--- title: "Abusive Language Detection on Arabic Social — Mubarak 2017" description: "The paper addresses the detection of obscene (vulgar and pornographic) language in Arabic social media." authors: ["Hamdy Mubarak", "Kareem Darwish", "Walid Magdy"] year: 2017 venue: "Proceedings of the First Workshop on Abusive Language Online" source: "https://aclanthology.org/W17-3008/" doi: "10.18653/v1/w17-3008" 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/mubarak-2017-abusive-language-detection-arabic-social/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/mubarak-2017-abusive-language-detection-arabic-social/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/2/) / Abusive Language Detection on Arabic Social Media # Abusive Language Detection on Arabic Social Media **Authors:** [Hamdy Mubarak](https://md.research.elamur.ai/free/en/authors/hamdy-mubarak/), [Kareem Darwish](https://md.research.elamur.ai/free/en/authors/kareem-darwish/), [Walid Magdy](https://md.research.elamur.ai/free/en/authors/walid-magdy/) · **Year:** 2017 · **Venue:** Proceedings of the First Workshop on Abusive Language Online · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/W17-3008/) · [PDF](https://aclanthology.org/W17-3008.pdf) · [DOI 10.18653/v1/w17-3008](https://doi.org/10.18653/v1/w17-3008) · [OpenAlex](https://openalex.org/W2747187574) **Citations:** 295 citations (OpenAlex, 2026-10-14) · [116](https://md.research.elamur.ai/free/en/papers/mubarak-2017-abusive-language-detection-arabic-social/cited-by/) in our library, of the 140884 papers whose bibliographies we read ## Summary The paper addresses the detection of obscene (vulgar and pornographic) language in Arabic social media. Hand-made blacklists rarely keep up with dialectal vocabulary, culture-specific slurs and creative misspellings. The authors start from 175 million Arabic tweets. They harvest a seed lexicon by looking at words that follow typical insult templates (for example, "you, son of…"), then check these by hand. They split prolific Twitter users into those who never use a seed term and those who use one at least once. Comparing word statistics between the two groups with a Log Odds Ratio surfaces candidate obscenities that occur only among the "obscene" users. They evaluate on a new crowd-annotated set of 1,100 Egyptian tweets. Simple lexicon matching gives very high precision (about 0.97) but recall below one half. Adding the mined unigrams raises recall slightly with no loss of precision. The authors release the lexicons, the tweet set, and 32K annotated comments that Aljazeera.net moderators had deleted. These are rare public resources for Arabic abusive-language research. ## Key points - **Seed lexicon:** 288 manually verified obscene words and phrases, plus 127 hashtags used by the TweetMogaz aggregator to block pornographic content. - **User groups for mining:** only users with at least 100 tweets were included. - Clean group: 166K users with 86M tweets. - Obscene group: 23K users with 16M tweets. - One LOR iteration over unigrams and bigrams (minimum count 10) produced 3,430 candidates with infinite LOR. In a sample of 100 candidates, 59 were judged truly obscene. - **Tweet test set:** 1,100 tweets, made up of 100 tweets from 10 controversial Egyptian accounts plus 1,000 replies to them. - Each tweet was labeled by 3 Egyptian CrowdFlower annotators who could see the reply thread. - Labels: 19.1% obscene, 40.3% offensive, 40.6% clean. - Inter-annotator agreement was 85%. - **Extrinsic results:** - SeedWords alone: P 0.97, R 0.43, F1 0.59. - SeedWords + LOR unigrams: P 0.97, R 0.44, F1 0.60. - Bigram lists lowered precision to 0.89. - **Aljazeera deleted comments:** 32K comments (3–200 characters long) were selected from 400K comments on about 10K articles. - Labels: 2% obscene, 79% offensive, 19% clean. - Agreement was 87%. - The comments are written in MSA and various dialects. ## Interesting ideas - **Template-based seed harvesting:** Collecting the words that fill typical insult slots (such as "son of X") quickly produces a candidate lexicon for manual review. This could bootstrap abuse lexicons for other low-resource languages or dialects, including Ukrainian and surzhyk. - **Contrasting users instead of documents:** Users who