Biomedical knowledge discovery based on Sentence‐BERT. Issue 1 (22nd October 2020)

• Record Type: Journal Article
• Authors: Shen, Si; Liu, Xiao; Sun, Hao; Wang, Dongbbo
• Is Part Of: Proceedings of the Association for Information Science and Technology. Volume 57:Issue 1(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-10-22
• Subjects: knowledge discovery -- knowledge graph -- Sentence‐BERT; Information science -- Congresses; Information technology -- Congresses; Information science; Information technology; Conference papers and proceedings; 020
• DOI: 10.1002/pra2.362
• Languages: English
• ISSNs: 2373-9231

Abstract

Abstract: In the advent of the big data era, the volume of biomedical information grows exponentially, which means that the researchers have to spend a large amount of time in biomedical information mining. Effective mining of biomedical information and building the knowledge graph in the biomedical field are of high importance for biomedical knowledge discovery. In this study, datasets were constructed by using the text similarity corpus of CDD. The average Pearson's correlation coefficient was calculated by using the Sentence‐BERT model with sentence embeddings. Then a comparison was performed among the wordnet‐based similarity method (WBSM), UMLS‐based similarity method (UBSM), and the combination of these two methods. The results showed that the average Pearson correlation r calculated by the Sentence‐BERT model was the highest, the value of which is 74.37%. Triplets consisting of diseases, genes and proteins were extracted by using the Sentence‐BERT model. On this basis, the biomedical knowledge graph was constructed. It was found that this knowledge graph had a good effect on the testing dataset.