The exponential growth of online communication platforms has led to a significant rise in toxic and offensive language, posing serious challenges for content moderation and user safety. Detecting such harmful text requires intelligent models capable of understanding linguistic subtleties, cultural context, and user intent. This paper presents a comprehensive review of research developments in toxic and offensive language detection, spanning traditional machine learning approaches, deep learning architectures, transformer models, and recent advancements in large language models (LLMs). The survey emphasizes the evolution of methods from lexical feature engineering to contextual representation learning and multilingual modeling. Key issues such as dataset imbalance, bias mitigation, interpretability, and efficiency are critically analyzed. Furthermore, the review highlights the importance of fairnessaware learning, explainable AI, and parameter-efficient finetuning for scalable real-world applications. The findings underline the ongoing transition toward hybrid, interpretable, and ethically aligned systems that integrate automation with human oversight to ensure safe and responsible online communication.
Keywords
^KEYWORDS^
Conclusion
^CONCLUSION_TEXT^
References
[1]S. Gaydhani et al., “Detecting Hate Speech and Offensive Language on Twitter Using Machine Learning,” IEEE SIIE, 2020. [2]S. Sadiq et al., “Automated Classification of Harmful Online Content Using Machine Learning,” Proceedings of ICDIS, 2019. [3]D. Bonetti et al., “Rule-Based and Statistical Models for Hate Speech Detection,” Information Journal, 2023. [4]D. D’Sa et al., “Deep Learning Architectures for Hate and Offensive Speech Classification,” EMNLP Findings, 2021. [5]J. Risch et al., “Offensive Language Identification Using Deep Neural Networks,” ACL Workshop on TRAC, 2020. [6]R. Sharma et al., “Multilingual Transfer Learning for Offensive Language Detection,” Applied Sciences, 2023. [7]A. Alansari and M. Luqman, “Active Learning for Arabic Offensive Speech Classification,” Applied Sciences, 2025. [8]A. Bhat et al., “ToxiScope: Workplace Toxicity Dataset and BERT-Based Classification,” Information, 2024. [9]N. Garg et al., “Bias and Fairness in Hate Speech Detection,” Mathematics, 2023. [10]J. Risch et al., “Explainability in Toxic Language Detection,” Findings of EMNLP, 2021. [11]A. Banerjee et al., “Adversarial Robustness and Interpretability in Neural Models,” IEEE Access, 2022. [12]R. Sharma et al., “Attention Visualization for Explainable Transformers,” Applied Sciences, 2024. [13]T. Hussain et al., “LLaMA-3-8B with QLoRA for Roman Urdu Toxicity Detection,” arXiv preprint, 2025. [14]M. Jahan et al., “Advancements in Offensive Language Detection Using GPT and Mistral,” ACL Long Papers, 2025. [15]M. Luqman et al., “Comprehensive Review and Experimental Analysis of Offensive Language Detection,” Applied Sciences, 2025
How to Cite This Paper
Shivam , Shresth Sharma , Shrey Swami , Shubham Yadav, Ankush Gupta (2026). Toxicity and Offensive Word Detection. International Journal of Computer Techniques, 13(2). ISSN: 2394-2231.
