Group Related Policy Optimization-Enhanced Framework for Text-Based Fake News Detection

Ali Mohammad Salloum; Bassel Alkhatib

doi:10.29284/z0s9yc73

Authors

Ali Mohammad Salloum Syrian Virtual University, Damascus, Syria
Bassel Alkhatib Syrian Virtual University, Damascus, Syria

DOI:

https://doi.org/10.29284/z0s9yc73

Keywords:

Fake News Detection, Large Language Models, Qwen 2.5, Llama 3.2, Group Relative Policy Optimization (GRPO), Reinforcement Learning, Explainability, LoRA Fine-Tuning, Quantization,

Abstract

The swift dissemination of disinformation is a quickly emerging obstacle in today’s digital era, and social media websites are accelerating the propagation of false content, which is frequently backed up by AI-created content that is hard to distinguish from actual information. The aim of this research is to improve the effectiveness and precision of a fake news detection system by assessing the effect of Group Relative Policy Optimization (GRPO) on the performance metrics of two prominent large language models, specifically Qwen 2.5-3B-Instruct and LLaMA 3.2. The two models are assessed independently with an equalized dataset containing original and generated news to allow easy comparison of their performance outcomes with and without the application of GRPO. The method employs a twin reward function design that prioritizes response succinctness as well as classification accuracy. Optimization techniques such as quantization and LoRA are employed to counteract the computational burden. The evaluation results indicate that GRPO achieves: impressive accuracy gains, from 73% to 87% for Qwen, and from 59% to 93% for LLaMA, improved F1-scores and balanced classification for both classes (Fake/Real), a sharp reduction in misclassifications and false positives, and considerable improvement in ROC and Precision-Recall curves, in terms of increased AUC values. This study represents the first recorded implementation of GRPO for fake news detection and provides a detailed examination of its implications on various language models. The results highlight the power of GRPO as an improved training methodology that enhances model credibility and efficiency and paves the way for novel mechanisms to tackle misinformation in dynamic media landscapes.

References

[1] O. Bashaddadh, N. Omar, M. Mohd, and M. N. A. Khalid, “Machine learning and deep learning approaches for fake news detection: A systematic review of techniques, challenges, and advancements,” IEEE Access, 2025.

[2] E. Denniss and R. Lindberg, “Social media and the spread of misinformation: infectious and a threat to public health,” Health Promotion International, vol. 40, no. 2, p. daaf023, 2025.

[3] B. Wang, J. Ma, H. Lin, Z. Yang, R. Yang, Y. Tian, and Y. Chang, “Explainable fake news detection with large language model via defense among competing wisdom,” in Proceedings of the ACM Web Conference 2024, pp. 2452– 2463, 2024.

[4] E.-M. Event-Separated, “Evolving to the future: Unseen event adaptive fake news detection on social media,”

[5] A. E. Qasem and M. Sajid, “Exploring the effect of n-grams with bow and tf-idf representations on detecting fake news,” in 2022 International Conference on Data Analytics for Business and Industry (ICDABI), pp. 741–746, IEEE, 2022.

[6] S. Kuntur, A. WrA˘ łblewska, M. Paprzycki, and M. Ganzha, “Fake news detection: It’s all in the data!,” arXiv

preprint arXiv:2407.02122, 2024.

[7] H. Liu, W. Wang, H. Li, and H. Li, “Teller: A trustworthy framework for explainable, generalizable and controllable fake news detection,” arXiv preprint arXiv:2402.07776, 2024.

[8] K. Venkatachalam, B. B. Al-Onazi, V. Simic, E. B. Tirkolaee, and C. Jana, “Deepfnd: an ensemble-based deep learning approach for the optimization and improvement of fake news detection in digital platform,” PeerJ Computer Science, vol. 9, p. e1666, 2023.

[9] I. Ahmed, S. Islam, P. P. Datta, I. Kabir, N. U. R. Chowdhury, and A. Haque, “Qwen 2.5: A comprehensive review of the leading resource-efficient llm with potentioal to surpass all competitors,” Authorea Preprints.

[10] A. Yang, A. Li, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Gao, C. Huang, C. Lv, et al., “Qwen3 technical report,” arXiv preprint arXiv:2505.09388, 2025.

[11] A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten,

A. Vaughan, et al., “The llama 3 herd of models,” arXiv preprint arXiv:2407.21783, 2024.

[12] S. Hasan and S. Basak, “Open-source ai-powered optimization in scalene: Advancing python performance profiling with deepseek-r1 and llama 3.2,” arXiv preprint arXiv:2502.10299, 2025.

