Research On Guidance Model Based On Deep Learning

With the development of Internet+medicine,online medical treatment has gradually become a new development direction in the medical industry.Before using online medical services,patients must first choose the right department to register according to their condition,but patients often lack professional medical knowledge and often have the problem of registering wrongly,which in turn delays the medical treatment of their condition.Therefore,online medical platforms provide patients with intelligent triage services.Most of the current triage models require patients to actively provide information about medical entities such as symptoms or diseases,and then the triage models recommend departments to patients based on this information.The problem of how to build an accurate medical triage model to provide professional help to patients and avoid the waste of medical resources has become a pressing issue to be solved.The research work in this paper revolves around building a deep learning-based guidance model,aiming to accurately recommend the corresponding department according to the patient’s medical expression in order to improve the patient’s medical experience.(1)To address the problem that patients are prone to registering wrongly due to their lack of medical knowledge,this paper proposes a guidance model based on ERNIE-Bi GRU.The text vector representation is generated by ERNIE pre-training model,and the proposed model achieves the best experimental results after extracting the deep features of text through Bi GRU network and comparing experiments with five embedding models and three classification models on two data sets.(2)To address the problem of low accuracy of medical entity recognition in traditional guidance models,this paper proposes an improved model of medical entity recognition incorporating attention mechanism,which is based on the traditional medical entity recognition model Bi LSTM-CRF,adding attention mechanism to improve the recognition ability of the model,and conducting comparison experiments with several medical entity recognition models to verify the effectiveness of the improved model.(3)To address the problem of low efficiency of single-task training,we propose a multitask learning-based medical guide model,train both the medical guide model and the entity recognition model and explore two different ways of parameter sharing.The experimental results show that the multi-task learning-based medical guide model outperforms the singletask based medical guide model and validate the effectiveness of multi-task learning.Three multi-task loss functions are also compared,combined with the multi-task learning model proposed in this paper for comparison experiments,and the experimental results show that the method of dynamically adjusting the weights of the loss function using homoscedastic uncertainty works better.