Comprehensive overview of A Transformer-Based Model for Detecting Fake news and Sentiment Trends on the Social Media Platform

Abstract

The spread of information, including fake content and deceptive narratives, has been greatly expedited by the quick development of social media platforms, especially Twitter. It is a difficult but critical responsibility to detect such misleading information while also investigating sentiment patterns. This review paper looks at transformer-based unified models that incorporate sentiment analysis and false tweet detection. Transformer architecture creation, usage in social media analytics, datasets, methodologies, assessment criteria, and obstacles are all discussed. The study stresses how transformer-based models, such as BERT, RoBERTa, and XLNet, outperform traditional machine learning algorithms due to their attention mechanisms and contextual knowledge. Finally, future research directions are discussed, including explainable AI and multimodal learning.

Introduction

This research focuses on detecting fake news on social media while simultaneously analyzing the sentiment of users' posts, particularly tweets. Misinformation, including fake news and rumors, can influence emotions, decisions, and social behavior, making automated and reliable detection increasingly important. Traditional machine-learning methods such as Naïve Bayes, SVM, and Logistic Regression often struggle with context, sarcasm, polysemy, long-range dependencies, and differences between domains.

Recent research shows that transformer-based models such as BERT, RoBERTa, and XLNet provide better contextual understanding and generally outperform conventional machine-learning and older deep-learning approaches for fake-news detection and sentiment analysis. Hybrid approaches, including BERT-CNN, BERT-LSTM, RoBERTa-GNN, attention mechanisms, and explainable-AI methods such as SHAP and LIME, further improve performance. Some reported models achieve very high accuracy, including approximately 98–99% on selected benchmark datasets.

However, the literature identifies several important limitations:

• Most studies treat fake-news detection and sentiment analysis as separate tasks rather than combining them into one model.
• Large transformer models require substantial computational resources, making real-time social-media deployment difficult.
• Transformer models often have poor explainability, functioning as black boxes.
• Most existing systems focus only on text, ignoring images, videos, and temporal information.
• Dataset bias and domain adaptation can reduce performance when models are applied to new datasets or languages.
• Multilingual and real-time misinformation detection remain insufficiently explored.

Proposed Research

To address these gaps, the research proposes a Hybrid Unified Transformer Model with Explainability for Twitter. The proposed framework combines:

• RoBERTa and DeBERTa transformer encoders for contextual text representation.
• Graph Neural Networks (GNNs) to capture relationships and social propagation patterns.
• Multimodal data fusion to incorporate textual, visual, and potentially temporal information.
• Explainable AI (XAI) to make model predictions easier to understand.
• A unified architecture that performs fake-news detection and sentiment analysis simultaneously.

Conclusion

Transformer-based unified models\' increased ability to capture contextual semantics, linguistic relationships, and hidden patterns inside social media text makes them a significant advancement in the fields of sentiment analysis and fraudulent tweet identification on Twitter. Transformer architectures like BERT, RoBERTa, and XLNet employ self-attention mechanisms to comprehend the contextual relationship between words, in contrast to conventional machine learning and deep learning techniques. This enables more precise detection of fake information and emotional polarity in tweets.By concurrently identifying fake news and categorizing sentiment using shared representations, multi-task learning increases model efficiency by decreasing redundancy and boosting generalization capacity. Additionally, transformer-based models can handle noisy and unstructured Twitter data that includes caustic words, hashtags, abbreviations, and emojis—all of which are difficult to process with conventional techniques. Future studies should therefore concentrate on developing multimodal, scalable, and explainable transformer-based systems that can perform sentiment analysis and real-time misinformation detection. While multimodal learning frameworks that integrate textual, visual, and temporal input can significantly increase detection accuracy and resilience, Explainable Artificial Intelligence (XAI) techniques like SHAP and LIME can enhance model transparency and user confidence. Additionally, the application of Large Language Models (LLMs), federated learning, and Graph Neural Networks (GNNs) may improve cross-domain adaption, propagation analysis, and contextual comprehension. These advancements will aid in the creation of social media monitoring systems that are more practical, intelligent, dependable, and efficient.

