Research On Named Entity Recognition Based On Lattice-BiGRU Electronic Medical Records

As the most direct actual medical records,electronic medical records contain extremely rich medical knowledge and clinical experience.However,such texts are mostly unstructured data,and internal information cannot be fully utilized.Therefore,in practical application,manual information extraction is still the main method,which has low efficiency and high cost.In order to improve the utilization rate of electronic medical record data,reduce labor costs,and provide the foundation for the construction of the smart medical platform,the research work in this thesis focuses on the Lattice-BiGRU-CRF based named entity recognition of electronic medical record.The main innovative work is as follows:1.A named entity recognition method for electronic medical records based on LatticeBiGRU-CRF is proposed.The BiGRU based on character method is the backbone network.The word based methed and the character-based method are connected through Lattice.At the same time,the word information and word information are used for entity recognition.The sequence labeling probability is calculated by CRF layer to get the final recognition result.The ambiguity recognition problem in named entity recognition is avoided to some extent.The influence of word information on character information is fully considered in the proposed algorithm.The recognition efficiency is improved by it.And the complexity of the original Lattice model is reduced,and good accuracy,recall,and F1 value can be obtained.2.Self-built medical dictionary is used to identify and correct professional words that cannot be identified by Lattice-BiGRU-CRF model,which further improves the accuracy of named entity recognition.On the basis of reducing the model complexity,the word vector matching of the model is further lightened.In view of the particularity of the electronic medical record text,the Lattice-BiGRU-CRF model involves some medical field word vectors by adding the medical dictionary while keeping the medical information of the words embedded in the file unchanged.The model filters out word vectors that are not pertinent to identifying named entities in the electronic medical record,such as names and locations.By lightening the irrelevant embedded word vectors in the model,the recognition effect of the named entities of the electronic medical record text is enhanced,and the running speed of the model is accelerated again,which is much less than that of the original model.At the same time,considering the professionalism of named entities in electronic medical records,a medical dictionary was built to fine-tune the recognition results of the Lattice model,which further reduced the recognition errors.Experimental results show that the algorithm proposed in this thesis has advantages over the existing algorithms in the accuracy,recall and F1 value of objective indicators,while the model complexity is reduced and the training speed is effectively improved.