Guided Intelligent Collection Of Elements Of Doctor-patient Dispute Cases

The doctor-patient dispute mediation center’s doctor-patient dispute mediation platform is a platform that assists mediators,doctor-patient parties,and managers at all levels in online registration,online case handling,online mediation,and online supervision and guidance of dispute.In order to solve the problems of inconvenient use,inaccurate entry,and unstructured data in the existing doctor-patient dispute registration system in this platform,this paper designs and implements a guided intelligent collection scheme for elements of doctor-patient dispute cases.In order to achieve online case registration while retaining the guidance and interaction features of offline mediator registration,and to help users focus on the efficient completion of registration tasks,the scheme uses a dialogue robot to serve patient dispute registration: the dialogue robot can use the dialogue method provide patients with key information entry guidance,multiple rounds of inquiry and confirmation to complete specific dispute information registration tasks.The dialogue robot can intelligently extract structured key information from the patient’s statement during the interaction process.Due to the specific field of doctor-patient disputes,the research object of this paper can be positioned as a task dialogue in a closed domain.RASA is an open source framework for intelligent dialogue systems that has emerged in the Bot field in recent years.Compared with other dialogue platforms such as API.AI and UNIT,it has the advantages of supporting local deployment optimization,customizable components and strategies,high flexibility,and relatively mature development.Therefore,this article research how to build a human-machine dialogue system for the registration of doctor-patient dispute cases based on RASA.Among them,the accuracy of the collection of case elements mainly depends on the dialogue understanding module in the dialogue system.Whether the guidance is reasonable and accurate mainly depends on the dialogue management module.First of all,in terms of dialogue understanding,user input is mostly short text,and there are difficulties in parsing and parsing and lack of grammar.RASA’s built-in intent recognition and entity extraction components are based on general statistical learning and machine learning algorithms.In the field of disputes,both the accuracy of intention recognition and the f1 value of entity extraction are lacking.To this end,this paper introduces the deep learning algorithm Bi LSTM-CRF and Bert pre-training model to construct a new intent recognition and entity extraction component to improve its accuracy and f1 value.In addition,the entities and intentions in the dialogue are usually related to each other,and the intention recognition model and the entity extraction model often use mutually independent assumptions,which will ignore the internal relationship between them,resulting in a situation where the intention and entity identified in the same sentence do not match.Therefore,this paper focuses on the multi-condition fusion Bi LSTM-CRF intent recognition and entity extraction joint model,using one model to do both intent recognition and entity extraction to improve efficiency and accuracy.Secondly,in terms of dialogue management,the built-in dialogue strategy model in RASA’s multi-round dialogue management is based on simple supervised learning,which is difficult to meet the complex business needs in the field of doctor-patient disputes,and it needs to provide dialogue scene corpus in various situations,which consumes a lot of labor.When applied to the dialogue scene of doctor-patient disputes with limited corpus,the effect is not ideal.However,the single-round question-and-answer strategy has a weak ability to deal with complex scenarios and a poor user experience.Besides,the action prediction strategy based on probability model or deep reinforcement learning,which is the core of dialogue management,is difficult to be controlled and difficult to be applied to actual business scenarios.In order to achieve efficient and accurate collection of information on the elements of medical disputes,this paper still adopts a multiround dialogue management scheme,and proposes a multi-classification model based on the self-attention mechanism to improve performance and effects based on the characteristics of medical disputes.Finally,based on the RASA framework,this article builds a dialogue service for the registration link of doctor-patient disputes,realizes intent identification,entity extraction,and management of multiple rounds of dialogue,which can guide users reasonably,achieve more accurate intelligent information collection,and directly obtain key information of structured cases.Experiments show that the optimized independent model components and the multi-condition fusion Bi LSTM-CRF joint model effectively improve the accuracy of intention recognition and the f1 value of entity extraction in the field of doctor-patient;compared with the RASA built-in scheme,the multi-round dialogue management scheme based on the self-attention mechanism has a greater improvement in the accuracy of motion prediction,which verifies the effectiveness and feasibility of the scheme proposed in this paper.