References

[1] Z. Liu, T. Zhang, K. Yang, P. Thompson, Z. Yu, and S. Ananiadou, “Emotion detection for misinformation?: A review,” vol. 107, no. November 2023, 2024. [2] M. Cheng, S. Wang, X. Yan, T. Yang, and W. Wang, “A COVID-19 Rumor Dataset,” vol. 12, no. May, pp. 1–10, 2021, doi: 10.3389/fpsyg.2021.644801. [3] B. Pang, L. Lee, H. Rd, and S. Jose, “Thumbs up?? Sentiment Classification using Machine Learning Techniques,” no. July, pp. 79–86, 2002. [4] J. Yi, T. Nasukawa, R. Bunescu, and W. Niblack, “Sentiment analyzer: Extracting sentiments about a given topic using natural language processing techniques,” Proc. - IEEE Int. Conf. Data Mining, ICDM, pp. 427–434, 2003, doi: 10.1109/icdm.2003.1250949. [5] L. Zhang and L. Corporation, “Deep Learning for Sentiment Analysis?: A Survey”. [6] B. Pang and L. Lee, “Opinion mining and sentiment analysis,” vol. 2, no. 1, 2008. [7] A. Vaswani, “Attention Is All You Need,” no. Nips, 2017. [8] X. Glorot, “Domain Adaptation for Large-Scale Sentiment Classification?: A Deep Learning Approach,” no. 1, 2011. [9] S. Fitria and N. Azizah, “Performance Analysis of Transformer Based Models ( BERT , ALBERT and RoBERTa ) in Fake News Detection,” pp. 1–6, 2018. [10] Z. Yang, Z. Dai, Y. Yang, and J. Carbonell, “XLNet?: Generalized Autoregressive Pretraining for Language Understanding,” no. NeurIPS, pp. 1–18, 2019. [11] Y. Liu et al., “RoBERTa: A Robustly Optimized BERT Pretraining Approach,” no. 1, 2019. [12] F. B. J. Camacho-collados and L. N. L. Espinosa-anke, “T WEET E VAL?: Unified Benchmark and Comparative Evaluation for Tweet Classification,” pp. 1644–1650, 2020. [13] X. Zhou and R. Zafarani, “A Survey of Fake News: Fundamental Theories, Detection Methods, and Opportunities,” 2020. [14] U. Comments, S. K. Hamed, M. Juzaiddin, A. Aziz, and M. R. Yaakub, “Fake News Detection Model on Social Media by Leveraging Sentiment Analysis of News Content and Emotion Analysis of,” 2023. [15] F. A. K. E. N. E. W. S. D. Etection, “BERT-BASED BLENDED APPROACH FOR FAKE NEWS DETECTION,” vol. 2, no. 1, pp. 7–15, 2024, doi: 10.54116/jbdai.v2i1.27. [16] J. Rout and M. Mishra, “Towards Reliable Fake News Detection?: Enhanced Attention-Based Transformer Model,” pp. 1–22, 2025. [17] A. Saadi, “Enhancing Fake News Detection with Transformer Models and Summarization,” vol. 15, no. 3, pp. 23253–23259, 2025. [18] C. Kumar, M. Bansal, M. A. Khan, and V. Kaushik, “Graph-augmented transformer ensemble framework for robust and scalable fake news detection in social media ecosystems,” pp. 1–31, 2026. [19] T. Li, Y. Sun, S. Hsu, Y. Li, and R. C. Wong, “Fake News Detection with Heterogeneous Transformer”. [20] S. Raza, D. Paulen-patterson, C. Ding, and C. Street, “Fake News Detection?: Comparative Evaluation of BERT-like Models and Large Language Models with Generative AI-Annotated Data,” pp. 1–30. [21] M. C. Kenton, L. Kristina, and J. Devlin, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” no. Mlm, 1953. [22] R. Uddin, A. Basit, Y. Khan, S. J. Shazib, and S. Hossain, “BERT-Based Fake News Detection?: A Transformer-Driven Approach for Misinformation Classification on Twitter,” vol. 16, no. 1, pp. 1–13. [23] J. Chen, Y. Zhang, Y. Pan, P. Xu, and C. Guan, “A Transformer-based deep neural network model