Preserving cultural heritage through AI: A novel framework integrating LangChain and ChatGPT for analyzing and digitalizing the Javanese Pawukon calendar system

Hastuti, Khafiizh and Hidayat, Erwin Yudi and Alzami, Farrikh and Risqa, Ifan and Ahmad, Sabrina and Fahmi, Hanif (2026) Preserving cultural heritage through AI: A novel framework integrating LangChain and ChatGPT for analyzing and digitalizing the Javanese Pawukon calendar system. Expert Systems with Applications, 331 (133228). pp. 1-23. ISSN 0957-4174

Text 00402160620261028223297.pdf Download (9MB)

Abstract

The preservation and analysis of cultural heritage documents present complex methodological challenges, particularly in maintaining linguistic authenticity while enabling computational accessibility. This research develops and validates an integrated framework for analyzing the Javanese Pawukon calendar system, implementing a novel approach that combines LangChain and ChatGPT with specialized cultural preservation techniques. The methodology processed 621 pages of Javanese-language Pawukon manuscripts through a sophisticated pipeline incorporating context-aware OCR, semantic chunking with 1000-character segmentation, and OpenAI’s text-embedding-ada-002 model optimized for cultural entity preservation. The framework introduced three key technical innovations: (1) direct semantic processing of low-resource language content without traditional linguistic preprocessing, (2) a hierarchical vector store architecture using PGVector for cultural entity indexing, and (3) prompt engineering methodologies optimized for traditional knowledge systems. Experimental validation demonstrated robust performance in preserving cultural semantics, achieving a BERTScore F1 of 86.46% for semantic preservation and a ROUGEScore F1 of 43.24% for lexical accuracy, with statistical significance confirmed through ANOVA (F = 5.2613, p  <  0.001). Human expert validation by three independent cultural validators from recognized Javanese cultural institutions yielded an overall score of 4.925 out of 5.000, with all ten generated responses accepted without revision, confirming that the BERTScore–ROUGE-L divergence reflects deliberate paraphrasing rather than cultural distortion. The framework’s successful implementation not only advances the field of digital cultural preservation but also provides a replicable methodology for similar preservation initiatives, particularly in processing complex traditional knowledge systems documented in low-resource languages. This research contributes both theoretical insights and practical methodologies for cultural heritage preservation, demonstrating how modern AI technologies can effectively preserve and make accessible traditional knowledge systems while maintaining their cultural and linguistic integrity.

Item Type:	Article
Uncontrolled Keywords:	LangChain, ChatGPT, Pawukon, Semantic modeling, Cultural heritage, Knowledge preservation, Low-resource languages
Divisions:	Faculty of Information and Communication Technology
Depositing User:	Norfaradilla Idayu Ab. Ghafar
Date Deposited:	04 Sep 2026 01:15
Last Modified:	04 Sep 2026 01:15
URI:	http://eprints.utem.edu.my/id/eprint/30387
Statistic Details:	View Download Statistic

Actions (login required)

View Item