Construction of Korean Medicine Knowledge Graph and Development of Prescription Recommendation System based on RippleNet

Sang-Kyun Kim; DongHyoung Lee; Anna Kim; SeJin Nam; Kim, Sang-Kyun; Lee, DongHyoung; Kim, Anna; Nam, SeJin

doi:10.13048/jkm.25017

JKM > Volume 46(2); 2025 > Article

Kim, Lee, Kim, and Nam: Construction of Korean Medicine Knowledge Graph and Development of Prescription Recommendation System based on RippleNet

Original Article

The Journal of Korean Medicine 2025; 46(2): 51-62.

Published online: June 1, 2025

DOI: https://doi.org/10.13048/jkm.25017

Construction of Korean Medicine Knowledge Graph and Development of Prescription Recommendation System based on RippleNet

Sang-Kyun Kim¹

, DongHyoung Lee²

, Anna Kim¹

, SeJin Nam²

¹KM Data Division, Korea Institute of Oriental Medicine

²Data Convergence Team, Seoul National University Bundang Hospital

Correspondence to: Sang-Kyun Kim, 1672, Yuseong-daero, Yuseong-gu, Daejeon, South Korea, Tel: +82-42-868-9526, E-mail: skkim@kiom.re.kr

Received March 26, 2025 Revised April 16, 2025 Accepted May 22, 2025

Abstract

Objectives

We aim to propose an artificial intelligence system for recommending Korean medicine prescriptions based on the RippleNet algorithm using a knowledge graph.

Methods

The Korean medicine knowledge graph was constructed with information on prescriptions, medicinal materials, chemical compounds, gene targets, and modern diseases from the TM-MC database. The RippleNet algorithm was trained with the knowledge graph and the prescription history of diseases from TM-MC. The optimized hyperparameter values of the algorithm were derived using the grid search method. The recommendation system was implemented with the knowledge graph and the parameterized model.

Results

The Korean medicine knowledge graph contains 16,977 nodes and 142,924 edges, and the prescription histories of diseases include 2,101 true values and 3,413 false data points. The recommendation algorithm showed the best performance with an AUC of 0.919 when trained with parameters of 4 hops and a ripple set size of 32.

Conclusions

In this study, we implemented a system that recommends prescriptions based on artificial intelligence algorithms using the knowledge graph. In the future, we plan to improve the quality of the knowledge graph and explore a variety of modern AI algorithms to enhance recommendations.

Keywords: Recommendation system, Prescription, Knowledge graph, RippleNet, Korean medicine

Fig. 1

Progress of the preference propagation in knowledge graph

Fig. 2

Example of Mawhang-tang in Korean medicine knowledge graph

Fig. 3

AUC variation graph of RippleNet algorithm parameters

Fig. 4

Screenshot of the prescription recommendation system

Table 1

Types and the Number of Nodes in Knowledge Graph

노드 타입	노드 개수
처방	2,466
약재	534
성분	6,713
타겟	4,221
질병	3,043

Table 2

AUC and ACC for Different Hop Numbers

홉 개수	AUC	ACC
2	0.908	0.846
3	0.913	0.828
4	0.919	0.848
5	0.914	0.827
6	0.917	0.819

Table 3

AUC and ACC for Different Sizes of Ripple Set

리플셋 크기	AUC	ACC
8	0.882	0.799
16	0.898	0.826
32	0.919	0.848
64	0.910	0.832
128	0.887	0.807

Table 4

AUC and ACC for Different Embedding Dimensions

임베딩 차원	AUC	ACC
4	0.890	0.819
8	0.919	0.848
16	0.569	0.542
32	0.552	0.504

Table 5

AUC and ACC for Different Batch Sizes

배치 크기	AUC	ACC
32	0.919	0.848
64	0.900	0.812
128	0.877	0.794
256	0.874	0.795

Table 6

AUC and ACC for Different L2 Regularization Weights

L2 정규화 계수	AUC	ACC
10-3	0.847	0.753
10-4	0.879	0.795
10-5	0.919	0.848
10-6	0.856	0.794
10-7	0.635	0.581

Table 7

AUC and ACC for Different KGE Weights

KGE 계수	AUC	ACC
0.01	0.898	0.811
0.02	0.915	0.847
0.03	0.919	0.848
0.04	0.902	0.823
0.05	0.897	0.813

Table 8

AUC and ACC for Different Learning Rates

학습률	AUC	ACC
0.01	0.842	0.768
0.02	0.870	0.797
0.03	0.890	0.802
0.04	0.884	0.780
0.05	0.919	0.848
0.06	0.908	0.824
0.07	0.885	0.814

