Applying S.R. Ranganathan's Classification Theory to Investigate the Epistemology of Knowledge Organization in Large Language Models (LLMs)

Abstract

It has been explored that how S.R. Ranganathan's faceted classification (Colon Classification) and epistemological ideas can inform the design and validation of Large Language Models (LLMs) . Ranganathan's five fundamental categories - Personality, Matter, Energy, Space, Time (PMEST)—and his three planes of work (Idea, Verbal, Notational) offer a framework for organising knowledge that can anchor and verify AI outputs. A conceptual model in which every LLM-generated claim is mapped onto a structured faceted scheme or knowledge graph (KG) via a Faceted Classification Module (FCM) and a Classification Validator is proposed. This validator tags each claim with PMEST facets, checks semantic and epistemic consistency using modern NLP techniques (detailed in Appendix A), and rates certainty using a seven-level epistemic hierarchy (detailed in Appendix B). Using the query "Uses of turmeric in Indian medicine" as a running example, it is demonstrated how Ranganathan's planes of work apply to LLM interactions and illustrate each pipeline step with technical specifications. This approach has been compared with the existing KG-augmented RAG methods, particularly Microsoft's GraphRAG. It has been further discussed how classification-based systems complement graph-based retrieval for improved fact-checking and cultural context preservation. Recent empirical findings show that advanced LLMs suffer high hallucination rates (GPT-4: ~29% hallucinations, ~13% precision on scientific reference tasks) and that knowledge graph-based approaches yield substantial improvements in complex question-answering performance. Overall, it is concluded that LLM knowledge organisation is a bibliographic process grounded in library epistemology, aiming to detect hallucinations, enhance explainability, and integrate diverse knowledge systems—including Indian Knowledge Systems—systematically.