摘要：

Abstract:

【Purpose/significance】To explore the knowledge extraction and knowledge fusion methods for multi-source heterogeneous

medical health data, optimize the automatic construction process of knowledge graph, and improve the ability to integrate multi-source

heterogeneous data and the efficiency and quality of information retrieval by users【. Method/process】Firstly, based on UMLS medical

terms and medical concepts, the topic analysis of unstructured data is carried out by using BTM topic model to assist ontology design.

Then, by comparing the embedding effect of BERT-Base, BioBERT, and MC-BERT as the coding end of CasRel model, the Mc-bert

casrel model with the best effect is selected to extract the entity and relational triad from the unstructured data, and data reorganization

is carried out on the semi-structured data. Establish the relationship between entities; Then SapBERT model and Levenshtein editing

distance algorithm are used for data fusion of triples. Finally complete the construction of knowledge graph【. Result/conclusion】Based

on the method proposed in this paper, the knowledge map of "digestive system diseases" containing 10010 entities and 29044 relation⁃

ships is constructed, and the knowledge retrieval application is realized【. Innovation/limitation】This paper focuses on the integration of

multi-source heterogeneous medical and health data, and provides a new idea and method for the construction process of knowledge

graph in the vertical field of medical and health under the Internet environment. However, the sample size of data sources is limited, so

knowledge extraction and knowledge fusion on larger data sets can be considered in the future.