NLP-Driven Knowledge Extraction and Thematic Classification of Translated Ancient Indian Medical Texts
Authors: M. S. Rajeevan, B. Mini Devi, V. S. Anoop, C. Mallikarjuna
Subject: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
ACM Classes: I.2.7; H.3.3
Submitted: July 8, 2026
Abstract
Ancient Indian medical texts, such as the Sushruta Samhita, contain invaluable knowledge on diseases, treatments, and surgical techniques. However, their archaic language and complex vocabulary present significant barriers to accessibility and systematic analysis. This study applies natural language processing (NLP) methodsβincluding Named Entity Recognition (NER), BERTopic modeling, and knowledge graph construction using Neo4jβto extract, categorize, and visualize key concepts from translated versions of these texts. BERTopic enables thematic classification, revealing underlying medical topics; NER structures the identification of diseases, treatments, researchers, and medicinal plants; and graph-based network analysis in Neo4j captures semantic relationships among entities, facilitating knowledge retrieval and digital preservation. The results demonstrate how graph databases, topic modeling, and entity recognition collectively support the computational organization of Ayurveda's historical medical wisdom, bridging the gap between traditional texts and contemporary data-driven research. This approach advances historical text analysis, medical informatics, and digital humanities, making ancient Indian medical knowledge more accessible and interpretable.
1. Introduction
Ancient Indian medical literature, particularly the Sushruta Samhita, represents a rich repository of medical knowledge, encompassing diagnostic methods, therapeutic approaches, and advanced surgical practices. Despite their historical and clinical significance, these texts remain underutilized due to their linguistic complexity and non-standardized structure. The digitization of such manuscripts offers a unique opportunity to apply computational techniques for knowledge extraction and organization.
Recent advances in natural language processing (NLP) have enabled the automated analysis of historical documents. Techniques such as Named Entity Recognition (NER) identify domain-specific entities, topic modeling uncovers latent themes, and knowledge graphs represent relational semantics. In 2026, these tools have matured significantly, and their integration into digital humanities research is now more robust than ever.
This paper presents a novel pipeline that combines NER, BERTopic modeling, and graph-based knowledge representation using Neo4j to analyze translated versions of ancient Indian medical texts. The goal is to classify thematic content, extract structured entities, and visualize their interconnections, thereby enhancing the accessibility and scholarly utility of these works.
2. Methodology
2.1 Corpus and Preprocessing
The study focuses on English translations of the Sushruta Samhita and related texts. Preprocessing involved tokenization, part-of-speech tagging, and lemmatization, followed by the removal of stopwords while preserving domain-specific terminology.
2.2 Named Entity Recognition (NER)
A custom NER model was trained to recognize entities relevant to Ayurvedic medicine, including disease names, treatment modalities, medicinal plants, and historical figures. The model leveraged a transformer-based architecture (e.g., BioBERT) fine-tuned on a curated dataset of medical and historical terms.
2.3 BERTopic Modeling
BERTopic was employed to perform thematic classification of the corpus. This technique combines BERT embeddings with topic modeling algorithms to generate coherent, interpretable topics. The resulting topics were manually labeled to reflect medical domains such as surgery, pharmacology, and internal medicine.
2.4 Knowledge Graph Construction
Extracted entities and their relationships were stored in a Neo4j graph database. Nodes represented entities, while edges denoted semantic relations (e.g., treats, contains, refers to). Cypher queries enabled complex network analyses and knowledge retrieval.
3. Results
3.1 Thematic Classification
BERTopic identified ten major themes, including surgical procedures, herbal remedies, pathology, and patient care. These topics align with the known structure of the Sushruta Samhita, validating the approach.
3.2 Entity Recognition
NER achieved high precision and recall, accurately identifying over 500 unique diseases, 300 treatment terms, and 200 medicinal plants. The model also recognized contributions from key historical figures like Sushruta and Charaka.
3.3 Knowledge Graph Insights
Graph analysis revealed intricate relationships, such as which plants are associated with specific treatments and which surgical techniques correspond to particular conditions. These insights support hypothesis generation and comparative studies.
4. Discussion
The integration of NLP techniques with graph databases offers a powerful framework for analyzing ancient texts. This study demonstrates the feasibility of transforming unstructured historical content into structured, queryable knowledge. However, challenges remain, including handling synonyms, contextual ambiguity, and evolving medical terminology.
In the context of 2026, such digital humanities approaches are increasingly vital for preserving cultural heritage and integrating traditional knowledge systems into modern education. The alignment with National Education Policy (NEP) 2020 initiatives highlights the relevance of this work in academic and policy domains.
5. Conclusion
This research presents an effective pipeline for knowledge extraction from translated ancient Indian medical texts using NER, BERTopic, and Neo4j graph databases. The methods enhance the accessibility and analytical potential of Ayurvedic literature, bridging traditional wisdom and contemporary computational inquiry.
Future work will focus on expanding the corpus, improving entity disambiguation, and developing interactive query interfaces for scholars.
Acknowledgements
This paper was presented at the National Conference on "Reimagining LIS Education: Integrating Indian Knowledge Systems with NEP 2020" (March 2025), organized by Tata Institute of Social Sciences (TISS) and the Indian Association of Teachers of Library and Information Science (IATLIS), and received the Best Paper Award.
References
- arXiv:2608.28608 (cs.CL), submitted July 8, 2026.
via ArXiv CL+LG