MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641094259 A) filed by Karpagam College Of Engineering; Karpagam Academy Of Higher Education; Dr Senthil P; and Santhiya S on August 04, 2026, for Detection Of Abusive Content In Social Media.

Inventors include Dr Senthil P; and Santhiya S.

The application for the patent was published on August 07, 2026, under issue no. 32/2026.

Abstract: Social media is a big part of people’s lives today and is widely used to share thoughts, opinions, and information. However, many users also post abusive or harmful content predominantly targeting vulnerable users such as children and women. It is important to detect such messages early to keep online spaces safe and respectful. In recent years, automated content moderation systems and compliance frameworks are emerging, they often lack sensitivity to region-specific legal standards and cultural nuances. This work presents a novel method to fuse Indian legal norms for social media exploitation against women and children into prompt graphs to construct an organized prototype to guide AI and NLP models for jurisdiction conscious conduct. Prompt graph architecture aligns both with Indian legislative laws and World human rights frameworks to introduce a scalable, content-aware, and ethical AI model. This research compares advanced ML models like Bidirectional Encoder Representations from Transformers (BERT), Robustly Optimized BERT Pretraining Approach (RoBERTa), A Lite BERT (ALBERT), eXtreme Language Net (XLNet), and Distilled BERT (DistilBERT). A carefully labeled dataset containing 30,000+ social media messages, including posts with abusive, offensive, or harmful language, is used to train and evaluate the models. The performance of each model is measured using metrics such as accuracy, precision, recall, sensitivity, and F1- score. Among all the models, RoBERTa achieves the highest results, with 95% accuracy, 93% sensitivity, 95% precision, 93% recall, and a 94% F1- score. It performs especially well in identifying abusive content, especially messages targeting women and children, while keeping errors low. These findings suggest that RoBERTa is well-suited for building effective and responsible content moderation systems for social media.

Disclaimer: Curated by HT Syndication.