Abstract

Social media provides a channel for communicating sentiments and perspectives; nevertheless, it is also utilized by certain individuals to disseminate hate, directing it towards individuals, organizations, towns, or nations. Consequently, it is imperative to recognize such content and implement corrective measures. In recent years, many methods have been evolved to automatically detect abusive words, offensive remarks, and toxic talks across online platforms. Despite that most prior research has primarily focused on English-language texts. The leading cause are the lack of similar work in scarcity of additional dialects is the root of the issue of sufficient resources. Although Hindi represents a single of the most widely spoken languages globally, there are very few data repository available for detecting hate speech in it, and none are specifically designed to address hate speech directed at women. This study seeks to fill that space by introducing a structured and explicate dataset, termed HANS (Hate speech Against Women in Hindi social media) which is meant to find hate words against women in Hindi. To evaluate the effectiveness of this dataset, a range of Neural Networks, Traditional, Attention, hybrid approaches and models that use transformers is employed. The findings demonstrate that HANS is highly effective in identifying hate speech directed at women in Hindi, thereby supporting the objective of developing a dedicated resource for this purpose.

References

• Abro, S., Shaikh, S., Khand, Z.H., Zafar, A., Khan, S., & Mujtaba, G. (2020). A comparative study of automatic hate speech detection using machine learning. Int J Adv Computer Sci Appl. https://doi.org/10.14569/ijacsa.2020.0110
• . Ali, A., & Syed, A.M. (2020). Detection of cyberbullying with machine learning techniques. Pak J Eng Technol, 3(2), 45–50.
• Anusha, M.D., & Shashirekha, H.L. (2020). Developed an ensemble model for the identification of hate speech and offensive content in Indo-European languages. In: FIRE (Working Notes), 253–259.
• Aurpa, T.T., Sadik, R., & Ahmed, M.S. (2022). Detection of abusive Bangla comments on Facebook with transformer-based deep learning models. Social Network Analysis and Mining, 12(1), 24.
• Bagora, A., Shrestha, K., Maurya, K., & Desarkar, M.S. (2022). Detection of hostility in online Hindi-English code-mixed conversations. In: Proceedings of the 14th ACM Web Science Conference, June 26, pp. 390–400.
• Bhattacharyya, P., Kumar, P., & Bhatnagar, V. (2022). Detection of malevolent posts in Hindi. Neurocomputing, 474, 60–81.
• Bohra, A., Vijay, D., Singh, V., Akhtar, S.S., & Shrivastava, M. (2018). A social media text dataset that is code-mixed in Hindi and English to identify hate speech. In: Proceedings of the Second Workshop on Computational Modeling of People's Opinions, Personality, and Emotions in Social Media, pp. 36–41.
• .Chakravarthi, B.R., Kumaresan, P.K., Sakuntharaj, R., Madasamy, A.K., Thavareesan, S., Navaneethakrishnan, S.C., & Mandl, T. (2021). Summary of the HASOC DravidianCodeMix joint effort focused on the detection of objectionable language in Tamil and Malayalam. In: Working Notes of FIRE 2021, CEUR.
• Chawla, N.V., Bowyer, K.W., Hall, L.O., & Kegelmeyer, W.P. (2002). SMOTE: Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16, 321–357.
• Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., ... & Stoyanov, V. (2019). Large-scale unsupervised cross-lingual representation learning. arXiv preprint arXiv:1911.02116.
• Dadkhah, S., Shoeleh, F., Yadollahi, M.M., Zhang, X., & Ghorbani, A.A. (2021). A system for the analysis and detection of real-time hostile behaviors. Applied Soft Computing, 104, 107175.
• Dadvar, M., de Jong, F.M., Ordelman, R., & Trieschnigg, D. (2012). Enhanced cyberbullying detection utilizing gender information. In: Proceedings of the Twelfth Dutch-Belgian Information Retrieval Workshop (DIR 2012), Ghent University, pp. 23–24.
• Davidson, T., Warmsley, D., Macy, M., & Weber, I. (2017). Automated identification of hate speech and the issue of foul language. In: Proceedings of the International AAAI Conference on Web and Social Media, 11, 512–515.
• Devlin, J., Chang, M.W., Lee, K., & Toutanova, K. (2018). BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.
• Farooqi, Z.M., Ghosh, S., & Shah, R.R. (2021). Utilizing transformers for the identification of hate speech in conversational code-mixed tweets. arXiv preprint arXiv:2112.09986.
• Fleiss, J.L. (1971). Assessing nominal scale concordance among many evaluators. Psychological Bulletin, 76(5), 378.
• Gitari, N.D., Zuping, Z., Damien, H., & Long, J. (2015). A lexicon-driven methodology for the identification of hate speech. International Journal of Multimedia and Ubiquitous Engineering, 10(4), 215–230.
