Harnessing text and image based multimodal techniques for enhanced fake news identification

Abstract

In the digital age, the proliferation of fake news poses a significant threat to the integrity of information, influencing public opinion and decision-making processes.The study is grounded in a comprehensive dataset, comprising textual and visual content related to the Ukraine-Russia conflict, which has been meticulously labeled for authenticity. The research employs sophisticated preprocessing techniques to standardize and augment the data, ensuring robustness and generalizability across diverse inputs. Through a series of experiments, the thesis demonstrates that multimodal models outperform traditional single-modality approaches, effectively capturing the nuanced interdependencies between text and images that are often exploited in the dissemination of fake news. This thesis explores the integration of Natural Language Processing (NLP) and computer vision techniques to enhance the detection of fake news through a multimodal approach that leverages both textual and visual data. By combining Long Short-Term Memory (LSTM) networks, Bidirectional Encoder Representations from Transformers (BERT), XGBoost and advanced image classifiers like ResNet-50, ResNet-101, Logistic Regression etc.This research aims to improve the accuracy of fake news identification across various languages and cultural contexts. The findings reveal that the combined use of text and image data significantly enhances the detection accuracy, particularly in complex scenarios where either modality alone might fail. Moreover, The contributions of this research extend beyond the specific case of fake news detection, offering insights into the broader application of multimodal techniques in various domains of machine learning and artificial intelligence.