MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202641087402 A) filed by Vallurupalli Nageswara Rao Vignana Jyothi Institute Of Engineering And Technology on July 16, 2026, for Ai-Driven Framework For Automated Mental Health Detection From Unstructureddata.
Inventors include Mr. Karnam Akhil; Dr. N. Venkata Sailaja; Dr. Ch. Suresh; N Koushik; Reshma Chowdary; Mohammed Talhah; and R Venkata Giridhar.
The application for the patent was published on July 24, 2026, under issue no. 30/2026.
Abstract: The increasing volume of unstructured text data poses significant challenges in domains requiring precise classification and timely analysis particularly in the social media and sectors. In there are delays due to extensive case backlogs which demand efficient automated document classification to support professionals. In social media especially for early detection of depression, mental health conventional methods often fall short in processing unstructured data from social platforms such as Reddit and Facebook. This investigates domain-specific text classification across three domains-datasets: the Reddit depression data, Facebook teenage mental health dataset and the documents. Each domain presents unique challenges related to text length, vocabulary complexity, and class imbalance. We employ specialized transformer-based architectures tailored to domain characteristics and enhance them with cross-attention layers to enrich contextual understanding. Explainable AI (XAI) techniques are integrated to improve model transparency, with visual attention maps highlighting the most influential features in classification. Our results demonstrate that the Reddit dataset achieves the highest classification accuracy at 98%, followed by the Facebook dataset at 96.9%, while texts show comparatively lower performance due to their structural and linguistic complexity. This work underscores the importance of domain adaptation, interpretability, and feature-aware modelling in improving classification performance across diverse unstructured text domains.
Disclaimer: Curated by HT Syndication.
