背景:随着社会经济的蓬勃发展，医疗健康逐渐成为人民最为关心的话题之一。 目前我国医疗行业仍存在较多问题，其中由于国情使然，供需结构失衡、医疗资 源分配失衡等问题较为突出。面对当前困境，政府积极推进多种举措，其中包括 优化门诊分诊制度。然而目前，在线下面对大量的待诊患者时，医院的导诊人员 依然应接不暇。现有的导诊系统，智能性不高。患者通常难以做出正确的选择， 这不仅增加了治疗的成本，也造成了医疗资源的浪费。 目的:随着计算机技术的发展，医工交叉领域的研究逐渐成为热点，越来越多的 研究者使用人工智能等技术解决传统的医疗服务中的痛点问题。智能导诊系统为 指导患者进行挂号，提供了新的助力。基于传统人工导诊方式难以满足现有医院 和患者的导诊需求，以及现有智能导诊模型存在不足的现状，本研究提出了一种 基于深度学习的知识融合型智能导诊模型。 方法:本研究提出通过引入医学知识图谱和来提升导诊系统的智能化程度。首先， 提出使用基于深度学习的联合型医学信息提取模型和面向序列标记的五位标记 法及相关信息提取准则，构建医学知识图谱。其次，使用首次提出的面向导诊的 医学知识图谱推理方法，进行知识推理并且和传统的文本特征相融合，建立融合 知识的文本特征。最后使用多种深度学习模型，包括卷积神经网络、循环神经网 络、注意力机制等，构造了基于深度学习的知识融合型智能导诊模型。 结果:本研究构建了涉及 32 个科室的，包含 122895 条真实问诊数据的数据库。 在此数据库基础上，构建了基于深度学习的知识融合型智能导诊模型。模型最优 结果中加权准确率达到了 0.9629，加权召回率达到了 0.9627，加权 F1 达到了 0.9626，相比较传统的深度学习模型分别提升了 0.48%、0.50%和 0.49%。 结论:通过实验证明，本研究提出的基于深度学习的知识融合型智能导诊模型， 相较于传统的模型，在提升智能导诊性能的同时，增加了模型的可解释性。本模 型的设计构建有助于减轻导诊人员负担、提升医院分诊效率、改善医疗资源分配 失衡等问题。 

Objective: With the rapid development of social economy, medical care has always been one of the topics that people care most about.China has a large population, and there are still many problems in the medical industry, such as the imbalance of supply and demand structure, the imbalance of medical resource distribution and so on.Facing the current difficulties, the government has actively promoted various measures, including optimizing the outpatient triage system and establishing Internet hospitals. In the offline face of a large number of patients waiting for treatment, hospital counselors are often overwhelmed. Due to the complexity and professionalism of medical knowledge, it is often difficult for patients to make a correct choice when using the Internet medical platform, which not only increases the cost of treatment, but also causes the waste of medical resources.

Content: With the development of computer technology, the research in the cross field of medicine and industry has gradually become a hot spot. More and more researchers use artificial intelligence and other technologies to solve the pain points in traditional medical services. Intelligence guidance system to guide patients to register, online consultation, etc., provides a new power. Based on the fact that traditional manual guidance is difficult to meet the guidance needs of existing hospitals and patients, online consultation platforms need to classify consultation data into departments, and existing intelligent guidance models are inadequate, this study proposes a knowledge fusion intelligent guidance model based on deep learning.

Methods: In this study, medical knowledge map was introduced to improve the intelligence of the guidance system. Firstly, a medical information extraction method based on deep learning model and conditional random field was proposed to construct medical knowledge map. Secondly, a method of knowledge feature extraction based on medical knowledge graph is proposed, and the extracted features are fused with the traditional text features to establish the text feature of fusion knowledge. Finally, combined with a variety of deep learning models, including convolutional neural network, recurrent neural network, attention mechanism, etc., a knowledge fusion intelligent guidance model based on deep learning is constructed.

Results: A database containing 122895 real consultation data from 32 departments was constructed in this study. Based on this database, a knowledge fusion intelligent guidance model based on deep learning is constructed. The optimal results of the model are as follows: the weighted accuracy rate reaches 0.9629, the weighted recall rate reaches 0.9627, and the weighted F1 reaches 0.9626, which are respectively improved by 0.48%, 0.50% and 0.49% compared with the traditional deep learning model. Conclusion: The experiment proves that the knowledge fusion intelligent guidance model proposed in this study based on deep learning not only improves the performance of intelligent guidance, but also increases the interpretability of the model compared with the traditional model. This model is helpful to reduce the burden of medical guidance personnel, improve the efficiency of Internet hospital triage, and improve the imbalance of medical resource allocation.