Elamur Research⌕
☰
home / papers / 2024 / Using Sarcasm to Improve Cyberbullying Detection

Using Sarcasm to Improve Cyberbullying Detection

Paper · 2024 · Xiaoyu Guo and Susan Gauch

Summary

Automatic cyberbullying detectors often fail on messages whose literal wording does not match their intent. Guo and Gauch first benchmark three transformer classifiers (vanilla BERT, HateBERT and CyberBERT) on three binary corpora (cyberbully vs. non-cyberbully) taken from Twitter, YouTube and Kaggle. They fine-tune each model on a 70/30 split that keeps the original class ratio. They score the models with both accuracy and F1, because every corpus is heavily skewed toward the benign class. A close look at the misclassified posts shows that many are sarcastic: some sound friendly but hide hostility, and others sound rude but are meant kindly. To address this, the authors add a neural sarcasm detector. Its output and the cyberbullying model's judgement go into a small multilayer perceptron that makes the final decision. This combination outperforms the BERT-based baselines on both accuracy and F1. The work matters because it turns error analysis into a concrete design fix and identifies figurative language as a clear source of moderation errors.

Key points

  • Baselines compared: vanilla BERT, HateBERT (retrained on RAL-E, a corpus of banned Reddit comments, with the MLM objective) and CyberBERT (BERT plus a fully connected layer and a softmax classifier).
  • The datasets and their cyberbully / non-cyberbully counts are:
    • Twitter (Wang et al., 2020): 7945 / 38,072
    • YouTube (Dadvar et al., 2014): 417 / 3047
    • Kaggle: 2806 / 5993
  • Preprocessing steps:
    • Posts of four words or fewer are discarded.
    • Hashtag symbols are stripped.
    • Emojis are converted to text with the Python emoji package.
    • User mentions become "@USR" and links become "URL".
  • The split is 70% training and 30% testing, sampled per class to keep the imbalance. Evaluation uses accuracy plus F1, since accuracy alone hides poor results on the minority class.
  • The proposed system feeds a sarcasm classifier's signal into an MLP for the final cyberbullying label. It improves both accuracy and F1 over the benchmarks.

Interesting ideas

  • I1. The fix was chosen by inspecting misclassified examples and naming their shared trait (sarcasm), not by tuning blindly; an error taxonomy can point to which missing signal to add.
  • I2. An auxiliary pragmatic detector is combined with the main classifier through a small late-fusion layer, so neither model needs retraining and the extra signal's contribution stays inspectable.
  • I3. Sarcasm is treated as a mismatch that runs both ways: hostile intent under benign words, and benign intent under harsh words.
  • I4. Accuracy is paired with F1, and splits are stratified per class, because skewed data lets a majority-class guesser look strong.
  • I5. Emojis are rewritten as words, and mentions and links are replaced with fixed placeholders, so noisy social-media tokens become normal lexical material.

For Mova

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

  • I1 (learn): After each evaluation run, cluster the failed items by a linguistic trait and use the dominant cluster to decide which repair rule or teacher tier to invest in next.
  • I2 (frames): An explicit irony flag in prag could sit beside the speech-act frame and, when raised, invert the evaluative polarity, with the flag recorded in the explanation trace.
  • I3 (answer): Readers answering questions about a character's attitude in fables could check whether the stated sentiment conflicts with what the character actually does, before trusting the literal words.
  • I4 (learn): Evaluations of rare phenomena (e.g., ironic or negated frames) should report per-class F1 on class-preserving splits, with a majority-class baseline as a negative control.
  • I5 (tokens): The en tokenizer could map emojis to textual names and normalise @-mentions and URLs to placeholder tokens with a fixed POS, so the tagger and parser handle them predictably.

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, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksInformation Systems · 1,730 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 1,035 works

Cite

Cite as Xiaoyu Guo and Susan Gauch. 2024. Using Sarcasm to Improve Cyberbullying Detection. Proceedings of the Fourth Workshop on Threat, Aggression & Cyberbullying @ LREC-COLING-2024.

