MC-eLDA: Towards Pathogenesis Analysis in Traditional Chinese Medicine by Multi-Content Embedding LDA
Traditional Chinese medicine (TCM) is well-known for its unique theory and effective treatment for complicated diseases. In TCM theory, “pathogenesis” is the cause of patient’s disease symptoms and is the basis for prescribing herbs. However, the essence of pathogenesis analysis is not well depicted by current researches. In this paper, we propose a novel topic model called Multi-Content embedding LDA (MC-eLDA), aiming to collaboratively capture the relationships of symptom-pathogenesis-herb triples, relationship between symptom-symptom, and relationship between herb-herb, which can be used in auxiliary diagnosis and treatment. By projecting discrete symptom words and herb words into two continuous semantic spaces respectively, the semantic equivalence can be encoded by exploiting the contiguity of their corresponding embeddings. Compared with previous models, topic coherence in each pathogenesis cluster can be promoted. Pathogenesis structures that previous topic modeling can not capture can be discovered by MC-eLDA. Then a herb prescription recommendation method is conducted based on MC-eLDA. Experimental results on two real-world TCM medical cases datasets demonstrate the effectiveness of the proposed model for analyzing pathogenesis as well as helping make diagnosis and treatment in clinical practice.
KeywordsTopic modeling Embedding Traditional Chinese medicine
This work was supported by National Key R&D Program of China (No. 2017YFC0803700), NSFC grants (No. 61532021 and 61472141), Shanghai Knowledge Service Platform Project (No. ZF1213), and SHEITC.
- 3.Li, Y., Li, H., Wang, Q., et al.: Traditional Chinese medicine formula evaluation using multi-instance multi-label framework. In: 2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 484–488. IEEE (2016)Google Scholar
- 4.Wang, S., Huang, E.W., Zhang, R., et al.: A conditional probabilistic model for joint analysis of symptoms, diseases, and herbs in traditional Chinese medicine patient records. In: 2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 411–418. IEEE (2016)Google Scholar
- 5.Everitt, B.S., Dunn, G.: Principal components analysis. Appl. Multivar. Data Anal. Second Ed. 48–73 (1993)Google Scholar
- 6.Fakoor, R., Ladhak, F., Nazi, A., et al.: Using deep learning to enhance cancer diagnosis and classification. In: Proceedings of the International Conference on Machine Learning (2013)Google Scholar
- 9.Das, R., Zaheer, M., Dyer, C.: Gaussian LDA for topic models with word embeddings. In: ACL, vol. 1, pp. 795–804 (2015)Google Scholar
- 10.Porteous, I., Newman, D., Ihler, A., et al.: Fast collapsed Gibbs sampling for latent Dirichlet allocation. In: Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 569–577. ACM (2008)Google Scholar
- 11.Vincent, P., Larochelle, H., Bengio, Y., et al.: Extracting and composing robust features with denoising autoencoders. In: Proceedings of the 25th International Conference on Machine Learning, pp. 1096–1103. ACM (2008)Google Scholar
- 13.Mikolov, T., Chen, K., Corrado, G., et al.: Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 (2013)
- 15.Jiang, Z., Zhou, X., Zhang, X., et al.: Using link topic model to analyze traditional Chinese medicine clinical symptom-herb regularities. In: 2012 IEEE 14th International Conference on e-Health Networking, Applications and Services (Healthcom), pp. 15–18. IEEE (2012)Google Scholar