Source-linked AI summary
NLP-Driven Knowledge Extraction and Thematic Classification of Translated Ancient Indian Medical Texts
M. S. Rajeevan, B. Mini Devi, V. S. Anoop, C. Mallikarjuna
TL;DR
Ancient Indian medical texts are difficult to access and organize because of complex language, transliteration, and structure. The paper applies NER, BERTopic, and Neo4j knowledge graphs to translated Sushruta Samhita text, producing structured entities, thematic groupings, and queryable relationships. It concludes that this combination supports systematic, scalable preservation and computational analysis of Ayurvedic knowledge.
Problem
Complex language, transliteration, and organization make ancient Indian medical knowledge difficult to access and systematically arrange.
Method
The study applies NER, BERTopic, and Neo4j knowledge graphs to extract entities, discover themes, and represent relationships in translated Sushruta Samhita text.
Results
The methods produced structured Ayurvedic entities, interpretable thematic groupings, and a graph-based representation supporting semantic querying and inferencing.
Takeaways & Limitations
NLP-driven classification and knowledge graphs provide a systematic and scalable solution for organizing, preserving, and analyzing intricate historical medical knowledge.
Takeaways & Limitations
Transliteration ambiguity and Sanskrit-derived terminology limited conventional NER performance, requiring custom preprocessing, rule-based fine-tuning, and domain-specific models.
Abstract
from arXiv · showhide
Ancient Indian medical texts like Sushruta Samhita have extensive information on diseases, treatments, and surgical techniques. Yet, their ancient format and use of intricate vocabulary pose difficulties in accessibility and systematic ordering. The research here utilizes Natural Language Processing (NLP) methods like Named Entity Recognition (NER), BERTopic modeling, and Knowledge Graph development in Neo4j to extract, categorize, and visualize important concepts based on translated versions. Thematic classification with BERTopic allows for the identification of the underlying medical topics, whereas NER supports the structured entity recognition of diseases, treatments, researchers, and medicinal plants. Graphbased network analysis with Neo4j also allows for the semantic representation of relationship among extracted entities, supporting knowledge retrieval and digital preservation. The findings illustrate how graph databases, topic modeling, and entity recognition facilitate the computational organization of Ayurveda's historical medical wisdom, closing the gap between the conventional texts and contemporary data-driven inquiry. The suggested method promotes historical text analysis, medical informatics, and digital humanities to make ancient Indian medical wisdom more accessible and understandable.
1. Introduction
Ancient Indian medical texts contain valuable Ayurvedic knowledge but remain difficult to access because of complex language, organization, and multilingual terminology. This study combines NLP methods to extract, classify, visualize, and preserve that knowledge computationally.
- Sushruta Samhita contains detailed information on surgery, pharmacology, and disease treatment, but its complex language and organization hinder access.
- NER, classification, and knowledge graphs can identify diseases, treatments, and practitioners in historical medical documents.
- The study applies NER, BERTopic, and Neo4j knowledge graphs to interpret Ayurvedic knowledge structurally and computationally.
- NER recognizes diseases, treatments, medicinal plants, and scholars, while BERTopic discovers medical themes and Neo4j captures entity relationships.
- Combining NER, topic modeling, and knowledge graphs provides a scalable approach for analyzing historical medical literature and supporting digital preservation.
2. Related Studies
Related research shows that NLP, topic modeling, and knowledge graphs are increasingly used to digitize, structure, and retrieve information from historical and Ayurvedic texts. These studies motivate computational approaches that improve access to traditional medical knowledge.
- NLP and knowledge graphs help extract, classify, and organize information from otherwise unstructured Sanskrit and Ayurvedic manuscripts.
- Meltemi 7B uses a large Greek text corpus to support comprehension, segmentation, and entity recognition in historical documents.
- Dynamic topic modeling over BERT embeddings has been applied to Roman literature to obtain qualitative insights from historical corpora.
- Prior Sanskrit knowledge-graph frameworks support query-based retrieval, while Āyurjñānam constructed an Ayurvedic graph through manual drug annotation.
- The Āyurjñānam program combined knowledge graphs with interfaces for querying and annotation, and TKDL uses standardized traditional-medical terminologies.
- Existing studies establish semantic annotation, machine learning, and knowledge models as empirical tools for extracting and preserving information from ancient manuscripts.
3. Materials and Methods
The methodology extracts and categorizes entities from a translated Sushruta Samhita, discovers themes with BERTopic, and represents relationships in a Neo4j knowledge graph. The workflow combines preprocessing, entity recognition, dimensionality reduction, clustering, keyword extraction, and graph analysis.
