A Deep Learning Methodology-Based Unsupervised Misinformation Detection (Umd) Technique for Identifying Fabricated Narratives

The widespread dissemination of misinformation, commonly referred to as “fake news,” through social media platforms has had profound negative effects on individuals’ daily lives and has even contributed to instances of civil unrest. In response to this pressing issue, this paper introduces an innovative approach called Unsupervised Misinformation (Fake-News) Detection Technique (UMD). This method combines user data, textual content, news dissemination patterns, and images using Bidirectional GRU (Bi-GRU) and Self-Attention layers within an autoencoder framework to extract latent features and interrelations among them. Furthermore, the approach utilizes residual reconstruction to identify instances of fake news. Empirical evaluations conducted on two real-world datasets employing four different methodologies demonstrate promising outcomes. The term “misinformation” encompasses fabricated narratives designed to deceive or mislead individuals. This study presents a Deep Learning-based strategy to combat the proliferation of fake news. Crafting fake news poses a considerable challenge, demanding models to possess a nuanced understanding of Natural Language intricacies. Leveraging technologies such as Artificial Intelligence (AI) and Natural Language Processing (NLP) tools holds great promise for researchers seeking automated solutions for identifying false information. However, detecting fake news remains a formidable task, as it necessitates models to contextualize news content and discern its authenticity relative to genuine news. Moreover, the subjective nature of comparing proposed news with original sources adds complexity to the classification process, introducing elements of subjectivity and bias. The overall accuracy of this model is $\mathbf{9 2. 8 2 \%}$ which is approximately 3% better when compared to the existing models like transformer model and hybrid CNN-RNN model. The proposed model is proven to be outstanding in terms of accuracy, precison and F1-score.