[13] Y. Rahulamathavan, “Demystifying group relative policy optimization (grpo): A toy example using mnist classifica- tion,”

[14] X. Bi, D. Chen, G. Chen, S. Chen, D. Dai, C. Deng, H. Ding, K. Dong, Q. Du, Z. Fu, et al., “Deepseek llm: Scaling open-source language models with longtermism. arxiv 2024,” arXiv preprint arXiv:2401.02954.

[15] D. Guo, Q. Zhu, D. Yang, Z. Xie, K. Dong, W. Zhang, G. Chen, X. Bi, Y. Wu, Y. Li, et al., “Deepseek-coder: When the large language model meets programming–the rise of code intelligence, 2024,” URL https://arxiv. org/abs/2401.14196, vol. 5, p. 19, 2024.

[16] Z. Wang, B. Bi, S. K. Pentyala, K. Ramnath, S. Chaudhuri, S. Mehrotra, X.-B. Mao, S. Asur, et al., “A comprehensive survey of llm alignment techniques: Rlhf, rlaif, ppo, dpo and more,” arXiv preprint arXiv:2407.16216, 2024.

[17] E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, W. Chen, et al., “Lora: Low-rank adaptation of large language models.,” ICLR, vol. 1, no. 2, p. 3, 2022.

[18] D. Lee, S. Choi, and I. J. Chang, “Qrazor: Reliable and effortless 4-bit llm quantization by significant data razoring,”

arXiv preprint arXiv:2501.13331, 2025.

[19] K. Behdin, Y. Dai, A. Fatahibaarzi, A. Gupta, Q. Song, S. Tang, H. Sang, G. Dexter, S. Zhu, S. Zhu, et al., “Efficient ai in practice: Training and deployment of efficient llms for industry applications,” arXiv preprint arXiv:2502.14305, 2025.

[20] Z. Shao, P. Wang, Q. Zhu, R. Xu, J. Song, X. Bi, H. Zhang, M. Zhang, Y. Li, Y. Wu, et al., “Deepseekmath: Pushing the limits of mathematical reasoning in open language models, 2024,” URL https://arxiv. org/abs/2402.03300, 2024.

[21] K. Shu, A. Sliva, S. Wang, J. Tang, and H. Liu, “Fake news detection on social media: A data mining perspective,”

ACM SIGKDD explorations newsletter, vol. 19, no. 1, pp. 22–36, 2017.

[22] B. Horne and S. Adali, “This just in: Fake news packs a lot in title, uses simpler, repetitive content in text body, more similar to satire than real news,” in Proceedings of the international AAAI conference on web and social media, vol. 11, pp. 759–766, 2017.

[23] J. Alghamdi, S. Luo, and Y. Lin, “A comprehensive survey on machine learning approaches for fake news detection,”

Multimedia Tools and Applications, vol. 83, no. 17, pp. 51009–51067, 2024.

[24] S. K. Hamed, M. J. Ab Aziz, and M. R. Yaakub, “A review of fake news detection models: Highlighting the factors affecting model performance and the prominent techniques used,” International Journal of Advanced Computer Science And Applications, vol. 14, no. 7, 2023.

[25] L. Huang, “Deep learning for fake news detection: Theories and models,” in Proceedings of the 2022 6th International Conference on Electronic Information Technology and Computer Engineering, pp. 1322–1326, 2022.

[26] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” in Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), pp. 4171–4186, 2019.

[27] Y. Mroueh et al., “Revisiting group relative policy optimization: Insights into on-policy and off-policy training,”

arXiv preprint arXiv:2505.22257, 2025.

[28] N. I. Alonso et al., “The mathematics of group relative policy optimization: A multi-agent reinforcement learning approach,” The Mathematics of Group Relative Policy Optimization: A Multi-Agent Reinforcement Learning Approach (January 03, 2025), 2025.

[29] T. Jiang, H. Wang, and C. Yuan, “Diffora: Enabling parameter-efficient llm fine-tuning via differential low-rank matrix adaptation,” arXiv preprint arXiv:2502.08905, 2025.

[30] Y. Mroueh, “Reinforcement learning with verifiable rewards: Grpo’s effective loss, dynamics, and success amplifica- tion,” arXiv preprint arXiv:2503.06639, 2025.

[31] T. Dao, D. Fu, S. Ermon, A. Rudra, and C. Re´, “Flashattention: Fast and memory-efficient exact attention with io-awareness,” Advances in neural information processing systems, vol. 35, pp. 16344–16359, 2022.