for SSVEP classification”. [24] A. Khalil, M. Jarrah, and M. Aldwairi, “Hybrid Neural Network Models for Detecting Fake News Articles,” Human-Centric Intell. Syst., vol. 4, no. 1, pp. 136–146, 2024, doi: 10.1007/s44230-023-00055-x. [25] K. Aalijah, “International Journal of Research Publication and Reviews Leveraging Social Media Analytics for Sustainability Trend Detection in Saudi Arabia ’ s Evolving Market,” vol. 6, no. 3, pp. 5540–5548, 2025. [26] M. A. Alonso, D. Vilares, and C. Gómez-rodríguez, “Sentiment Analysis for Fake News Detection,” 2021. [27] D. Amangeldi, A. Usmanova, and P. Shamoi, “Understanding Environmental Posts?: Sentiment and Emotion Analysis of Social Media Data,” IEEE Access, vol. 12, no. February, pp. 33504–33523, 2024, doi: 10.1109/ACCESS.2024.3371585. [28] R. Anggrainingsih, G. Mubashar, and H. Amitava, Transformer based models for combating rumours on microblogging platforms?: a review, vol. 57, no. 8. Springer Netherlands, 2024. doi: 10.1007/s10462-024-10837-9. [29] S. V Balshetwar, A. Rs, and D. J. R, “Fake news detection in social media based on sentiment analysis using classifier techniques Fake news detection in social media based on sentiment analysis using classifier techniques,” no. March, 2023, doi: 10.1007/s11042-023-14883-3. [30] A. Bhardwaj, S. Bharany, and S. Kim, “Heliyon Fake social media news and distorted campaign detection framework using sentiment analysis & machine learning,” Heliyon, vol. 10, no. 16, p. e36049, 2024, doi: 10.1016/j.heliyon.2024.e36049. [31] Z. Boukouvalas and A. Shafer, “Role of Statistics in Detecting Misinformation?: A Review of the State of the Art , Open Issues , and Future Research Directions,” pp. 27–50, 2024. [32] Y. Cao, J. R. W. Ii, and D. Shallcross, “Can Large Language Models Detect Misinformation in Scientific News Reporting??”. [33] K. L. Chandrashekar and S. Darshan, “Twitter Sentiment Analysis Using NLP Models and Real-Time Tweet Fetching,” 2025. [34] P. Feldman, S. Tiwari, C. S. L. Cheah, J. R. Foulds, and S. Pan, “Analyzing COVID-19 Tweets with Transformer-based Language Models,” vol. 2, no. Figure 1, pp. 1–6. [35] M. Gupta, I. Researcher, and U. States, “Understanding Social Media Behavior?: A Review of Sentiment , Engagement , and Temporal Trends”. [36] F. Hafeez, M. Hafeez, M. Shaff, B. Imran, A. Qasim, and N. Hussain, “A Comparative Study of Deep Learning and Transformer Models for Twitter Sentiment Analysis,” vol. 17, no. April, pp. 85–89, 2026. [37] S. K. Hamed, M. Juzaiddin, A. Aziz, and M. R. Yaakub, “Fake News Detection Model on Social Media by Leveraging Sentiment Analysis of News Content and Emotion Analysis of Users ’ Comments”. [38] M. Islam, M. Abaker, and A. Daud, “Unified Large Language Models for Misinformation Detection in Low-Resource Linguistic Settings”. [39] A. Liu, Q. Sheng, and X. Hu, “Preventing and Detecting Misinformation Generated by Large Language Models,” pp. 3001–3004, 2024, doi: 10.1145/3626772.3661377. [40] M. S. Mredula, N. Dey, S. Rahman, I. Mahmud, and Y. Cho, “A Review on the Trends in Event Detection by Analyzing Social Media Platforms ’ Data Media Platforms ’ Data”.

Copyright

Copyright © 2026 Gurpreet Kaur, Dr. Jagdev Singh Rana. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.