• Jahan, M., Ahamed, I., Bishwas, M.R., & Shatabda, S. (2019). Detection of abusive remarks in Bangla-English code-mixed and transliterated text. In: 2nd International Conference on Innovation in Engineering and Technology (ICIET), IEEE, pp. 1–6.
• Kakwani, D., Kunchukuttan, A., Golla, S., Gokul, N.C., Bhattacharyya, A., Khapra, M.M., & Kumar, P. (2020). IndicNLPSuite: Monolingual corpora, assessment benchmarks, and pre-trained multilingual language models for Indian languages. In: Proceedings of EMNLP 2020, pp. 4948–4961.
• Khan, M.M., Shahzad, K., & Malik, M.K. (2021). Detection of hate speech in Roman Urdu. ACM Transactions on Asian Low-Resource Language Information Processing (TALLIP), 20(1), 1–9.
• Khanuja, S., Bansal, D., Mehtani, S., Khosla, S., Dey, A., Gopalan, B., ... & Gupta, S. (2021). MuRIL: Multilingual representations for Indian languages. arXiv preprint arXiv:2103.10730.
• Koufakou, A., Pamungkas, E.W., Basile, V., & Patti, V. (2020). HurtBERT: Integrating lexical attributes with BERT for the identification of abusive language. In: Proceedings of the Fourth Workshop on Online Abuse and Harms, ACL, pp. 34–43.
• Kumari, K., & Singh, J.P. (2020). AI_ML_NIT_Patna at HASOC 2020: BERT models for the identification of hate speech in Indo-European languages. In: FIRE (Working Notes), pp. 319–324.
• Mandl, T., Modha, S., Majumder, P., Patel, D., Dave, M., Mandlia, C., & Patel, A. (2019). An overview of the HASOC track at FIRE 2019: Identification of hate speech and objectionable content in Indo-European languages. In: Proceedings of the 11th Annual Meeting of FIRE, pp. 14–17.
• Mandl, T., Modha, S., Shahi, G.K., Madhu, H., Satapara, S., Majumder, P., Schäfer, J., Ranasinghe, T., Zampieri, M., Nandini, D., & Jaiswal, A.K. (2021). Summary of the HASOC subtrack at FIRE 2021: Identification of hate speech and objectionable content. arXiv preprint arXiv:2112.09301.
• Mathew, B., Saha, P., Yimam, S.M., Biemann, C., Goyal, P., & Mukherjee, A. (2021). HateXplain: A benchmark dataset for explainable hate speech detection. In: Proceedings of the AAAI Conference on Artificial Intelligence, 35, 14867–14875.
• Mehedi, M.H., Al Taawab, A., Tasnia, L., & Dhar, M. (2020). Transliterated classification of Bengali comments from social media. In: 10th Region 10 Humanitarian Technology Conference (R10-HTC), IEEE, pp. 365–371.
• Mishra, A.K., Saumya, S., & Kumar, A. (2020). IIIT_DWD@HASOC 2020: Identifying objectionable content in Indo-European languages. In: FIRE (Working Notes), pp. 139–144.
• Mohtaj, S., Woloszyn, V., & Möller, S. (2020). TUB at HASOC 2020: Character-based LSTM for hate speech identification in Indo-European languages. In: FIRE (Working Notes), pp. 298–303.
• Mollas, I., Chrysopoulou, Z., Karlos, S., & Tsoumakas, G. (2022). ETHOS: A dataset for the detection of multi-label hate speech. Complex Intelligent Systems, 8(6), 4663–4678.
• Mundra, S., Singh, N., & Mittal, N. (2021). Optimize BERT for the classification of hate speech in Hindi-English code-mixed text. In: Forum for Information Retrieval Evaluation (FIRE) (Working Notes), CEUR.
• Nahar, V., Li, X., & Pang, C. (2013). A proficient method for identifying cyberbullying. Communicate Information Science Management Engineering, 3(5), 238.
• Pamungkas, E.W., Basile, V., & Patti, V. (2020). Detection of misogyny on Twitter: A multilingual and cross-domain analysis. Information Processing & Management, 57(6), 102360. [34.] Pathak, V., Joshi, M., Joshi, P., Mundada, M., & Joshi, T. (2020). KBCNMUJAL@HASOC-Dravidian-CodeMix-FIRE20: Employing machine learning for identification of hate speech and offensive code-mixed content. In: FIRE 2020 (Working Notes), CEUR, pp. 351–361.
• Qian, J., Bethke, A., Liu, Y., Belding, E., & Wang, W.Y. (2019). A benchmark dataset for intervention strategies in online hate speech. arXiv preprint arXiv:1909.04251. [36.] Ramadan, S.T., Sakib, T., Rahat, M.A., Hossain, M.M., Rahman, R., & Rahman, M.M. (2022). An embedded system for abusive Bengali speech detection with NLP and deep learning. In: 25th International Conference on Computer and Information Technology (ICCIT), IEEE, pp. 698–703.
