Enhanced Arabic-language cyberbullying detection: deep embedding and transformer (BERT) approaches

Summary

Most automatic cyberbullying detectors target English, and Arabic tools remain scarce. This paper tries to close that gap for Arabic posts on X. The authors collected 10,662 X posts, cleaned them, and labeled them as bullying or not, using Cohen's kappa to check annotation consistency. They then ran four experiments:

classical machine-learning baselines;
LSTM and Bi-LSTM networks paired with different word representations (TF-IDF, GloVe, Araword2Vec, FastText);
fine-tuned Arabic BERT models (Arabbertv02 and CAMeL-da);
hybrids that feed BERT representations into LSTM or Bi-LSTM layers.

Bi-LSTM with FastText embeddings scored best at 98% accuracy. LSTM-BERT and Bi-LSTM-BERT reached about 97%, and every deep configuration beat the classical baselines. The work matters for two reasons: it adds an annotated Arabic dataset, and it shows that subword-aware embeddings and Arabic-pretrained transformers handle noisy, dialect-mixed social media text well. One caveat is that the dataset is only available on request.

Key points

The dataset has 10,662 Arabic X posts with binary bullying labels, checked with the kappa statistic. It covers general Arabic rather than one dialect and is not public for privacy reasons.
Among the baselines, linear SVC was strongest and RF weakest at 76%. LSTM and Bi-LSTM on TF-IDF features were also unimpressive.
Embedding choice changed results sharply:
- Bi-LSTM scored 98% with FastText, 97% with Araword2Vec, and 81% with GloVe.
- LSTM scored 86% with FastText, 83% with Araword2Vec, and 82% with GloVe.
Of the transformers, the dialect-pretrained CAMeL-da beat Arabbertv02 and reached 97%.
The hybrid LSTM/Bi-LSTM + CAMeL-da models gave 97% accuracy and an F1 of 96.8%. Adding the transformer lifted plain LSTM up to Bi-LSTM's level and sped up training, with early stopping at epoch 12.

Interesting ideas

I1. FastText builds word vectors from character n-grams, so misspelled, inflected, or unseen word forms in noisy posts still get meaningful vectors; here it clearly beat whole-word embeddings. — For Mova (tokens): Subword vectors could serve as a fallback for the tagger and lemmatizer on out-of-vocabulary or misspelled English forms.
I2. The authors cross every architecture (LSTM, Bi-LSTM) with every representation (TF-IDF, GloVe, Araword2Vec, FastText, BERT) in one grid, which separates the effect of the encoder from the effect of the input features. — For Mova (learn): Evaluating parser or tagger variants as a full features-by-model grid would show which component actually drives a gain instead of crediting the whole pipeline.
I3. Inter-annotator agreement (kappa) is used as a quality gate before any training, rather than trusting single-annotator labels. — For Mova (learn): A kappa-style agreement score between witness parsers could decide which silver trees are trusted enough to enter training.
I4. Pretraining on informal, dialect-rich text (CAMeL-da) beat a more general model, showing that matching the pretraining register to the target text pays off. — For Mova (tree): The gparse transformer encoder could be pretrained or adapted on informal English (social media, dialogue) before parsing such text.
I5. Stacking a recurrent layer on top of transformer representations let a weaker LSTM catch up with Bi-LSTM and converge in fewer epochs. — For Mova (frames): A light sequential head over contextual embeddings could cheaply flag hostile or ironic speech acts, then pass the flag into the explainable frame layer.

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.11591/ijai.v14.i3.pp2258-2269 · arXiv 2510.02232

@misc{aljohani2025enhanced,
  title = {{Enhanced Arabic-language cyberbullying detection: deep embedding and transformer (BERT) approaches}},
  author = {Aljohani, Ebtesam Jaber and Yafoo, Wael M. S.},
  year = {2025},
  doi = {10.11591/ijai.v14.i3.pp2258-2269},
  eprint = {2510.02232},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.11591/ijai.v14.i3.pp2258-2269}
}

