Research On Guidance Model Integrating Medical Entity Concept Knowledge

With the development of economy,the improvement of people’s living standards and the continuous increase of average life expectancy,the demand for hospitals and related medical institutions is also increasing,and the phenomenon of short-term medical resources is emerging.Among them,the hospital guide resources run is a direct reflection of this phenomenon.The referral service is a window for patients to use medical resources.Its main function is to provide patients with simple department recommendations and other functions.Although some hospitals can provide referral services,based on the huge population of our country,there is often a run on the referral resources in the hospital.Therefore,more and more patients begin to choose online medical treatment.Through various medical websites on the Internet,they can enjoy high-quality medical services without leaving home.However,before using medical services,they need to understand their own conditions to choose the appropriate medical department,and then choose their favorite doctors from the department for consultation.Therefore,constructing a guidance model that can classify patients into the corresponding departments according to the patient’s own condition description information is an important means to solve this problem.The early guidance model is to understand the questions raised by patients through manually constructed rule templates,but this method requires a lot of human and material resources,and requires professional domain knowledge,and the effect of the model is limited by the size of the template library and keyword library.In recent years,with the development of deep learning technology,more and more researchers use deep learning to implement a guidance system.However,the general deep learning guidance model ignores the help of the conceptual knowledge of medical entities in the text to the classification of departments.Therefore,this paper will use the conceptual knowledge of medical entities to enrich the feature information in the original short text,and combine with deep learning to propose a guidance model.The main work of this paper is as follows:(1)Aiming at the problem that the short text of disease description provided by patients may have sparse features,this paper proposes a method to enrich the feature information of the original text through external medical knowledge.Firstly,the required medical entities need to be identified in the text.Therefore,this paper introduces a multi-head attention mechanism into the traditional Bi GRU-CRF named entity recognition model to improve the recognition ability of the model,so as to obtain the medical entities in the text more accurately.Finally,the annotated entity words are used to obtain the conceptual knowledge of the medical entity corresponding to the CN-Probase knowledge graph.Finally,this paper verifies the effectiveness of the named entity recognition model used in medical text through a series of experiments.The accuracy,recall and F1 value of the named entity recognition model proposed in this paper on the CMe EE dataset reach 89.37%,90.41%and 89.89%respectively.(2)MCBERT model is used to represent the word vector.The conceptual knowledge text and illness description text of medical entities are obtained by the method in Chapter 3,and the disease description text vector and the conceptual knowledge vector are spliced to obtain a composite vector.Then,the concatenated word vector is input into the MCBERT-MHA-Bi LSTM-CNN guide model.The model takes into account the long-distance association relationship and local feature information in the text context,and integrates the multi-head attention mechanism to improve the extraction ability of key features.In addition,Swish activation function is introduced to solve the problems of Re LU activation function.Because the input integrates external knowledge,it is easy to generate noise,and the traditional pooling layer for text feature dimension reduction will lose part of the semantic information,so the singular value decomposition algorithm is used in Text CNN to replace the pooling layer to realize dimension reduction,noise reduction and feature extraction.And the model has achieved an accuracy of 86.9%on the dataset,which has better results than other models,and the effectiveness of the improved method proposed in this paper is verified by experiments.