Leveraging DistilBERT-Multilingual for Robust and Efficient AI-Based Fake News Detection

Abstract

In this research, a machine learning–based framework for fake news detection using the DistilBERT-base-multilingual-cased Transformer model is proposed. The rapid growth of social media platforms has significantly increased the spread of misinformation, making accurate and scalable fake news detection a critical challenge. To address this issue, the proposed system focuses on multilingual text classification while maintaining low computational complexity, making it suitable for real-world applications. A large and diverse training dataset is created by merging multiple open-source datasets, enabling the model to effectively learn linguistic patterns across different languages and sources. Prior to training, the textual data is carefully pre-processed through cleaning, normalization, and tokenization to improve classification accuracy and reduce unnecessary computational overhead. The DistilBERT model is then fine-tuned to classify news content as real or fake based on semantic and contextual information. Extensive experiments conducted on social media–based text data demonstrate that the proposed approach achieves an accuracy of 88.97% and a precision of 0.99 in detecting fake misinformation. These results highlight the effectiveness of lightweight Transformer-based architectures in identifying misinformation while maintaining efficiency and scalability. The study confirms that artificial intelligence–driven and deep learning–based models can play a significant role in mitigating the spread of fake news and improving the reliability of online information ecosystems.