deadline, a tuned CHBA network improved on it, reaching up to 88.72% on test. The work shows that cheap data augmentation and dense char+word lexical features are strong for dialectal, noisy Arabic text.

## Key points
- The official Arabic OffensEval training data has 1,915 OFF and 8,085 NOT tweets. With external data, the final split was 16,046 train, 1,000 validation and 2,000 test messages.
- External data came from two sources:
  - L-HSAB, with 8,846 Levantine tweets whose Hate and Abusive labels were merged into OFF;
  - WideBot's private balanced set of 1,600 + 1,600 chat messages, of which only the OFF part was used.
- The SVM used TF-IDF over character 1–6-grams and word 1–5-grams, with pruning of low-scoring features. It reached 87.32% on validation and 86.91% on test, compared with 76.59% on test for the baseline.
- CHBA works as follows:
  - Inputs are FastText Arabic skip-gram word vectors (300-d, up to 50 words) and trainable character embeddings (20-d, up to 10 characters per word).
  - The character embeddings pass through a CNN and a highway network.
  - Both branches are concatenated and fed to a Bi-LSTM with attention.
  - FastText worked better than AraVec in their tests.
  - The reported test scores are 88.13% with the original data only and 88.72% with external data. The prose and the table disagree on which setting produced the best run.
- The SVM's confusion matrix on test shows it handles NOT far better than OFF. It missed 103 of the 402 offensive tweets and gave 58 false alarms.

## Interesting ideas
- **Flooding.** Training uses a "flooding" objective |J − b| + b. Once the loss drops below a small level b, the gradient flips to ascent, which stops the network from driving training loss to zero and memorising the data.
- **Bias initialisation on logits.** The output-layer bias is set so that the initial predictions already match the class prior. The network then skips the early epochs it would otherwise spend learning only the imbalance.
- **Mapping external labels onto the task scheme.** Before merging the corpora, the external label sets are converted to the target scheme, for example Hate + Abusive → OFF. Only the slice that fixes the class imbalance is added.
- **Normalisation for noisy user text.** Runs of repeated characters are collapsed, and visually or orthographically similar letters are unified before features are computed. This makes informal, misspelled input map to stable forms.
- **Error analysis by class.** Reporting the confusion matrix per class exposed that the model mostly fails on subtle, context-dependent offensive tweets rather than failing evenly.

## For Mova

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

- **I1** (learn): A floor on the loss could keep the gparse or tagger training from overfitting small silver sets. Comparing several values of b on a held-out treebank would be cheap.
- **I2** (learn): Rare-label classifiers, such as infrequent dependency relations or rare frame roles, could start from their prior frequencies to converge faster and more stably.
- **I3** (tree): When mixing treebanks from different UD dialects, an explicit label-conversion step should be paired with selective sampling of under-represented relations, rather than merging everything wholesale.
- **I4** (tokens): A reversible normalisation pass of this kind (e.g. "sooooo" → "so"), recorded so the change stays explainable, could help the en tokenizer and lemmatizer on chat-like English.
- **I5** (frames): Evaluating irony or implicature detection the same way would show whether misses are concentrated in veiled cases that require pragmatic reasoning.

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

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,954 works · [🤝 Communication](https://md.research.elamur.ai/free/en/shelves/social-sciences/communication/) · 104 works

🧭 Topics (2): [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 1,035 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works

## Cite

Cite as: Yasser Otiefy, Ahmed Abdelmalek and Islam El Hosary. 2020. WOLI at SemEval-2020 Task 12: Arabic Offensive Language Identification on Different Twitter Datasets. Proceedings of the Fourteenth Workshop on Semantic Evaluation.

