A Hybrid Multi-Dimensional Fake News Detection Framework Using Transformer-Based Contextual Learning and Multi-Source Credibility Fusion

Fake news dissemination has become a critical challenge in modern digital ecosystems, significantly influencing public opinion, political stability, and public health outcomes. The problem is particularly severe in multilingual and socio-culturally diverse regions such as India, where linguistic variability, code-mixed text, and the widespread use of encrypted communication platforms limit the effectiveness of existing detection systems. This paper proposes a novel multi-dimensional hybrid framework for fake news detection that integrates four complementary analytical components: content intelligence, source credibility assessment, claim verification, and social propagation modelling. Unlike traditional approaches that rely on a single analytical dimension, the proposed system combines transformer-based contextual understanding using DistilBERT with traditional machine learning techniques and behavioural signal analysis to capture both semantic and structural characteristics of misinformation. A comprehensive experimental evaluation is conducted on a combined dataset of over 45,000 labelled news articles collected from multiple sources. The proposed framework achieves an accuracy of 97.8%, outperforming baseline machine learning models and demonstrating improved robustness over standalone transformer-based approaches. Additionally, a detailed ablation study is performed to quantify the contribution of each module, showing that the hybrid integration leads to a measurable performance gain of 3–5% compared to content-only models. To address the challenge of model interpretability, the framework incorporates a weighted scoring mechanism that provides transparent, module-wise contributions to the final decision. This enhances trust and usability in real-world applications. The results demonstrate that the proposed approach not only improves detection accuracy but also ensures scalability, interpretability, and adaptability in dynamic and multilingual information environments. The framework’s efficacy is demonstrated through extensive testing on standardized datasets, achieving a 99.1% accuracy rate, providing a robust foundation for deployment in complex digital ecosystems like India.