---

Discontinuity, Nonlinearity, and Complexity 14(4) (2025) 745--756 | DOI:10.5890/DNC.2025.12.010

Shah Hemal Girishkumar, Dr. Hiren Joshi

Department of Computer Science Gujarat University, Navrangpura, Ahmedabad, Gujarat 380009

Download Full Text PDF

 

Abstract

Detecting unhealthy conversations online presents significant challenges, especially regarding class imbalance and the nuanced features of social media language. Class imbalance can lead to biased models and poor performance, particularly in identifying minority class instances. Furthermore, existing methods often struggle to accurately detect unhealthy conversations due to the complexity of language nuances and the sheer volume of online discourse. To address these challenges, this paper presents the Stratified RoBERTa Enhanced Framework for detecting unhealthy conversations online. This framework employs stratified sampling during data pre-processing to ensure proper distribution and preservation of minority classes, effectively mitigating the negative impact of class imbalance. Additionally, we introduce a novel technique using the Hugging Face Auto Tokenizer to enhance tokenisation efficiency. The proposed approach utilizes a neural network architecture that integrates a pre-trained DistilRoBERTa-base model, followed by a hidden layer with ReLU activation. Fine-tuning with the Adam optimizer further enhances the model's adaptability to varying learning rates. Experimental results, illustrated through Receiver Operating Characteristic (ROC) graphs, demonstrate improved true positive rates and false positive rates, affirming the efficacy of the proposed framework in accurately detecting unhealthy conversations.

References

1. [1]	Thompson, J.B. (2020), Mediated interaction in the digital age, Theory, Culture \& Society, 37(1), 3-28.

2. [2]	Laor, T. (2022), Twitter as a clique: Journalists' patterns of Twitter use in Israel, Online Information Review, 46(1), 40-58.

3. [3]	Alkharashi, A.A. (2021), Exploring the Characteristics of Abusive Behaviour in Online Social Media Settings, (Doctoral dissertation, University of Glasgow).

4. [4]	Herhausen, D., Ludwig, S., Grewal, D., Wulf, J., and Schoegel, M. (2019), Detecting, preventing, and mitigating online firestorms in brand communities, Journal of Marketing, 83(3), 1-21.

5. [5]	Qiu, H., Zhao, T., Li, A., Zhang, S., He, H., and Lan, Z. (2023, October), A benchmark for understanding dialogue safety in mental health support, in CCF International Conference on Natural Language Processing and Chinese Computing, (Cham: Springer Nature Switzerland), 1-13.

6. [6]	Shukla, V.K., Fekih, A., Joshi, M.C., and Mishra, P.K. (2024), Study of finite-time synchronization between memristive neural networks with leakage and mixed delays, International Journal of Dynamics and Control, 12(5), 1541-1553.

7. [7]	Shukla, V.K., Joshi, M.C., Mishra, P.K., and Xu, C. (2024), Adaptive fixed-time difference synchronization for different classes of chaotic dynamical systems, Physica Scripta, 99(9), 095264.

8. [8]	Marshan, A., Nizar, F.N.M., Ioannou, A., and Spanaki, K. (2023), Comparing machine learning and deep learning techniques for text analytics: detecting the severity of hate comments online, Information Systems Frontiers, 1-19.

9. [9]	Jaques, N., Shen, J.H., Ghandeharioun, A., Ferguson, C., Lapedriza, A., Jones, N., and Picard, R. (2020), Human-centric dialog training via offline reinforcement learning, arXiv preprint arXiv:2010.05848.

10. [10]	Saveski, M., Roy, B., and Roy, D. (2021), The structure of toxic conversations on Twitter, in Proceedings of the Web Conference 2021, 1086-1097.

11. [11]	Gilda, S., Giovanini, L., Silva, M., and Oliveira, D. (2022), Predicting different types of subtle toxicity in unhealthy online conversations, Procedia Computer Science, 198, 360-366.

12. [12]	Kovacs, E.R., Cotfas, L.A., and Delcea, C. (2023), January 6th on Twitter: measuring social media attitudes towards the Capitol riot through unhealthy online conversation and sentiment analysis, Journal of Information and Telecommunication, 1-22.

13. [13]	Leung, S. and Papapolyzos, F. (2023), Hashing it out: predicting unhealthy conversations on Twitter, arXiv preprint arXiv:2311.10596.

14. [14]	Saravani, S.M., Ray, I., and Ray, I. (2021, December), Automated identification of social media bots using deepfake text detection, in International Conference on Information Systems Security, (Cham: Springer International Publishing), 111-123.

15. [15]	Sadeghi, F., Bidgoly, A.J., and Amirkhani, H. (2022), Fake news detection on social media using a natural language inference approach, Multimedia Tools and Applications, 81(23), 33801-33821.

16. [16]	Fadhli, I., Hlaoua, L., and Omri, M.N. (2023), Deep learning-based credibility conversation detection approaches from social network, Social Network Analysis and Mining, 13(1), 57.

17. [17]	Jang, Y., Park, C.H., Lee, D.G., and Seo, Y.S. (2021), Fake news detection on social media: a temporal-based approach, Computers, Materials \& Continua, 69(3).

18. [18]	Monti, F., Frasca, F., Eynard, D., Mannion, D., and Bronstein, M.M. (2019), Fake news detection on social media using geometric deep learning, arXiv preprint arXiv:1902.06673.

19. [19]	Ozbay, F.A. and Alatas, B. (2019), A novel approach for detection of fake news on social media using metaheuristic optimization algorithms, Elektronika ir Elektrotechnika, 25(4), 62-67.

20. [20]	Truica, C.O. and Apostol, E.S. (2022), MisRoBERTa: transformers versus misinformation, Mathematics, 10, 569.

21. [21]	Sanh, V., Debut, L., Chaumond, J., and Wolf, T. (2019), DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter, in Proceedings of the Workshop on Energy Efficient Machine Learning and Cognitive Computing, Vancouver, BC, Canada, 13 December 2019, 1-5.

22. [22]	Agbesi, V.K., Chen, W., Yussif, S.B., Hossin, M.A., Ukwuoma, C.C., Kuadey, N.A., and Al-antari, M.A. (2023), Pre-trained transformer-based models for text classification using low-resourced Ewe language, Systems, 12(1), 1.

23. [23]	Price, I., Gifford-Moore, J., Fleming, J., Musker, S., Roichman, M., Sylvain, G., and Sorensen, J. (2020), Six attributes of unhealthy conversation, arXiv preprint arXiv:2010.07410.

24. [24]	Kovacs, E.R., Cotfas, L.A., and Delcea, C. (2024), January 6th on Twitter: measuring social media attitudes towards the Capitol riot through unhealthy online conversation and sentiment analysis, Journal of Information and Telecommunication, 8(1), 108-129.