Joint extraction method of entity relationship in Chinese Medicine based on Data Augmentation and Deep Learning

TCM texts are rich in evidence-based information, and a crucial step in knowledge mining is using high-tech tools to organize and store TCM texts in a systematic manner. Knowledge graphs are better suited to organizing and preserving the knowledge of TCM texts with complex relationships than typical databases are. Building high-quality knowledge maps requires accurate and efficient entity relationship extraction, and completely automated entity relationship extraction necessitates the creation of a sizable amount of high-quality corpus data, which increases expenses and lowers productivity. Because the same entity can generate many relations in the joint extraction of TCM entity relations and there are insufficient corpus data, these issues must be addressed. The joint extraction model of TCM entity interactions proposed in this paper is based on deep learning and data augmentation. Using a multi-head selective bidirectional long and short-term memory network (multi-head-BILSTM), the relationship overlap problem is first solved, and the data is then enhanced using five mechanisms: entity replacement, random addition, random deletion, random replacement, and integrated enhancement. We quantitatively assess the benefits and drawbacks of several relationship extraction algorithms as well as the performance improvements of TCM entity relationship joint extraction brought about by five alternative data augmentation mechanisms. In conclusion, our research has the potential to significantly enhance the efficiency of joint TCM entity relationship extraction.