Natural Language Processing Of Ancient Books Of Chinese Traditional Medicine Based On Deep Learning

Objective: Chinese medicine has a rich history spanning thousands of years,resulting in an extensive collection of ancient medical texts.These texts contain valuable theoretical and clinical knowledge that can be mined through natural language understanding research.In this study,we preprocess and annotate the ancient Chinese medical text "Synopsis of the Golden Chamber," create a standardized dataset,and employ a deep learning network model for named entity recognition and relationship extraction tasks.The extracted information is combined with manual review to construct a knowledge graph of the ancient Chinese medical text.Methods:For the named entity recognition task,we propose an ALBERT-BiLSTM-CRF model that uses ALBERT pre-trained model for word embedding,a bi-directional LSTM for contextual information,and CRF label inference to solve output dependency problems between labels.Multiple sets of comparison experiments are conducted on the preprocessed ancient Chinese medical text dataset.For the relationship extraction task,we use the Cas Rel deep learning neural network,which includes Subject Tagger and Relation-Specific Object Tagger,to extract entity relationship triads(Subject,Relation,Object).After named entity identification and relationship extraction,the data of entities and relationships are stored in the Neo4 j graph database.Results: The proposed ALBERT-BiLSTM-CRF model achieves an accuracy rate of95.65%,a recall rate of 93.97%,and an F1 value of 94.80% in recognizing all entities on the dataset.Compared to other models,the F1 value is improved by up to 6.3 percentage points.The Cas Rel model achieves an accuracy rate of 71.64%,a recall rate of 76.12%,and an F1 value of 73.81%.The knowledge graph constructed contains 1411 entity nodes and 2081 relationships.Conclusion: The experimental results show that the ALBERT-BiLSTM-CRF model and Cas Rel model are effective in named entity recognition and relationship extraction,respectively.The knowledge graph constructed provides groundwork for subsequent applications such as assisted diagnosis and treatment and intelligent question and answer systems.