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

```bibtex
@inproceedings{otiefy2020woli,
  title = {{WOLI at SemEval-2020 Task 12: Arabic Offensive Language Identification on Different Twitter Datasets}},
  author = {Yasser Otiefy, Ahmed Abdelmalek and Islam {El Hosary}},
  booktitle = {Proceedings of the Fourteenth Workshop on Semantic Evaluation},
  year = {2020},
  doi = {10.18653/v1/2020.semeval-1.298},
  url = {https://doi.org/10.18653/v1/2020.semeval-1.298}
}
```

- [← previous Shallow Discourse Parsing for Under-Resourced Languages: Combining Machine Translation and Annotation Projection](https://md.research.elamur.ai/free/en/papers/gathje-2020-shallow-discourse-parsing-under-resourced/)
- [next → A Lexicon-Based Approach for Detecting Hedges in Informal Text](https://md.research.elamur.ai/free/en/papers/islam-2020-lexicon-based-approach-detecting-hedges/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/8/)

## References

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

1. Ehab A. Abozinadah and James H. Jones. 2016. Improved micro-blog classification for detecting abusive arabic twitter accounts. International Journal of Data Mining & Knowledge Management Process, 6:17–28. [find](https://scholar.google.com/scholar?q=Ehab+A%2E+Abozinadah+and+James+H%2E+Jones%2E+2016%2E+Improved+micro%2Dblog+classification+for+detecting+abusive+arabic+twitter+accounts%2E+International+Journal+of+Data+Mining+%26+Knowledge+Management+Process%2C+6%3A17)
2. Ehab A. Abozinadah and James H. Jones. 2017. A statistical learning approach to detect abusive twitter accounts. In ICCDA ’17. [find](https://scholar.google.com/scholar?q=Ehab+A%2E+Abozinadah+and+James+H%2E+Jones%2E+2017%2E+A+statistical+learning+approach+to+detect+abusive+twitter+accounts%2E+In+ICCDA+%E2%80%9917%2E)
3. Sweta Agrawal and Amit Awekar. 2018. Deep learning for detecting cyberbullying across multiple social media platforms. In ECIR. [find](https://scholar.google.com/scholar?q=Sweta+Agrawal+and+Amit+Awekar%2E+2018%2E+Deep+learning+for+detecting+cyberbullying+across+multiple+social+media+platforms%2E+In+ECIR%2E)
4. Areej Al-Hassan and Hmood Al-Dossari. 2019. Detection of hate speech in social networks: A survey on multilingual corpus. [find](https://scholar.google.com/scholar?q=Areej+Al%2DHassan+and+Hmood+Al%2DDossari%2E+2019%2E+Detection+of+hate+speech+in+social+networks%3A+A+survey+on+multilingual+corpus%2E)
5. Azalden Alakrot, Liam Murray, and Nikola S. Nikolov. 2018. Towards accurate detection of offensive language in online communication in arabic. In ACLING. [find](https://scholar.google.com/scholar?q=Azalden+Alakrot%2C+Liam+Murray%2C+and+Nikola+S%2E+Nikolov%2E+2018%2E+Towards+accurate+detection+of+offensive+language+in+online+communication+in+arabic%2E+In+ACLING%2E)
6. Nuha Albadi, Maram Kurdi, and Shivakant Mishra. 2018. Are they our brothers? analysis and detection of religious hate speech in the arabic twittersphere. 2018 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), pages 69–76. [find](https://scholar.google.com/scholar?q=Nuha+Albadi%2C+Maram+Kurdi%2C+and+Shivakant+Mishra%2E+2018%2E+Are+they+our+brothers%3F+analysis+and+detection+of+religious+hate+speech+in+the+arabic+twittersphere%2E+2018+IEEE%2FACM+International+Conference+on+Adva)