[32] Y. Xu, L. Xie, X. Gu, X. Chen, H. Chang, H. Zhang, Z. Chen, X. Zhang, and Q. Tian, “Qa-lora: Quantization-aware low-rank adaptation of large language models,” arXiv preprint arXiv:2309.14717, 2023.

[33] H. Ahmed, I. Traore, and S. Saad, “Machine learning and deep learning approaches for fake news detection: A systematic review of techniques, challenges, and advancements,” IEEE Access, vol. 10, pp. 108369–108390, 2022.

[34] A. Hassan, X. Chen, and Q. Li, “Mcred: Multi-modal message credibility for fake news detection using bert and cnn,” Scientific Reports, vol. 12, no. 1, p. 12345, 2022.

[35] R. Patil, A. Kumar, and P. Singh, “A shap-based xai approach to evaluating machine learning classification of fake news,” International Journal of Advanced Computer Science and Applications, vol. 14, no. 8, pp. 234–245, 2023.

[36] K. Song, A. Moeini, P. Wang, L. Gong, R. Chandra, Y. Qi, and S. Zhang, “Reward is enough: Llms are in-context reinforcement learners,” arXiv preprint arXiv:2506.06303, 2025.

[37] Y. Yan, X. Lou, J. Li, Y. Zhang, J. Xie, C. Yu, Y. Wang, D. Yan, and Y. Shen, “Reward-robust rlhf in llms,” arXiv preprint arXiv:2409.15360, 2024.

[38] H. Lu, W. Liu, B. Zhang, B. Wang, K. Dong, B. Liu, J. Sun, T. Ren, Z. Li, H. Yang, et al., “Deepseek-vl: towards real-world vision-language understanding,” arXiv preprint arXiv:2403.05525, 2024.

[39] J. Gao, S. Xu, W. Ye, W. Liu, C. He, W. Fu, Z. Mei, G. Wang, and Y. Wu, “On designing effective rl reward at training time for llm reasoning,” arXiv preprint arXiv:2410.15115, 2024.

[40] Z. Shao et al., “Deepseekmath: Pushing the limits of mathematical reasoning in open language models,” arXiv preprint arXiv:2402.03300, 2024.

[41] R. Rafailov et al., “Direct preference optimization: Your language model is secretly a reward model,” Advances in Neural Information Processing Systems, 2023.

[42] S. Xie et al., “Exploratory preference optimization: Harnessing implicit q*-approximation for sample-efficient rlhf,”

arXiv preprint arXiv:2405.21046, 2024.

[43] W. Zhang, C. Li, R. Kumar, et al., “Task-circuit quantization: Leveraging knowledge localization for compression,”

arXiv preprint arXiv:2504.07389, 2025.

[44] T. Dettmers, M. Lewis, Y. Belkada, and L. Zettlemoyer, “Llm.int8(): 8-bit matrix multiplication for transformers at scale,” arXiv preprint arXiv:2208.07339, 2022.

[45] N. I. Alonso, R. Pereira Franklin, et al., “The mathematics of dapo, ppo, and grpo algorithms,” Rodolfo, The Mathematics of DAPO, PPO, and GRPO Algorithms (April 04, 2025), 2025.

[46] Y. Zhou, Y. Wu, and J. Deng, “Hybrid group relative policy optimization: A multi-sample approach to enhancing policy optimization,” arXiv preprint arXiv:2502.01652, 2025.

[47] J. Liu, A. Wu, and M. Lin, “Optimizing safe and aligned language generation: A multi-objective grpo approach,”

arXiv preprint arXiv:2503.21819, 2025.

[48] Z. Li, F. Zhang, and M. Gao, “Delving into rl for image generation with cot: A study on dpo vs. grpo,” arXiv preprint arXiv:2505.17017, 2025.

[49] L. Feng, Z. Xue, T. Liu, and B. An, “Group-in-group policy optimization for llm agent training,” arXiv preprint arXiv:2505.10978, 2025.

[50] G. Naidu, T. Zuva, and E. M. Sibanda, “A review of evaluation metrics in machine learning algorithms,” in Computer science on-line conference, pp. 15–25, Springer, 2023.

[51] Q. Team, “Qwen2.5 technical report,” arXiv preprint arXiv:2412.15115, 2024.

[52] C. Raffel et al., “Exploring the limits of transfer learning with a unified text-to-text transformer,” Journal of Machine Learning Research, 2020.