MTL‑rtFND: Multimodal Transfer Learning for Real-Time Fake News Detection on Social Media

Social media platforms have become crucial channels for the rapid dissemination of news in various formats, including text, images, audio, and video. Ensuring the authenticity of news content at primary stage is crucial to prevent the spread of false information. In order to gather semantic and contextual data for the identification of false news, current state-of-the-art majorly concentrated on text-based techniques, leveraging pre-trained word embedding and language models. However, these approaches suffer from the limitations viz. inefficiency in extracting context-based features, reliance on pre-trained models trained on more compact corpora, and static-masking utilization. In order to get over these issues, a new Framework for Transfer Learning based on Content for real-time False or Fake News Detection (MTL-rtFND) that integrates multimodal transfer learning methodologies has been proposed in this paper. It consists of a Data preprocessing block, Multimodal Feature Extraction Block (MFEB) and a Classification Block (CB). In the MFEB, a multimodal pre-trained mode, such as a fusion of visual and textual representations is leveraged to efficiently capture context-based features. This pre-trained model is trained on larger-scale multimodal datasets, enabling the extraction of richer contextual information. The resulting multimodal feature vectors from the MFEB are fed into the CB that employs a real time classification of news articles as fake or legitimate using deep neural network. The proposed MTL-rtFND model evaluation for real world datasets, demonstrating its effectiveness for real-time detection of fake news. The experimental results show significant improvements compared to state-of-the-art methods, obtaining an average gain in accuracy of 6.34% across multiple text embedding techniques. The findings demonstrate the capacity of multimodal transfer learning to improve fake news detection in real-time scenarios, thus contributing to mitigate the undesirable impact of misinformation on social media platforms.