Jyothi Swaroopa Jinka supervised by Dr. Santosh Ravi Kiran  received her Master of Science in Computer Science and Engineering (CSE). Here’s a summary of her research work on Textual Attribute Recognition for Documents: 

This thesis addresses the challenge of Textual Attribute Recognition (TAR), which involves identifying formatting attributes such as bold, italic, underline, and strikeout that convey important semantic information in documents. Existing approaches either ignore contextual information or rely on computationally expensive full-document processing. To overcome these limitations, the thesis introduces TexTAR, a context-aware, multi-task Transformer framework that efficiently captures local document context using a novel Context Window-based approach. The model incorporates spatially aware positional encoding and jointly predicts multiple textual attributes, enabling robust performance across multilingual and layout-rich documents. The work also presents MMTAD, a large-scale multilingual and multi-domain dataset containing over one million word-level annotations. Extensive experiments demonstrate that TexTAR significantly outperforms existing methods while remaining computationally efficient, advancing document understanding beyond traditional OCR toward richer, semantically aware document intelligence systems.

May 2026