- 3. Materials and Methods: The study uses NER, BERTopic, and Neo4j visualization to categorize entities and their interconnectivity in ancient Indian literature.
- 3.1 Data Processing: The translated Sushruta Samhita is preprocessed with Python, spaCy, ScispaCy, and NLTK before entities are categorized for structured analysis.
- 3.1 Data Processing: The workflow proceeds from data extraction through topic modeling and entity classification to knowledge-graph construction.
- 3.2 Named Entity Recognition (NER): NER identifies and classifies diseases, treatments, scholars, medicinal plants, and locations from text.
- 3.2 Named Entity Recognition (NER): NER frames each word sequence as labels assigned to entity types such as Disease, Treatment, and Scholar.
- 3.2 Named Entity Recognition (NER): The NER methods discussed include CRF and BiLSTM-CRF models, with CRF using learned feature weights for possible label sequences.
- 3.2 Named Entity Recognition (NER): ScispaCy identifies medical conditions, treatments, and plants, while custom lexicons address spelling variations and Sanskrit transliterations.
- 3.3 BERTopic Modeling for Thematic Analysis: BERTopic discovers topics using transformer-based embeddings and clustering models that group semantically related words.
4. Results
NER organized Ayurvedic entities into medical, historical, geographical, and textual categories, while BERTopic identified major themes and visualizations represented their semantic structure. A Neo4j knowledge graph then connected diseases, treatments, plants, scholars, texts, and places.
- Named Entity Recognition: NER classified diseases, treatments, scholars, medicinal plants, and places/texts into distinct thematic divisions.Examples included Kushtha, Prameha, and Jwara among diseases, and Susruta, Charaka, and Vagbhata among scholars.
- Named Entity Recognition: 565 NORP, 242 PRODUCT, 174 DATE, 30 TIME, 48 ORDINAL, 79 GPE, 64 LOC, and 58 WORK_OF_ART entities were identified.These categories represented traditions, medical products, chronology, geography, and medical texts.
- BERTopic Modeling: BERTopic identified themes spanning medical cures, diseases, scholarly works, medicinal plants, and historical or geographical sources.High-probability terms linked Vasti, Sneha, Triphala, and Ghrita with Medical Treatments, while Kushtha and Prameha belonged to Diseases & Disorders.
- BERTopic Modeling: Medical Treatments and Diseases & Disorders were the most pervasive topics, with surgical methods and herbal medicine also prominent.The topic visualizations represented dominant terms, topic prevalence, and semantic distributions across the text.
- BERTopic Modeling: The intertopic distance map showed overlapping medical and procedural themes alongside distinct groups for diseases, surgery, and herbal medicines.In the 3D UMAP projection, nearby documents represented greater thematic similarity and distant documents represented more divergent topics.
5. Recommendations.
The recommendations identify domain-specific modeling, ontology-informed topic classification, stronger graph reasoning, and standardized interdisciplinary representations as priorities for improving Ayurvedic knowledge extraction and preservation.
- General-purpose NER models misclassify Sanskrit-derived terminology and Ayurvedic jargon, motivating personalized entity recognition and transliteration methods.
- Ontology-based and rule-based classification could make BERTopic outputs more readable and reduce uncertainty in theme identification.
- Adding causal correlations, semantic reasoning, and advanced queries would make the Neo4j knowledge graph more practically effective for exploring Ayurvedic relationships.
- A standardized Ayurvedic knowledge model and interdisciplinary collaboration could support integration with historical, linguistic, and biomedical research.
6. Discussion and Conclusion
The study applies NLP, BERTopic, and Neo4j knowledge graphs to structure the English translation of the Sushruta Samhita. It extracts entities, identifies thematic structures, and organizes relations, while highlighting transliteration and domain-model limitations and future research needs.
- NER, BERTopic, and knowledge graphs extract structured entities, reveal thematic structures, and organize relationships in the translated Sushruta Samhita.
- Transliteration ambiguity, inconsistent ancient text, weak conventional NER performance, and topic-label tuning require domain-specific models and preprocessing.
- The Neo4j knowledge graph connects diseases, treatments, medicinal plants, and scholars into a queryable and extensible representation.
- Integration with contemporary medical ontologies could support global incorporation of Ayurveda into contemporary healthcare systems.
- The combined approach provides a systematic and scalable solution for preserving and computationally analyzing ancient Indian medical literature.