Fake News Detection In Code-Mixed Hindi–English Text Using Mbert and Mt5: Cross-lingual, Script aware approach

The widespread use of social media platforms has transformed the way information is shared and consumed across the world. Although these platforms provide rapid access to news and communication, they have also become major channels for the spread of misleading and fabricated information. Fake news has become a critical challenge that can influence public opinion, disrupt social harmony, and create misinformation among communities. Detecting false information in online environments has therefore become an essential task in natural language processing.The challenge becomes even more complex when the textual content is written in code-mixed languages. In multilingual countries such as India, social media users frequently combine multiple languages within the same sentence, especially Hindi and English. These mixed-language texts often contain informal writing styles, transliteration, abbreviations, and inconsistent grammar. Traditional machine learning models struggle to process such data because they are primarily designed for single-language text analysis.To address this issue, a multilingual deep learning framework based on transformer architectures is utilized for detecting fake news in code-mixed Hindi–English text. The system employs advanced models such as Multilingual Bidirectional Encoder Representations from Transformers (MBERT) and Multilingual Text-to-Text Transfer Transformer (MT5). These models are capable of capturing contextual relationships between words across multiple languages and generating meaningful textual representations.The framework processes social media text through several stages including preprocessing, tokenization, contextual embedding generation, and classification. The transformer models analyze semantic relationships and identify linguistic patterns associated with misinformation. Experimental evaluation demonstrates that transformer-based approaches significantly improve classification accuracy compared to traditional machine learning techniques