Enhancing Social Media Fake News Detection Using PsychoLinguistic Multiplicative Attentive Net (PLiMA Net) with LIWC Features

Detecting fake news on social media platforms remains a critical challenge due to the rapid dissemination of information and the intricate nature of multimedia content. Traditional approaches inadequately capture the psycholinguistic features essential for understanding the social context, intent, and emotional undertones embedded in the text. To address these shortcomings, this study introduces a novel PsychoLinguistic Multiplicative Attentive Network (PLiMA Net) that incorporates psycholinguistic analysis and Luong attention mechanism into RNN for fake news detection systems, aiming to enhance the accuracy and reliability of identifying misleading content. Previous contributions: (a) The first work focused on text-based fake news detection through the "Integrated Hierarchical Granular Retentive XLNet with Fast Embedded Semantic Extraction" method. This approach utilized innovative components such as Structural Transition Enhanced Parsing (STEP), Multi-Granular Entity Resolution (MulGER), and Fast-X-Ref Semantic Feature Extraction, achieving a remarkable accuracy of 97.67%. (b) The second work extended this framework by incorporating emojis into sentiment analysis where the "Multi-Token Concatenated Embedding and SemantiXpert Probabilistic Classifier" method was proposed, featuring Multi-Run Byte Pair Encoding (MRBPE) and a Concat-ViLT model for merging textual and visual representations and achieved an accuracy of 95.83%. By incorporating psycholinguistic analysis with Luong attention mechanism into those previous two works, this PLiMA Net model improves its precision and adaptability to the evolving linguistic area of social media, ensuring robust and trustworthy detection of fake news.