Clinical Knowledge Graph Embeddings with Hierarchical Structure for Thyroid Treatment Recommendation

Thyroid knowledge graph constructed from EMRs contains abundant knowledge about patients. By applying representation learning model, the entities could be represented as embeddings for treatment recommendation, application of link prediction. However, the traditional representation learnings are limited in the field of thyroid disease: (1) clinical entities are different in magnitude and meanings (2) treatment entities are divided into two categories(operation and medicine). Thus, this paper proposed a framework apply to thyroid treatment recommendation with cold start based on TransD and network embedding with hierarchical structure, which are utilized respectively to solve the above problems. Finally, it provides the recommendation lists according to the ranking probabilities outputted by a binary classifier trained on the learned embeddings of the patients and treatments. Experiment compared with different components proved the applicability of the proposed framework.