Indian Language Ontology Extractor A Domain-Specific Implementation
Year: 2025
Type: article
Abstract: The exponential growth of digitized Hindi content across news portals, government platforms, and social media presents significant challenges for text mining. Effective stop-word removal is essential for improving processing efficiency. This study introduces a hybrid statistical and knowledge-based approach to generate domain-specific Hindi stop-word lists, leveraging multiple ranking methods normalized through a social choice-based voting mechanism. The lists are evaluated against text mining algorithms, revealing that conventional attribute extraction techniques often fail to identify potential stop words accurately. To address structural complexity in ontology extraction, we propose an unsupervised k-partite graph learning algorithm for transforming unstructured text into structured knowledge. The method partitions terms by information content and constructs ontologies by identifying subsumption relationships across partitions, reducing computational overhead and enhancing term refinement. Experimental validation on Hindi and English health-sector corpora demonstrates substantial precision gains compared to manually curated gold standards. Notably, the framework operates without advanced NLP tools such as parsers or Named Entity Recognition, ensuring portability across resource-limited languages.
Author: Viranchee V. Dave
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7125219794