A Study Of Biomedical Entity Relationship Joint Extraction Method Based On Tagging Strategy

Biomedical entity relation extraction is an important information extraction task that aims to extract biomedical entity relation triples from unstructured biomedical texts.The entity relationship extraction method for biomedical text can help people better obtain the hidden biomedical information.It has important applications in pharmaceutical research and development,intelligent diagnosis and treatment,and many other areas.The traditional pipeline method has problems such as error propagation.Therefore,the paper studies the joint extraction method for the biomedical field according to the characteristics of biomedical texts.The paper’s main work is as follows:(1)The paper proposed a biomedical entity relationship joint extraction method based on BioBERT and multi-link handshaking tagging.The sentence encoding layer used in joint extraction is constructed by BioBERT,which reduces error propagation and effectively use the biomedical information contained in it.In view of the complex semantics of biomedical text and the overlap entity relationship,using the multi-link handshaking tagging as a joint decoding method,which enables the method to better deal with the complex entity relationship in biomedical text.The paper verifies the effect of the multilink handshake annotation strategy on biomedical texts by comparing with the baseline model,and conducts experiments on three biomedical public datasets to verify the validity of the method.(2)Aiming at the problems of long-distance semantic information loss and entity category information not being used effectively when only using BioBERT,an improved method that integrates Bi LSTM and entity category annotation module is proposed.The method uses Bi LSTM to better capture the semantic dependencies between long texts,and is more suitable for biomedical long text sequences.The method uses multi-layer annotation to achieve effective use of entity category information.Experiments were carried out on three biomedical data sets,and the F1 value was improved to a certain extent,and the ablation experiments were carried out to verify the effect of each module.(3)In view of the fact that each sentence of some biomedical datasets has only one head entity,a single-head and multi-tail labeling strategy that is more suitable for this type of data set is proposed.A joint extraction method for medical entity relationship.This method can effectively extract overlap triples of such datasets.Compared with the methods proposed in 1 and 2,this method reduces the training time while ensuring the extraction results.