MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202641072286 A) filed by Vellore Institute Of Technology on June 10, 2026, for A System And Method For Federated Deep Learning-Based Fake News Detection Using Hybrid Character-Level Feature Extraction And Attention Mechanisms.
Inventors include Dr. Dhivyaa C R; and Dr. Nithya K.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: A System and Method for Federated Deep Learning-Based Fake News Detection Using Hybrid Character-Level Feature Extraction and Attention Mechanisms. The invention provides a federated deep learning framework comprising a plurality of client nodes and a central server configured to collaboratively train a classification model without sharing raw data. Each client node processes textual data through a pipeline including preprocessing, character-level embedding, convolutional feature extraction, sequential modeling using a stacked bidirectional long short-term memory network, attention-based feature weighting, and cost-sensitive classification. The character-level embedding captures fine-grained textual variations including misspellings and morphological patterns, while the attention mechanism enhances feature selection for improved classification accuracy. The central server aggregates model updates received from the client nodes and generates an updated global model for iterative training. The framework enables privacy-preserving, scalable, and robust detection of fake news across distributed and heterogeneous datasets while addressing class imbalance and improving model generalization. (FIG. 1)
Disclaimer: Curated by HT Syndication.