use a seed term at least once form a foreground corpus, and users who never do form the background. The contrast then reveals related vocabulary the seed list missed. Splitting by author rather than by message is a cheap form of weak supervision that could transfer to propaganda, spam or topic lexicons. - **Precision-first thresholding:** Keeping only terms that never appear in the background corpus (infinite LOR) yields a conservative list. The authors suggest relaxing the threshold later to gain recall. This gives a tunable knob between precision and recall for any lexicon-mining pipeline. - **Evaluating twice:** The mined list is checked both by sampling candidates for manual judgment (intrinsic) and by its effect on a labeled downstream test set (extrinsic). The two views together separate "are the words correct?" from "do they actually help detection?". - **Moderation logs as data:** Comments removed by professional moderators make a naturally enriched pool of inappropriate content for annotation. This approach could be reused with partnering news sites or forums in other languages. *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: [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/) Topics: [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 @inproceedings{mubarak2017abusive, title = {{Abusive Language Detection on Arabic Social Media}}, author = {Mubarak, Hamdy and Darwish, Kareem and Magdy, Walid}, booktitle = {Proceedings of the First Workshop on Abusive Language Online}, year = {2017}, doi = {10.18653/v1/w17-3008}, url = {https://doi.org/10.18653/v1/w17-3008} } ``` - [← previous Gender and Dialect Bias in YouTube's Automatic Captions](https://md.research.elamur.ai/free/en/papers/tatman-2017-gender-dialect-bias-youtube-s/) - [next → Unsupervised Neural Machine Translation](https://md.research.elamur.ai/free/en/papers/artetxe-2017-unsupervised-neural-machine-translation/) - [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/2/) ## References 14 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Ellen Spertus. 2007. Smokey: Automatic recognition of hostile messages. In In Proceedings of the Ninth Conference on Innovative Applications of Artificial Intelligence. [find](https://scholar.google.com/scholar?q=Ellen+Spertus%2E+2007%2E+Smokey%3A+Automatic+recognition+of+hostile+messages%2E+In+In+Proceedings+of+the+Ninth+Conference+on+Innovative+Applications+of+Artificial+Intelligence%2E) 2. Ehab A Abozinadah, Jr Jones, and H James. 2016. Improved microblog classification for detecting abusive arabic twitter accounts. In International Journal of Data Mining and Knowledge Management Process. IJDKP. [find](https://scholar.google.com/scholar?q=Ehab+A+Abozinadah%2C+Jr+Jones%2C+and+H+James%2E+2016%2E+Improved+microblog+classification+for+detecting+abusive+arabic+twitter+accounts%2E+In+International+Journal+of+Data+Mining+and+Knowledge+Management+Proces) 3. Zeerak Waseem. 2016. Are you a racist or am i seeing things? annotator influence on hate speech detection on twitter. In Proceedings of the First Workshop on NLP and Computational Social Science. Association for Computational Linguistics, Austin, Texas, pages 138–142. http://aclweb.org/anthology/W16-5618. [↗](https://aclanthology.org/W16-5618/) [our card](https://md.research.elamur.ai/free/en/papers/waseem-2016-are-you-racist-or-am/) 4. Ying Chen, Yilu Zhou, Sencun Zhu, and Xu Heng. 2012a. Detecting offensive language in social media to protect adolescent online safety. In In Privacy, Security, Risk and Trust (PASSAT), 2012 International Conference on and 2012 International Conference on Social Computing (SocialCom). IEEE, pages 71– 80. [find](https://scholar.google.com/scholar?q=Ying+Chen%2C+Yilu+Zhou%2C+Sencun+Zhu%2C+and+Xu+Heng%2E+2012a%2E+Detecting+offensive+language+in+social+media+to+protect+adolescent+online+safety%2E+In+In+Privacy%2C+Security%2C+Risk+and+Trust+%28PASSAT%29%2C+2012+Internati) 5. Hate Speech Watch. 2014. Hate crimes: Consequences of hate speech. In [http://www.nohatespeechmovement.org/hate-speech-watch/focus/](http://www.nohatespeechmovement.org/hate-speech-watch/focus/) consequences-of-hate-speech. [↗](http://www.nohatespeechmovement.org/hate-speech-watch/focus/) 6. Kareem Darwish, Walid Magdy, and Ahmed Mourad. 