참고문헌

1. Roy, D., & Dutta, M. (2022). A systematic review and research perspective on recommender systems. Journal of Big Data, 9, 59. 10.1186/s40537-022-00592-5

2. Jamali M., Ester M.2010. A matrix factorization technique with trust propagation for recommendation in social networks. In : Proceedings of the 4th ACM conference on Recommender systems; p. 135-142. 10.1145/1864708.1864736

3. Zhang F., Yuan J. N., Lian D., Xie X., Ma W. Y.2016. Collaborative knowledge base embedding for recommender systems. In : Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; p. 353-362. 10.1145/2939672.2939673

4. Sun, Y., Yuan, J. N., Xie, X., McDonald, K., & Zhang, R. (2017). Collaborative Intent Prediction with Real-Time Contextual Data. ACM Transactions on Information Systems, 35(4), 1-33. 10.1145/3041659

5. Guo, Q., Zhuang, F., Qin, C., Zhu, H., Xie, X., & Xiong, H., et al (2022). A Survey on Knowledge Graph-Based Recommender Systems. IEEE Transactions on Knowledge and Data Engineering, 34(8), 3549-3568. 10.1109/TKDE.2020.3028705

6. Wang H., Zhang F., Wang J., Zhao M., Li W., Xie X., et al2018. RippleNet: Propagating User Preferences on the Knowledge Graph for Recommender Systems. In : Proceedings of the 27th ACM International Conference on Information and Knowledge Management; p. 417-426. 10.1145/3269206.3271739

7. Kim, S. K., Lee, M. K., Jang, H., Lee, J. J., Lee, S., & Jang, Y., et al (2024). TM-MC 2: an enhanced chemical database of medicinal materials in Northeast Asian traditional medicine. BMC Complementary Medicine and Therapies, 24, 40. 10.1186/s12906-023-04331-y

8. Korea Institute of Oriental Medicine. (2013). Ontology-based Traditional Korean Medicine Knowledge Framework (Final report). Daejeon. Korea Institute of Oriental Medicine.

9. Szklarczyk, D., Santos, A., Mering, C., Jensen, L. J., Bork, P., & Kuhn, M. (2016). STITCH 5: augmenting protein-chemical interaction networks with tissue and affinity data. Nucleic Acids Research, 44, D380-384. 10.1093/nar/gkv1277

10. Piñero, J., Saüch, J., Sanz, F., & Furlong, L. I. (2021). The DisGeNET cytoscape app: exploring and visualizing Disease genomics data. Comput Struct Biotechnol J, 19, 2960-7. 10.1016/j.csbj.2021.05.015

11. Wang, H., Zhao, M., Xie, X., Li, W., & Guo, M. (2019). Knowledge Graph Convolutional Networks for Recommender Systems. The World Wide Web Conference, 3307-3313. 10.1145/3308558.3313417

12. Li, C., Cao, Y., Zhu, Y., Cheng, D., Li, C., & Morimoto, Y. (2024). Ripple Knowledge Graph Convolutional Networks for Recommendation Systems. Machine Intelligence Research, 21, 481-494. 10.1007/s11633-023-1440-x

13. Seo, J. S., Kim, S. K., Oh, Y. T., Kim, A. N., & Jang, H. C. (2014). Web based System for Supporting Medical Treatment in Korean Medicine based on Korean Medicine Ontology. Journal of Physiology & Pathology in Korean Medicine, 28(1), 113-121.

14. Kim, S. K., Lee, S., & Kim, A. (2024). A Prescription Support System using Synonyms of Indications in Prescriptions. The Journal of Herbal Formula Science, 32(4), 457-464. 10.14374/HFS.2024.32.4.457

15. Zhou, W., Yang, K., Zeng, J., Lai, X., Wang, X., & Ji, C., et al (2021). FordNet: Recommending traditional Chinese medicine formula via deep neural network integrating phenotype and molecule. Pharmacological Research, 173. doi.org/10.1016/j.phrs.2021.105752

16. Qian, Y., Wang, X., Cai, L., Han, J., Huang, Z., & Lou, Y., et al (2024). Model informed precision medicine of Chinese herbal medicines formulas–A multi-scale mechanistic intelligent model. Journal of Pharmaceutical Analysis, 14(4), 100914. doi.org/10.1016/j.jpha.2023.12.004

17. Fang, S. S., Dong, L., Liu, L., Guo, J. C., Zhao, L. H., & Zhang, J. Y., et al (2020). HERB: A high-throughput experiment- and reference-guided database of traditional Chinese medicine. Nucleic acids research, 49(1), D1197-D1206. 10.1093/nar/gkaa1063

18. Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., & Goyal, N., et al (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. 34th Conference on Neural Information Processing Systems, 793, 9459-9474. 10.5555/3495724.3496517