References

2 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

E. J. ALjohani, W. M. S. Yafooz, and A. Alsaeedi, “Cyberbullying detection approaches: a review,” in 2023 5th International Conference on Inventive Research in Computing Applications (ICIRCA), 2023, pp. 1310–1316, doi: 10.1109/ICIRCA57980.2023.10220688. T. K. H. Chan, C. M. K. Cheung, and Z. W. Y. Lee, “Cyberbullying on social networking sites: a literature review and future research directions,” Information and Management, vol. 58, no. 2, 2021, doi: 10.1016/j.im.2020.103411. C. Evangelio, P. Rodríguez-González, J. Fernández-Río, and S. Gonzalez-Villora, “Cyberbullying in elementary and middle school students: a systematic review,” Computers & Education, vol. 176, p. 104356, 2022, doi: 10.1016/j.compedu.2021.104356. S. I. Alqahtani, W. M. S. Yafooz, A. Alsaeedi, L. Syed, and R. Alluhaibi, “Children’s safety on Youtube: a systematic review,” Applied Sciences, vol. 13, no. 6, 2023, doi: 10.3390/app13064044. M. Mladenovic, V. Ošmjanski, and S. V. Stankovic, “Cyber-aggression, cyberbullying, and cyber-grooming,” ACM Computing Surveys, vol. 54, no. 1, 2021, doi: 10.1145/3424246. M. Alotaibi, B. Alotaibi, and A. Razaque, “A multichannel deep learning framework for cyberbullying detection on social media,” Electronics, vol. 10, no. 21, 2021, doi: 10.3390/electronics10212664. N. Rezvani and A. Beheshti, “Towards attention-based context-boosted cyberbullying detection in social media,” Journal of Data Intelligence, vol. 2, no. 4, pp. 418–433, 2021. M. M. Islam, M. A. Uddin, L. Islam, A. Akter, S. Sharmin, and U. K. Acharjee, “Cyberbullying detection on social networks using machine learning approaches,” in 2020 IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE), 2020, pp. 1–6, doi: 10.1109/CSDE50874.2020.9411601. M. A. Al-Ajlan and M. Ykhlef, “Deep learning algorithm for cyberbullying detection,” International Journal of Advanced Computer Science and Applications, vol. 9, no. 9, pp. 199–205, 2018, doi: 10.14569/ijacsa.2018.090927. W. M. S. Yafooz, A. Al-Dhaqm, and A. Alsaeedi, “Detecting kids cyberbullying using transfer learning approach: transformer fine-tuning models,” in Kids Cybersecurity Using Computational Intelligence Techniques, Cham: Springer, 2023, pp. 255–267. S. Afrifa and V. Varadarajan, “Cyberbullying detection on Twitter using natural language processing and machine learning techniques,” International Journal of Innovative Technology and Interdisciplinary Sciences, vol. 5, no. 4, pp. 1069–1080, 2022, doi: 10.15157/IJITIS.2022.5.4.1069-1080. A. Ali and A. M. Syed, “Cyberbullying detection using machine learning,” Pakistan Journal of Engineering and Technology, vol. 3, no. 2, pp. 45–50, 2022, doi: 10.51846/vol3iss2pp45-50. C. Raj, A. Agarwal, G. Bharathy, B. Narayan, and M. Prasad, “Cyberbullying detection: hybrid models based on machine learning and natural language processing techniques,” Electronics, vol. 10, no. 22, 2021, doi: 10.3390/electronics10222810. T. Kanan, A. Aldaaja, and B. Hawashin, “Cyber-bullying and cyber-harassment detection using supervised machine learning techniques in Arabic social media contents,” Journal of Internet Technology, vol. 21, no. 5, pp. 1409–1421, 2020, doi: 10.3966/160792642020092105016. A. M. Alduailaj and A. Belghith, “Detecting Arabic cyberbullying tweets using machine learning,” Machine Learning and Knowledge Extraction, vol. 5, no. 1, pp. 29–42, 2023, doi: 10.3390/make5010003. J. Hani, M. Nashaat, M. Ahmed, Z. Emad, E. Amer, and A. Mohammed, “Social media cyberbullying detection using machine learning,” International Journal of Advanced Computer Science and Applications, vol. 10, no. 5, pp. 703–707, 2019, doi: 10.14569/ijacsa.2019.0100587. M. E. AlFarah, I. Kamel, Z. Al Aghbari, and D. Mouheb, “Arabic cyberbullying detection from imbalanced dataset using machine learning,” Communications in Computer and Information Science, pp. 397–409, 2022, doi: 10.1007/978-3-031-05767-0_31. M. A. Al-Ajlan and M. Ykhlef, “Optimized Twitter cyberbullying detection based on deep learning,” in 2018 21st Saudi Computer Society National Computer Conference (NCC), 2018, pp. 1–5, doi: 10.1109/NCG.2018.8593146. M. S. Ahmed, S. M. Maher, and M. E. Khudhur, “Arabic cyberbullying detecting using ensemble deep learning technique,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 32, no. 2, pp. 1031–1041, 2023, doi: 10.11591/ijeecs.v32.i2.pp1031-1041. S. Neelakandan et al., “Deep learning approaches for cyberbullying detection and classification on social media,” Computational Intelligence and Neuroscience, vol. 2022, pp. 1–13, 2022, doi: 10.1155/2022/2163458. ↗
