Detecting Disinformation Using BERT-BiLSTMand ResNet-Based Deep Fusion Networks

Read full text Summarise with VIRAT
Authors: Saranya N., Karmukilan B., Kishore P., Mohammed Safik M., Nithan Anto J.
Cited by: 0
References: 3

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Ever since the arrival of the digital world, misleading information through the use of manipulated pictures and poorly constructed text on news websites and social media has become a world wide issue. Detection of such manipulated or false information is required in order to protect the public opinion, state security, as well as trust to digital environments, so in order to detect misinformation using a single input, which is a visual information that also has hidden text (memes, posters, and screenshots), this work suggests a deep learning dual-stream fusion model. The system recognizes the text on the image using optical character recognition (OCR) and derives the semantic meaning with the help of a BERT-BiLSTM model. Highlevel features of the image are simultaneously extracted by feeding it through a ResNet convolutional neural network. After that, the two features (text and visual) are fused on a deep fusion layer and maintain context and multi-modal dependency and generate a more robust and accurate classification. Such an integrated solution provides the final classifier that can distinguish the authenticity of supplied image-text input and prevent the further propagation of the misinformation introduced by multimedia sources and can greatly enhance the correctness of fake news recognition based on either Natural Language Understanding or Visual Semantics.

Fields of Study & Subjects

Misinformation and Its Impacts Digital Media Forensic Detection Multimodal Machine Learning Applications

Publication details

Journal / source
International Journal of Innovative Science and Research Technology (IJISRT)
Pages
2875–2875
Publication date
2 April 2026
Language
EN