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

@inproceedings{guo2024using,
  title = {{Using Sarcasm to Improve Cyberbullying Detection}},
  author = {Xiaoyu Guo and Susan Gauch},
  booktitle = {Proceedings of the Fourth Workshop on Threat, Aggression \& Cyberbullying @ LREC-COLING-2024},
  year = {2024},
  doi = {10.63317/29c6pg2kc9ks},
  url = {https://doi.org/10.63317/29c6pg2kc9ks}
}

References

39 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mohammed Ali Al-Garadi, Kasturi Dewi Varathan, and Sri Devi Ravana. 2016. Cybercrime detection in online communications: The experimental case of cyberbullying detection in the twitter network. Computers in Human Behavior, 63:433– 443. ↗
  2. Darkunde Mayur Ashok, Agrawal Nidhi Ghanshyam, Sayed Saniya Salim, Dun-garpur Burhanuddin Mazahir, and Bhushan S Thakare. 2020. Sarcasm detection using genetic optimization on lstm with cnn. In 2020 International Conference for Emerging Technology (INCET), pages 1–4. IEEE. find
  3. Vimala Balakrishnan, Shahzaib Khan, and Hamid R Arabnia. 2020. Improving cyberbullying detection using twitter users’ psychological features and machine learning. Computers & Security, 90:101710. ↗
  4. Jennifer Bayzick, April Kontostathis, and Lynne Edwards. 2011. Detecting the presence of cyberbullying using computer software. find
  5. Uwe Bretschneider, Thomas Wöhner, and Ralf Peters. 2014. Detecting online harassment in social networks. find
  6. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  7. Tommaso Caselli, Valerio Basile, Jelena Mitrović, and Michael Granitzer. 2020. Hatebert: Retraining bert for abusive language detection in english. arXiv preprint arXiv:2010.12472. ↗
  8. Saravanan Chandrasekaran, Aditya Kumar Singh Pundir, T Bheema Lingaiah, et al. 2022. Deep learning approaches for cyberbullying detection and classification on social media. find
  9. Computational Intelligence and Neuroscience, 2022. find
  10. Niladri Chatterjee, Tanya Aggarwal, and Rishabh Maheshwari. 2020. Sarcasm detection using deep learning-based techniques. Deep Learning-Based Approaches for Sentiment Analysis, pages 237–258. ↗
  11. Vikas S Chavan and SS Shylaja. 2015. Machine learning approach for detection of cyber-aggressive comments by peers on social media network. In 2015 International Conference on Advances in Computing, Communications and Informatics (ICACCI), pages 2354–2358. IEEE. find
  12. Maral Dadvar, FMG de Jong, Roeland Ordelman, and Dolf Trieschnigg. 2012a. Improved cyberbullying detection using gender information. In Proceedings of the Twelfth Dutch-Belgian Information Retrieval Workshop (DIR 2012). University of Ghent. find
  13. Maral Dadvar, Roeland Ordelman, Franciska De Jong, and Dolf Trieschnigg. 2012b. Towards user modelling in the combat against cyberbullying. In Natural Language Processing and Information Systems: 17th International Conference on Applications of Natural Language to Information Systems, NLDB 2012, Groningen, The Netherlands, June 26-28, 2012. Proceedings 17, pages 277–283. Springer. find
  14. Maral Dadvar, Dolf Trieschnigg, and Franciska De Jong. 2014. Experts and machines against bullies: A hybrid approach to detect cyberbullies. In Advances in Artificial Intelligence: 27th Canadian Conference on Artificial Intelligence, Canadian AI 2014, Montréal, QC, Canada, May 6-9, 2014. Proceedings 27, pages 275–281. Springer. find
  15. Maral Dadvar, Dolf Trieschnigg, Roeland Ordelman, and Franciska De Jong. 2013. Improving cyberbullying detection with user context. In Advances in Information Retrieval: 35th European Conference on IR Research, ECIR 2013, Moscow, Russia, March 24-27, 2013. Proceedings 35, pages 693–696. Springer. find
  16. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805. ↗
  17. Karthik Dinakar, Roi Reichart, and Henry Lieberman. 2011. Modeling the detection of textual cyberbullying. In Proceedings of the International AAAI Conference on Web and Social Media, volume 5, pages 11–17. find
  18. Suliman Mohamed Fati, Amgad Muneer, Ayed Alwadain, and Abdullateef O Balogun. 2023. Cyberbullying detection on twitter using deep learning-based attention mechanisms and continuous bag of words feature extraction. Mathematics, 11(16):3567. ↗
