MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202611089127 A) filed by Manipal University Jaipur on July 22, 2026, for A Hybrid Roberta Model For Emotion Detection Using Calibration And Uncertainty Estimation.
Inventors include Namita Narang; Riya Jain; and Nandini Babbar.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: The present invention relates to a hybrid RoBERTa model for emotion detection using calibration and uncertainty estimation that provides reliable and personalized emotional analysis for mental health monitoring and digital journaling applications. The system comprises a data preprocessing module for preparing textual inputs, a transformer-based emotion recognition engine for classifying emotions, a calibration module for improving prediction confidence using temperature scaling, and an uncertainty estimation module for identifying ambiguous emotional inputs. A hybrid fusion module selectively combines transformer-based predictions with lexical emotion analysis to improve recognition accuracy under uncertain conditions, while a large language model-based feedback module generates personalized emotional summaries, reflective insights, and context-aware recommendations. The system further incorporates continuous learning mechanisms to enhance prediction reliability and user interaction over time. By integrating contextual language understanding, confidence calibration, uncertainty estimation, hybrid emotion fusion, and personalized feedback generation within a unified framework, the invention provides an accurate, explainable, and adaptive emotion recognition solution suitable for mental health support, digital wellness platforms, and intelligent journaling systems.
Disclaimer: Curated by HT Syndication.
