Development of a knowledge extraction system from ancient Isan medical texts using Bi-LSTM-based named entity recognition
Article Sidebar
Main Article Content
Abstract
Ancient Isan medicinal manuscripts encapsulate traditional medical knowledge that reflects the indigenous wisdom of the Isan people in healthcare practices of the past. This body of knowledge continues to be preserved and applied in contemporary contexts, particularly through the effective and safe integration of various Thai herbs with modern medicine. Furthermore, these herbs hold significant potential for development into value-added products within the industrial sector. Nonetheless, the interpretation of ancient Isan medicinal manuscripts remains challenging due to their distinctive linguistic features and domain-specific terminologies, especially for individuals lacking foundational knowledge of the language and subject matter. This research aims to develop an automated knowledge extraction system from ancient Isan medicinal manuscripts, focusing on the identification of medicine names, disease names, and herbal names. The research methodology includes data collection and the training of a named entity recognition model using the Bi-LSTMs technique. The knowledge extraction system is implemented using the Python programming language, and the extracted knowledge is stored in an SQLiteStudio database. The system's performance is evaluated based on the model's precision using F1-scores, both per class and as a weighted average across all classes. User satisfaction is also assessed regarding the system and database design and development. The evaluation group consists of 12 participants, including experts in computer program design and development, specialists in ancient languages, and general users. The model’s performance was evaluated using the weighted average F1-score, which reached 94%. The class-wise F1-scores were 96% for herbal names, 82% for symptom names, and 7% for medicine names. User satisfaction with the system was also assessed. The results indicated that the system design and development were rated at the highest level (x̄ = 4.44, SD = 0.57), while the database design was rated at a high level (x̄ = 4.44, SD = 0.57). These findings demonstrate the potential of applying named entity recognition techniques to effectively develop an automated knowledge extraction system for ancient Isan medicinal texts.
Article Details
References
จินตนา พลศรี, นิศาชล จำนงศรี, และจีรยุทธ ไชยจารุวณิช. (2565). การพัฒนาคลังศัพท์สำหรับการแปลงคำปริวรรตของอักษรไทยน้อยเป็นคำอ่านภาษาไทยถิ่นอีสานด้วยคอมพิวเตอร์ [Development of a lexicon for computerized transliteration of Thai Noi script into Isan Thai dialect readings]. วารสารวิทยาศาสตร์ประยุกต์, 22(1), 1–14.
จินตนา พลศรี. (2564). กรอบการทำงานในการแปลงคำปริวรรตของอักษรไทยน้อยเป็นคำอ่านภาษาไทยถิ่นอีสาน [A framework for transliterating Thai Noi script into Isan Thai dialect readings] [Doctoral dissertation, Suranaree University of Technology].
สำนักงานคณะกรรมการพัฒนาการเศรษฐกิจและสังคมแห่งชาติ. (2555). แผนพัฒนาเศรษฐกิจและสังคมแห่งชาติ ฉบับที่ 11 พ.ศ. 2555–2559 [The 11th National Economic and Social Development Plan, 2012–2016]. https://www.nesdb.go.th/download/article/article_20160323112431.pdf
สำนักเลขาธิการนายกรัฐมนตรี ทำเนียบรัฐบาล. (2567). "เดชอิศม์" หนุนยกระดับสมุนไพรไทย นวดไทย ขับเคลื่อนเศรษฐกิจประเทศ [“Dechis” supports elevating Thai herbal medicine and Thai massage to drive the national economy]. https://www.thaigov.go.th/news/contents/details/89674
Chalamkate, T., Thinnachote, C., & Thamrongrattanarit, A. (2022). The development of geo-names extraction from Twitter texts data by conditional random fields. Journal of Engineering and Digital Technology (JEDT), 10(2), 1–10.
Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
Kannikaklang, N., Thamviset, W., & Wongthanavasu, S. (2024). BiLSTCAN: A novel SRS-based bidirectional long short-term capsule attention network for dynamic user preference and next-item recommendation. IEEE Access, 12, 6879–6899. https://doi.org/10.1109/ACCESS.2024.3351283
Kuran, E. C., Elen, A., & Kuran, U. (2024). A hybrid BiLSTM-CRF model with BERT embeddings for named entity recognition on confidential synthetic data. Proceedings of the V. International Science and Innovation Congress (INSI). https://www.researchgate.net/publication/385101301
Li, J., Sun, A., Han, J., & Li, C. (2020). A survey on deep learning for named entity recognition. IEEE Transactions on Knowledge and Data Engineering, 34(1), 50–70. https://doi.org/10.1109/TKDE.2020.2981314
Likert, R. (1961). New patterns of management. McGraw-Hill.
Makarenkov, V., Rokach, L., & Shapira, B. (2019). Choosing the right word: Using bidirectional LSTM tagger for writing support systems. Engineering Applications of Artificial Intelligence, 84, 1–10.
Mansouri, A., Affendey, L. S., & Mamat, A. (2008). Named entity recognition approaches. International Journal of Computer Science and Network Security, 8(2), 339–344.
Polsri, J. (2024). CRF-named entity recognition model for ancient Isan medicine texts. In Proceedings of the 2024 12th International Electrical Engineering Congress (iEECON) (pp. 1–5). IEEE.
Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11), 2673–2681. https://doi.org/10.1109/78.650093
Werbos, P. J. (1990). Backpropagation through time: What it does and how to do it. Proceedings of the IEEE, 78(10), 1550–1560. https://doi.org/10.1109/5.58337
Yadav, V., & Bethard, S. (2018). A survey on recent advances in named entity recognition from deep learning models. In Proceedings of the 27th International Conference on Computational Linguistics (pp. 2145–2158). Association for Computational Linguistics.
Zhang, Y., & Wang, L. (2023). Named entity recognition in electronic health records: A hybrid approach. National Center for Biotechnology Information. https://pmc.ncbi.nlm.nih.gov/articles/PMC10651400/