Extraction of Relationship Between Esophageal Cancer and Biomolecules Based on BioBERT.

A large number of biomedical studies have presented the role of specific biomolecules (e.g., genes, miRNAs, and lncRNAs) in the occurrence and development of esophageal cancer, such as promoting and inhibiting the occurrence of esophageal cancer, as prognostic factors and biomarkers. The above information is of great significance for the early diagnosis and drug development of esophageal cancer. However, the information distribution is scattered, and the esophageal cancer text contains many specific terms and terms, so the entity recognition and relation extraction method proposed in the general language can't be directly used in esophageal cancer text mining. Therefore, an end-to-end joint relationship extraction model based on BioBERT was designed in this paper, and a biomolecular esophageal cancer relationship dataset was constructed to identify the relationship between molecules and esophageal cancer entities as well as extracted entities in biomedical literature. The designed model framework is mainly composed of a BioBERT pre-training language model coding layer, batch standardization layer, head entity decoding layer, and tail entity decoding layer, as well as constructs a corpus set of esophageal cancer relation tagging by manual tagging. The experimental results show that BioBERT SRO can not only effectively identify biomolecules and esophageal cancer entities in biomedicine, but also effectively solve the problem of relation overlap. The model has achieved high accuracy, recall and F1 value on the corresponding datasets.