  19. Ziyang Feng, Jintao Su, and Junkuo Cao. 2022. Bhf: Bert-based hierarchical attention fusion network for cyberbullying remarks detection. In Proceedings of the 2022 5th International Conference on Machine Learning and Natural Language Processing, pages 1–7. find
  20. Aniruddha Ghosh and Tony Veale. 2016. Fracking sarcasm using neural network. In Proceedings of the 7th workshop on computational approaches to subjectivity, sentiment and social media analysis, pages 161–169. ↗
  21. Ashish Goel and Latika Gupta. 2020. Social media in the times of covid-19. JCR: Journal of Clinical Rheumatology, 26(6):220–223. ↗
  22. Xiaoyu Guo, Usman Anjum, and Jusin Zhan. 2022. Cyberbully detection using bert with augmented texts. In 2022 IEEE International Conference on Big Data (Big Data), pages 1246–1253. IEEE. find
  23. Jeffrey T Hancock. 2004. Verbal irony use in face-to-face and computer-mediated conversations. Journal of Language and Social Psychology, 23(4):447–463. ↗
  24. Sameer Hinduja and Justin W Patchin. 2008. Cyberbullying: An exploratory analysis of factors related to offending and victimization. Deviant behavior, 29(2):129–156. ↗
  25. Akshi Kumar, Saurabh Raj Sangwan, Anshika Arora, Anand Nayyar, Mohamed Abdel-Basset, et al. 2019. Sarcasm detection using soft attention-based bidirectional long short-term memory model with convolution network. IEEE access, 7:23319–23328. find
  26. Amanda Lenhart, Kristen Purcell, Aaron Smith, and Kathryn Zickuhr. 2010. Social media & mobile internet use among teens and young adults. millennials. Pew internet & American life project. find
  27. Vicente J Llorent, Rosario Ortega-Ruiz, and Izabela Zych. 2016. Bullying and cyberbullying in minorities: Are they more vulnerable than the majority group? Frontiers in psychology, 7:1507. ↗
  28. Altaf Mahmud, Kazi Zubair Ahmed, and Mumit Khan. 2008. Detecting flames and insults in text. find
  29. Vinita Nahar, Xue Li, and Chaoyi Pang. 2013. An effective approach for cyberbullying detection. Communications in information science and management engineering, 3(5):238. find
  30. Sayanta Paul and Sriparna Saha. 2022. Cyberbert: Bert for cyberbullying identification: Bert for cyberbullying identification. Multimedia Systems, 28(6):1897–1904. find
  31. Kelly Reynolds, April Kontostathis, and Lynne Edwards. 2011. Using machine learning to detect cyberbullying. In 2011 10th International Conference on Machine learning and applications and workshops, volume 2, pages 241–244. IEEE. find
  32. Pradeep Kumar Roy and Fenish Umeshbhai Mali. 2022. Cyberbullying detection using deep transfer learning. Complex & Intelligent Systems, 8(6):5449–5467. ↗
  33. Stephen M Serra and Hein S Venter. 2011. Mobile cyber-bullying: A proposal for a pre-emptive approach to risk mitigation by employing digital forensic readiness. In 2011 Information Security for South Africa, pages 1–5. IEEE. find
  34. Boaz Shmueli, Lun-Wei Ku, and Soumya Ray. 2020. Reactive Supervision: A New Method for Collecting Sarcasm Data. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2553–2559, Online. Association for Computational Linguistics. ↗
  35. Peter K Smith, Jess Mahdavi, Manuel Carvalho, Sonja Fisher, Shanette Russell, and Neil Tippett. 2008. Cyberbullying: Its nature and impact in secondary school pupils. Journal of child psychology and psychiatry, 49(4):376–385. ↗
  36. Sterling Stauffer, Melissa Allen Heath, Sarah Marie Coyne, and Scott Ferrin. 2012. High school teachers’ perceptions of cyberbullying prevention and intervention strategies. Psychology in the Schools, 49(4):352–367. ↗
  37. Jason Wang, Kaiqun Fu, and Chang-Tien Lu. 2020. Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection. In 2020 IEEE International Conference on Big Data (Big Data), pages 1699–1708. IEEE. find
  38. Zhibiao Wu and Martha Palmer. 1994. Verb semantics and lexical selection. arXiv preprint cmp-lg/9406033.
  39. Rui Zhao, Anna Zhou, and Kezhi Mao. 2016. Automatic detection of cyberbullying on social networks based on bullying features. In Proceedings of the 17th international conference on distributed computing and networking, pages 1–6. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/guo-2024-sarcasm-improve-cyberbullying-detection/