Medical Knowledge Graph Construction Based on Traceable Conversion.

Medical knowledge graph (MKG) can provide ideal technical support for integrating multi-structure data and enhancing graph-based services. The construction of MKG usually requires information extracted from a large number of data sources, including structured data from medical databases (MDBs) and unstructured data from medical texts. However, the previous works used single data sources and simple format conversion when constructing MKG, and the MKG information constructed in this way is incomplete and untraceable. This paper proposes a method to build MKG based on traceable conversion to solve the above problems. For the structured data from MDB, the DB data is automatically converted into MKG nodes in the form of the RDF, which not only reduces the DB information loss in the conversion process but also enriches the types of graph nodes. When the data is efficiently converted, the converted nodes can also be traced back to the source. For the unstructured data from medical text, a strong deep learning model is used for entity and relation extraction. On the basis of avoiding the exposure bias and ensuring consistency of model training and prediction, the medical texts information is maximally extracted, and traceability is added, which reduces medical texts information loss and further complements the MKG. Based on the traceable conversion method for MKG construction, medical multi-structure data can be used more effectively to construct MKG.