C. Iwendi, G. Srivastava, S. Khan, and P. K. R. Maddikunta, “Cyberbullying detection solutions based on deep learning architectures,” Multimedia Systems, vol. 29, no. 3, pp. 1839–1852, 2023, doi: 10.1007/s00530-020-00701-5. B. Haidar, M. Chamoun, and A. Serhrouchni, “Multilingual cyberbullying detection system: detecting cyberbullying in Arabic content,” in 2017 1st Cyber Security in Networking Conference (CSNet), 2017, pp. 1–8, doi: 10.1109/CSNET.2017.8242005. R. Duwairi, A. Hayajneh, and M. Quwaider, “A deep learning framework for automatic detection of hate speech embedded in Arabic tweets,” Arabian Journal for Science and Engineering, vol. 46, no. 4, pp. 4001–4014, 2021, doi: 10.1007/s13369-021-05383-3. A. Muneer and S. M. Fati, “A comparative analysis of machine learning techniques for cyberbullying detection on Twitter,” Future Internet, vol. 12, no. 11, pp. 1–21, 2020, doi: 10.3390/fi12110187. D. Mouheb, R. Albarghash, M. F. Mowakeh, Z. Al Aghbari, and I. Kamel, “Detection of Arabic cyberbullying on social networks using machine learning,” in 2019 IEEE/ACS 16th International Conference on Computer Systems and Applications (AICCSA), 2019, pp. 1–5, doi: 10.1109/AICCSA47632.2019.9035276. C. Graëff, T. Lampert, J. P. Mazellier, N. Padoy, L. El Amiri, and P. Liverneaux, “The preliminary stage in developing an artificial intelligence algorithm: a study of the inter-and intra-individual variability of phase annotations in internal fixation of distal radius fracture videos,” Artificial Intelligence Surgery, vol. 3, no. 3, pp. 147–159, 2023, doi: 10.20517/ais.2023.12. K. T. Mursi, A. Y. Almalki, M. M. Alshangiti, F. S. Alsubaei, and A. A. Alghamdi, “ArCyb: a robust machine-learning model for Arabic cyberbullying tweets in Saudi Arabia,” International Journal of Advanced Computer Science and Applications, vol. 14, no. 9, pp. 1059–1067, 2023, doi: 10.14569/IJACSA.2023.01409110. M. O. Hegazi, Y. Al-Dossari, A. Al-Yahy, A. Al-Sumari, and A. Hilal, “Preprocessing Arabic text on social media,” Heliyon, vol. 7, no. 2, 2021, doi: 10.1016/j.heliyon.2021.e06191. A. Kulkarni and S. Mhaske, “Tweet sentiment analysis and study and comparison of various approaches and classification algorithms used,” International Research Journal of Engineering and Technology, vol. 7, no. 4, pp. 2619–2624, 2020. R. Corizzo, E. Zdravevski, M. Russell, A. Vagliano, and N. Japkowicz, “Feature extraction based on word embedding models for intrusion detection in network traffic,” Journal of Surveillance, Security and Safety, vol. 1, pp. 140–150, 2020, doi: 10.20517/jsss.2020.15. T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Distributed representations of words and phrases and their compositionality,” Neural information processing systems, vol. 1, pp. 1–9, 2006. N. Al-Twairesh, H. Al-Khalifa, A. Alsalman, and Y. Al-Ohali, “Sentiment analysis of Arabic tweets: feature engineering and a hybrid approach,” arXiv-Computer Science, pp. 1–11, 2018. B. Samal, A. K. Behera, and M. Panda, “Performance analysis of supervised machine learning techniques for sentiment analysis,” in 2017 Third International Conference on Sensing, Signal Processing and Security (ICSSS), 2017, pp. 128–133, doi: 10.1109/SSPS.2017.8071579. F. Qiao, B. Li, Y. Zhang, H. Guo, W. Li, and S. Zhou, “A fast and accurate recognition of ECG signals based on ELM-LRF and BLSTM algorithm,” IEEE Access, vol. 8, pp. 71189–71198, 2020, doi: 10.1109/ACCESS.2020.2987930. A. Rahali and M. A. Akhloufi, “End-to-end transformer-based models in textual-based NLP,” AI, vol. 4, no. 1, pp. 54–110, 2023, doi: 10.3390/ai4010004. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/aljohani-2025-enhanced-arabic-language-cyberbullying-detection/

📄 Enhanced Arabic-language cyberbullying detection: deep embedding and transformer (BERT) approaches