7. Pinkesh Badjatiya, Shashank Gupta, Manish Gupta, and Vasudeva Varma. 2017. Deep learning for hate speech detection in tweets. In WWW. [find](https://scholar.google.com/scholar?q=Pinkesh+Badjatiya%2C+Shashank+Gupta%2C+Manish+Gupta%2C+and+Vasudeva+Varma%2E+2017%2E+Deep+learning+for+hate+speech+detection+in+tweets%2E+In+WWW%2E)
8. Thomas Chadefaux. 2014. Early warning signals for war in the news. [find](https://scholar.google.com/scholar?q=Thomas+Chadefaux%2E+2014%2E+Early+warning+signals+for+war+in+the+news%2E)
9. Michael D. Conover, Jacob Ratkiewicz, Matthew R. Francisco, Bruno Gonçalves, Filippo Menczer, and Alessandro Flammini. 2011. Political polarization on twitter. In ICWSM. [find](https://scholar.google.com/scholar?q=Michael+D%2E+Conover%2C+Jacob+Ratkiewicz%2C+Matthew+R%2E+Francisco%2C+Bruno+Gon%C3%A7alves%2C+Filippo+Menczer%2C+and+Alessandro+Flammini%2E+2011%2E+Political+polarization+on+twitter%2E+In+ICWSM%2E)
10. Corinna Cortes and Vladimir Vapnik. 1995. [Support-vector networks](https://md.research.elamur.ai/free/en/papers/cortes-1995-support-vector-networks/). Machine Learning, 20:273–297.
11. Kareem Darwish, Dimitar Alexandrov, Preslav Nakov, and Yelena Mejova. 2017. Seminar users in the arabic twitter sphere. ArXiv, abs/1707.07276. [↗](https://arxiv.org/abs/1707.07276)
12. Takashi Ishida, Ikko Yamane, Tomoya Sakai, Gang Niu, and Masashi Sugiyama. 2020. Do we need zero training loss after achieving zero training error? ArXiv, abs/2002.08709. [↗](https://arxiv.org/abs/2002.08709)
13. Timothy Jay and Kristin Janschewitz. 2008. The pragmatics of swearing. [find](https://scholar.google.com/scholar?q=Timothy+Jay+and+Kristin+Janschewitz%2E+2008%2E+The+pragmatics+of+swearing%2E)
14. Yoon Kim, Yacine Jernite, David A Sontag, and Alexander M. Rush. 2016. [Character-aware neural language models](https://md.research.elamur.ai/free/en/papers/kim-2015-character-aware-neural-language-models/). In AAAI.
15. Irene Kwok and Yuzhou Wang. 2013. Locate the hate: Detecting tweets against blacks. In AAAI. [find](https://scholar.google.com/scholar?q=Irene+Kwok+and+Yuzhou+Wang%2E+2013%2E+Locate+the+hate%3A+Detecting+tweets+against+blacks%2E+In+AAAI%2E)
16. Bing Liu and Ian Lane. 2016. Attention-based recurrent neural network models for joint intent detection and slot filling. In INTERSPEECH. [find](https://scholar.google.com/scholar?q=Bing+Liu+and+Ian+Lane%2E+2016%2E+Attention%2Dbased+recurrent+neural+network+models+for+joint+intent+detection+and+slot+filling%2E+In+INTERSPEECH%2E)
17. Ping Liu, Wen Li, and Liang Zou. 2019. [Nuli at semeval-2019 task 6: Transfer learning for offensive language detection using bidirectional transformers](https://md.research.elamur.ai/free/en/papers/liu-2019-nuli-semeval-2019-task-6/). In SemEval@NAACL-HLT.
18. Shervin Malmasi and Marcos Zampieri. 2017. Detecting hate speech in social media. ArXiv, abs/1712.06427. [↗](https://arxiv.org/abs/1712.06427)
19. Hamdy Mubarak and Kareem Darwish. 2019. Arabic offensive language classification on twitter. In SocInfo. [find](https://scholar.google.com/scholar?q=Hamdy+Mubarak+and+Kareem+Darwish%2E+2019%2E+Arabic+offensive+language+classification+on+twitter%2E+In+SocInfo%2E)