• Ravikiran, M., & Annamalai, S. (2021). DOSA: Dataset for identifying objectionable spans in Dravidian code-mixed language. In: Proceedings of the First Workshop on Speech and Language Technologies for Dravidian Languages, pp. 10–17.
• Razavi, A.H., Inkpen, D., Uritsky, S., & Matwin, S. (2010). Detection of offensive language by multi-tiered classification. In: Advances in Artificial Intelligence: 23rd Canadian Conference on AI, Springer, pp. 16–27.
• Remon, N.I., Tuli, N.H., & Akash, R.D. (2022). Detection of Bengali hate speech on Facebook pages. In: International Conference on Innovations in Science, Engineering, and Technology (ICISET), IEEE, pp. 169–173.
• Romim, N., Ahmed, M., Talukder, H., & Islam, M.S. (2021). Hate speech. In: Advances in Computational Intelligence: IJCACI 2020, Springer, pp. 457–468.
• Sai, S., & Sharma, Y.S. (2020). HASOC-Dravidian-CodeMix-FIRE-2020: Identification of multilingual offensive speech in codemixed and romanized text. In: FIRE (Working Notes), pp. 336–343.
• Satapara, S., Modha, S., Mandl, T., Madhu, H., & Majumder, P. (2021). Synopsis of the HASOC subtrack at FIRE 2021: Detection of conversational hate speech in code-mixed language. In: Working Notes of FIRE, pp. 13–31.
• Sadiq, M.T., Naqvi, I.R., Akhter, M.P., Jiangbin, Z., & Abdelmajeed, M. (2020). Automatic detection of objectionable language in Urdu and Roman Urdu. IEEE Access, 15(8), 91213–91226.
• Sazzed, S. (2021). Detection of abusive text in a transliterated Bengali-English social media corpus. In: Proceedings of the Fifth Workshop on Computational Approaches to Linguistic Code-Switching, pp. 125–130.
• Sharma, D., Singh, A., & Singh, V.K. (2024). A high-quality Hindi-English code-mixed dataset for THAR-targeted hate speech against religion, utilizing deep learning models. ACM TALLIP. https://doi.org/10.1145/3653017
• Sharma, D., Gupta, V., & Singh, V.K. (2024). Detection of abusive comments in Tamil with deep learning techniques. In: Computational Intelligence Techniques for Sentiment Analysis in NLP Applications, Morgan Kaufmann, pp. 207–226.
• Sharma, D., Gupta, V., & Singh, V.K. (2022). Identification of homophobia and transphobia in Malayalam and Tamil: Investigating deep learning techniques. In: International Conference on Advanced Network Technologies and Intelligent Computing, Springer, pp. 217–226.
• Singh, A., Sharma, D., & Singh, V.K. (2024). MIMIC: Identification of misogyny in multimodal internet content in Hindi-English code-mixed language. ACM TALLIP. https://doi.org/10.1145/3656169
• Singh, P., & Bhattacharyya, P. (2021). CFILT IIT Bombay at HASOC-Dravidian-CodeMix FIRE 2020: Supporting transformers using random transliteration. In: FIRE (Working Notes), pp. 411–416.
• N. Tyagi, G. K. Sharma, and N. K. Sharma, "Combating hate speech: Challenges and solutions in detection techniques," in Proc. PiCET, pp. 1741-1746, 2025. doi:10.1049/icp.2025.1705.
• Sutejo, T.L., & Lestari, D.P. (2018). Detection of hate speech in Indonesia with deep learning techniques. In: International Conference on Asian Language Processing (IALP), IEEE, pp. 39–43.
• Vargas, F., de Góes, F.R., Carvalho, I., Benevenuto, F., & Pardo, T.A. (2021). Contextual lexicon methodology for abusive language detection. arXiv preprint arXiv:2104.12265.
• Neha Tyagi, Gopal Krishna Sharma, Narendra Kumar Sharma. A Hybrid MuRIL–Attention–Random Forest Framework for Hate Speech Detection Against Women in Hindi. International Journal of Computer Applications. 187, 96 (Apr 2026), 51-59. DOI=10.5120/ijca2ad422afaaf6
• Waseem, Z., & Hovy, D. (2016). Hateful symbols or malevolent individuals? Predictive attributes for the identification of hate speech on Twitter. In: Proceedings of the NAACL Student Research Workshop, pp. 88–93.
• Bölücü, N., & Canbay, P. (2021). Hate speech detection on Turkish Twitter using deep learning techniques. In Proceedings of the International Conference on Innovations in Intelligent Systems and Applications (ASYU), IEEE, pp. 1–6.
• Neha Tyagi, Gopal Krishna Sharma, and Narendra Kumar Sharma Predictive Policing 2.0: AI Techniques for Hate Speech Monitoring and Intervention, E-Book ISBN: 978-1-77964-784-9.