MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641100479 A) filed by Dr. Divya S; Emali Dutta; Tathagata Chatterjee; V Meena; S Saranya; and Dr. B. Gunasundari on August 20, 2026, for Ai - Poe=wered Misinformation Detection And Trust Scoring System For Social Media.
Inventors include Dr. Divya S; Emali Dutta; Tathagata Chatterjee; V Meena; S Saranya; and Dr. B. Gunasundari.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: The rapid spread of misinformation on social media platforms has become a major challenge, influencing public opinion, elections, healthcare decisions, and disaster response. Existing detection systems primarily focus on binary classification of fake and genuine content but often fail to provide a reliable measure of information credibility. To address this limitation, this paper proposes an AI-Powered Misinformation Detection and Trust Scoring System for Social Media that integrates advanced Natural Language Processing (NLP), Graph Neural Networks (GNNs), and Explainable Artificial Intelligence ()CAI) for accurate misinformation detection and credibility assessment. Initially, social media posts are collected from platforms such as Twitter/X, Facebook, and Reddit, followed by text preprocessing involving tokenization, stopword removal, lemmatization, URL elimination, and emoji normalization. Semantic representations are extracted using the RoBERTa transformer model, while contextual propagation patterns among users are modeled through a Graph Attention Network (GAT) to capture information diffusion and source reliability. The semantic and graph-based features are fused using a Multi-Head Attention Fusion Network, and misinformation is classified using an XGBoost classifier. To enhance transparency, SHAP (SHapley Additive Explanations) is employed to highlight influential words and user interactions contributing to each prediction. Furthermore, a Trust Score ranging from 0 to 100 is computed by combining content authenticity, source credibility, propagation consistency, user reputation, and historical reliability through a weighted confidence aggregation model. The proposed system provides an effective, interpretable, and scalable solution for identifying misinformation while enabling users to assess the credibility of social media content in real time.
Disclaimer: Curated by HT Syndication.
