Research On Extraction Of Drug Entity Relationship Based On Semantic Representation

In recent years,due to the gradual deepening of research in the field of biomedicine,the biomedical literature has exploded rapidly.And with the rise of the Internet,knowledge sharing around the world has become possible.However,due to the huge amount of medical literature,it becomes more and more difficult to accurately extract some knowledge in it.Although there are related medical literature search engines,the documents found still need to be read manually.This will consume a lot of manpower and time.How to extract relevant knowledge from unstructured documents has become an urgent problem.In medical experiments,the interaction between drugs is of great significance.Improper use of drugs can cause serious consequences.Therefore,extracting structured information about drugs in the medical literature can reduce medical costs and reduce drug-related issues and safety accidents,it is of great significance to accelerate the launch of new drugs.With the gradual maturity of deep learning technology,relevant knowledge mining in medical texts has become a popular application in current computer science.The structured information extraction of medical text in this paper is mainly divided into two tasks,drug entity recognition and drug-drug relationship extraction.The main research contents include the following parts:Using the latest medical field pre-trained model BioBERT and LSTM network and CRF tags relies on identifying drug entities in medical text.Compared with the generalpurpose pre-trained model BERT,it improves the accuracy of entity recognition and is more suitable for medical text.The groundbreaking use of abstract meaning representation to semantically represent the input text sequence,and by analyzing the structure of the abstract meaning representation graph,it is found that the shortest path between entities in the graph contains important semantic relationships,and the shortest path sequence is used as the model's One of the inputs constructs a medical text relation extraction model.A two-layer bidirectional LSTM attention model is used for relationship extraction.The input of a sentence divides the sentence's attention vector into three context subsequences for feature extraction based on the position of the entity.At the same time,it adds abstract relational semantics and uses the top-level LSTM network for overall feature representation.Experiments show that this method effectively improves the overall performance of the model.