2012. Language processing for arabic microblog retrieval. In Proceedings of the 21st ACM international conference on Information and knowledge management. ACM, pages 2427–2430. [find](https://scholar.google.com/scholar?q=Kareem+Darwish%2C+Walid+Magdy%2C+and+Ahmed+Mourad%2E+2012%2E+Language+processing+for+arabic+microblog+retrieval%2E+In+Proceedings+of+the+21st+ACM+international+conference+on+Information+and+knowledge+management) 7. Guang Xiang, Bin Fan, Ling Wang, Jason Hong, and Carolyn Rose. 2012. Detecting offensive tweets via topical feature discovery over a large scale twitter corpus. In In Proceedings of the 21st ACM international conference on Information and knowledge management. ACM. [find](https://scholar.google.com/scholar?q=Guang+Xiang%2C+Bin+Fan%2C+Ling+Wang%2C+Jason+Hong%2C+and+Carolyn+Rose%2E+2012%2E+Detecting+offensive+tweets+via+topical+feature+discovery+over+a+large+scale+twitter+corpus%2E+In+In+Proceedings+of+the+21st+ACM+inter) 8. Eslam Elsawy, Moamen Mokhtar, and Walid Magdy. 2014. Tweetmogaz v2: Identifying news stories in social media. In Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management. ACM. [find](https://scholar.google.com/scholar?q=Eslam+Elsawy%2C+Moamen+Mokhtar%2C+and+Walid+Magdy%2E+2014%2E+Tweetmogaz+v2%3A+Identifying+news+stories+in+social+media%2E+In+Proceedings+of+the+23rd+ACM+International+Conference+on+Conference+on+Information+and+K) 9. George Forman. 2008. Bns feature scaling: an improved representation over tfidf for svm text classification. In In Proceedings of the 17th ACM conference on Information and knowledge management. ACM. [find](https://scholar.google.com/scholar?q=George+Forman%2E+2008%2E+Bns+feature+scaling%3A+an+improved+representation+over+tfidf+for+svm+text+classification%2E+In+In+Proceedings+of+the+17th+ACM+conference+on+Information+and+knowledge+management%2E+ACM%2E) 10. Timothy Jay and Kristin Janschewitz. 2008. The pragmatics of swearing. In Journal of Politeness Research. Language, Behaviour, Culture 4, no. 2. [find](https://scholar.google.com/scholar?q=Timothy+Jay+and+Kristin+Janschewitz%2E+2008%2E+The+pragmatics+of+swearing%2E+In+Journal+of+Politeness+Research%2E+Language%2C+Behaviour%2C+Culture+4%2C+no%2E+2%2E) 11. Walid Magdy. 2013. Tweetmogaz: a news portal of tweets. In In Proceedings of the 36th international ACM SIGIR conference on Research and development in information retrieval. ACM. [find](https://scholar.google.com/scholar?q=Walid+Magdy%2E+2013%2E+Tweetmogaz%3A+a+news+portal+of+tweets%2E+In+In+Proceedings+of+the+36th+international+ACM+SIGIR+conference+on+Research+and+development+in+information+retrieval%2E+ACM%2E) 12. Altaf Mahmud, Kazi Zubair Ahmed, and Mumit Khan. 2008. Detecting flames and insults in text. In In Proceedings of the Sixth International Conference on Natural Language Processing. [find](https://scholar.google.com/scholar?q=Altaf+Mahmud%2C+Kazi+Zubair+Ahmed%2C+and+Mumit+Khan%2E+2008%2E+Detecting+flames+and+insults+in+text%2E+In+In+Proceedings+of+the+Sixth+International+Conference+on+Natural+Language+Processing%2E) 13. Björn Ross, Michael Rist, Guillermo Carbonell, Benjamin Cabrera, Nils Kurowsky, and Michael Wojatzki. 2016. Measuring the Reliability of Hate Speech Annotations: The Case of the European Refugee Crisis. In Michael Beißwenger, Michael Wojatzki, and Torsten Zesch, editors, Proceedings of NLP4CMC III: 3rd Workshop on Natural Language Processing for Computer-Mediated Communication. Bochum, volume 17 of Bochumer Linguistische Arbeitsberichte, pages 6–9. [find](https://scholar.google.com/scholar?q=Bjo%CC%88rn+Ross%2C+Michael+Rist%2C+Guillermo+Carbonell%2C+Benjamin+Cabrera%2C+Nils+Kurowsky%2C+and+Michael+Wojatzki%2E+2016%2E+Measuring+the+Reliability+of+Hate+Speech+Annotations%3A+The+Case+of+the+European+Refugee+Cris) 14. Sara Sood, Judd Antin, and Elizabeth Churchill. 2012b. Using crowdsourcing to improve profanity detection. In In AAAI Spring Symposium: Wisdom of the Crowd, volume SS-12-06 of AAAI Technical Report. AAAI. [find](https://scholar.google.com/scholar?q=Sara+Sood%2C+Judd+Antin%2C+and+Elizabeth+Churchill%2E+2012b%2E+Using+crowdsourcing+to+improve+profanity+detection%2E+In+In+AAAI+Spring+Symposium%3A+Wisdom+of+the+Crowd%2C+volume+SS%2D12%2D06+of+AAAI+Technical+Report%2E+A)