Research On Knowledge Extraction Technology For Chinese Medical Text

In the era of big data,the rapid development and extensive application of information technology push the medical industry to explore the direction of medical informatization,and it has become a mainstream trend.With the advancement of medical informatization,the medical field has accumulated a large amount of unstructured text data,which contains a lot of valuable knowledge.It is of great significance for the development of medical informatization to mine the effective knowledge from the mass of unstructured medical texts,and it is also a research hotspot in the field of natural language processing.As a semantic network with efficient knowledge representation,knowledge graph can store,manage,extend and apply knowledge effectively.Knowledge extraction is the core task in the construction of knowledge graph,and its extraction effect has a direct impact on the quality,expansion and application of knowledge graph.Therefore,this thesis mainly studies the technology of knowledge extraction for Chinese medical texts,and constructs a knowledge graph of medical domain based on this technology.The main work includes the following aspects:1.An attention-based Chinese medical named entity Recognition Model BGRU-CRF,BGRU-att-CRF,is proposed.The model first converts every character in the input medical text sentences into a character vector,and then uses the BGRU network to obtain the rich context information in the medical text sentences.Furthermore,the attention mechanism is used to select more relevant and dependent contextual semantic information.Finally,the CRF is used to obtain the global optimal solution of the medical entity label sequence to complete the recognition of the medical entity.The effectiveness of the proposed model in the task of Chinese medical named entity recognition is verified by comparing the overall performance and fine-grained performance with several reference models.2.A Chinese medical entity relation extraction model based on neural network and self-attention mechanism,BLSTM-MCATT-CNN,is proposed.The model first uses BLSTM to capture the context information and shallow semantic features of medical text sentences,and then applies CNN to capture the local phrase features of medical text sentences.Combined with multi-channel self-attention mechanism to capture the global information of medical text sentences,the semantic features of medical text are deeply mined.The optimal parameter combination of BLSTM-MCATT-CNN model is determined by parameter tuning experiments.Finally,the model is compared with several reference models in the overall performance.The validity of the model in the task of Chinese medical entity relation extraction is verified.3.Based on the above work of medical named entity recognition and medical entity relationship extraction,a disease-centered knowledge graph of medical domain is constructed by a top-down approach.The whole construction process includes pattern layer definition,knowledge extraction,knowledge fusion and knowledge storage.Based on this,a medical knowledge question answering system is designed and implemented,which can not only verify the practicability of knowledge graph,but also provide some help for patients to understand the knowledge of related diseases.