20. Hamdy Mubarak, Ammar Rashed, Kareem Darwish, Younes Samih, and Ahmed Abdelali. 2020. [Arabic offensive language on twitter: Analysis and experiments](https://md.research.elamur.ai/free/en/papers/mubarak-2021-arabic-offensive-language-twitter-analysis/). arXiv preprint arXiv:2004.02192. [↗](https://arxiv.org/abs/2004.02192)
21. Hala Mulki, Hatem Haddad, Chedi Bechikh Ali, and Halima Alshabani. 2019. [L-hsab: A levantine twitter dataset for hate speech and abusive language](https://md.research.elamur.ai/free/en/papers/mulki-2019-l-hsab-levantine-twitter-dataset/).
22. Chikashi Nobata, Joel R. Tetreault, Achint Oommen Thomas, Yashar Mehdad, and Yi Chang. 2016. Abusive language detection in online user content. In WWW. [find](https://scholar.google.com/scholar?q=Chikashi+Nobata%2C+Joel+R%2E+Tetreault%2C+Achint+Oommen+Thomas%2C+Yashar+Mehdad%2C+and+Yi+Chang%2E+2016%2E+Abusive+language+detection+in+online+user+content%2E+In+WWW%2E)
23. Georgios K. Pitsilis, Heri Ramampiaro, and Helge Langseth. 2018. Detecting offensive language in tweets using deep learning. ArXiv, abs/1801.04433. [↗](https://arxiv.org/abs/1801.04433)
24. Amir Hossein Razavi, Diana Inkpen, Sasha Uritsky, and Stan Matwin. 2010. Offensive language detection using multi-level classification. In Canadian Conference on AI. [find](https://scholar.google.com/scholar?q=Amir+Hossein+Razavi%2C+Diana+Inkpen%2C+Sasha+Uritsky%2C+and+Stan+Matwin%2E+2010%2E+Offensive+language+detection+using+multi%2Dlevel+classification%2E+In+Canadian+Conference+on+AI%2E)
25. Abu Bakr Soliman, Kareem Eissa, and Samhaa R. El-Beltagy. 2017. Aravec: A set of arabic word embedding models for use in arabic nlp. In ACLING. Rupesh Kumar Srivastava, Klaus Greff, and Jürgen Schmidhuber. abs/1505.00387. 2015. [Highway networks](https://md.research.elamur.ai/free/en/papers/srivastava-2015-highway-networks/). [↗](https://arxiv.org/abs/1505.00387)
26. ArXiv, Zeerak Waseem and Dirk Hovy. 2016. [Hateful symbols or hateful people? predictive features for hate speech detection on twitter](https://md.research.elamur.ai/free/en/papers/waseem-2016-hateful-symbols-or-hateful-people/). In SRW@HLT-NAACL.
27. Dawei Yin, Zhenzhen Xue, Liangjie Hong, Brian D. Davison, and Lynne Edwards. 2009. Detection of harassment on web 2.0. [find](https://scholar.google.com/scholar?q=Dawei+Yin%2C+Zhenzhen+Xue%2C+Liangjie+Hong%2C+Brian+D%2E+Davison%2C+and+Lynne+Edwards%2E+2009%2E+Detection+of+harassment+on+web+2%2E0%2E)
28. Marcos Zampieri, Preslav Nakov, Sara Rosenthal, Pepa Atanasova, Georgi Karadzhov, Hamdy Mubarak, Leon Derczynski, Zeses Pitenis, and Çağrı Çöltekin. 2020. SemEval-2020 Task 12: Multilingual Offensive Language Identification in Social Media (OffensEval 2020). In Proceedings of SemEval. [find](https://scholar.google.com/scholar?q=Marcos+Zampieri%2C+Preslav+Nakov%2C+Sara+Rosenthal%2C+Pepa+Atanasova%2C+Georgi+Karadzhov%2C+Hamdy+Mubarak%2C+Leon+Derczynski%2C+Zeses+Pitenis%2C+and+%C3%87ag%CC%86r%C4%B1+%C3%87%C3%B6ltekin%2E+2020%2E+SemEval%2D2020+Task+12%3A+Multilingual+Offensive)

*Updated 2 November 2026*