MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641113237 A) filed by Seshadri Rao Gudlavalleru Engineering College; A. Koteswaramma; V. Dadi Naga Siva Sai Pavan; N. Teja; Y. Nithya Sri; and T. P. Santhosh on September 22, 2026, for A Confidence-Based Two-Tier Hate Speech Detection System Using Svm And Bert With Confidence-Aware Routing For Real-Time Social Media Content Moderation.
Inventors include Seshadri Rao Gudlavalleru Engineering College; A. Koteswaramma; V. Dadi Naga Siva Sai Pavan; N. Teja; Y. Nithya Sri; and T. P. Santhosh.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: Automatic identification of harmful content on social platforms requires systems that maintain classification precision without sacrificing processing speed. In real- world moderation pipelines, this balance is critical because large-scale user-generated content must be filtered in near real time while preserving contextual understanding. Deep learning architectures such as BERT deliver superior accuracy but demand extensive computing resources, whereas conventional algorithms like SVMs execute rapidly yet fail to capture semantic context adequately. The present invention introduces a dual-stage classification framework employing confidence- based decision routing to optimize the trade-off between computational efficiency and detection performance. The methodology begins with an SVM Tier 1 Classifier that processes vectorized TF-IDF representations of clean preprocessed text for initial rapid assessment, then forwards uncertain low-confidence predictions to a fine- tuned BERT Tier 2 Classifier for comprehensive semantic evaluation, while routing high-confidence predictions directly to a Result Store. Both classification paths converge at a Router Combiner module that consolidates predictions for final Output Prediction. By restricting intensive transformer processing to ambiguous cases only, this strategy minimizes unnecessary computational expense. Evaluation conducted on the Davidson et al. benchmark dataset containing 24,783 annotated tweets across three categories—hate speech, offensive-but-not-hateful language, and neutral messages—demonstrates that the proposed framework achieves 93.2% classification accuracy with 30 ms mean inference time, approximating BERT-only performance while operating approximately five times faster than standalone transformer implementations. The invention demonstrates that hybrid systems with adaptive confidence-aware routing offer a practical and scalable solution for deploying sophisticated hate speech detection systems under real-world operational constraints.
Disclaimer: Curated by HT Syndication.
