Research On The Method Of Generating Medical Record Summarization Based On Discourse Analysis

By applying automatic text summarisation to medical texts,it is possible to advance the theoretical realisation of automatic text summarisation.The combination of automatic text summarisation with certain specific problems in the medical field can also solve some of the practical problems encountered in the medical field.For example,a patient’s electronic medical record is processed by means of a neural network,for example,to analyse the main information in the electronic medical record.After processing,the summarisation is extracted and presented to the medical staff,which can effectively assist them in their treatment services.Electronic medical record summarisation is of great practical value,but there is little research on Chinese electronic medical record summarisation in the Chinese language.The use of discourse analysis and other methods to study Chinese electronic medical records,as well as further analysis of the structural characteristics of Chinese electronic medical records,can effectively improve the effectiveness of electronic medical record summarisation generation.The elementary discourse unit(EDU)is the smallest unit in discourse analysis and also forms the basis of discourse.By introducing EDUs into text processing techniques,natural language processing tasks such as text summarisation and text compression can be effectively enhanced.Although some researchers have already conducted research on discourse analysis,there is no research on discourse analysis for Chinese electronic medical records,and likewise no research on EDUs for electronic medical records.Thus an EDU corpus for Chinese electronic medical records is lacking.At the same time,although combining EDU with text summarisation can enhance the final effect of text summarisation,different combinations will produce different enhancement effects.In response to the above,this paper makes the following contributions.· Combining punctuation and main statement bit theory,a method for constructing a corpus of EDUs for Chinese electronic medical records is proposed.According to this scheme,a corpus of EDUs for Chinese electronic medical records,CCEMR,is constructed.500 documents with nearly 150,000 EDUs are available in the CCEMR corpus,covering a wide range of diseases and involving patients of different ages,which is highly representative.Experiments on the EDU recognition task were conducted on the CCEMR corpus using the EDU recognition model.The experimental results show that the EDU recognition model can basically reach the human recognition level when performing EDU recognition on Chinese electronic medical records.This demonstrates the effectiveness of the corpus construction method.· By combining multi-grain heterogeneous graphs and graph attention networks,a multi-grain Chinese electronic medical record summary model MCMS is proposed,which can further capture the potential information in Chinese electronic medical records by constructing multi-grain graph structures of words,EDUs and sentences.A dataset with 5000 electronic medical records was also constructed for experimental validation.The final results of the MCMS model on this dataset were obtained as R1 value 47.7,R2 value 30.6 and RL value 39.1.The experimental results show that the MCMS model is able to achieve an advanced level among the extractive abstraction models of its type for the Chinese electronic medical record abstraction task.