MeDical Question Answering System Based On Knowledge Graph

The emergence and development of search engines have brought great convenience to people’s information retrieval.The search needs of users in the medical and health field have increased year by year,and the current search engines return mostly related pages,and users can identify and filter the returned results.The question and answer system provides a solution to the above problems,and can provide users with accurate and timely medical consulting services.However,the current existing medical consulting services have problems such as strong subjectivity in content and low efficiency in communication between doctors and patients.This paper studies and implements a medical question and answer system based on a knowledge graph,establishes a deep learning model to understand user questions,and retrieves answers from a knowledge graph composed of a large amount of medical data.Compared with the results returned by search engines,it is more suitable for users.Retrieve requirements.The most important part of question answering system is the understanding of the user’s natural language,and the question raised by the user is transformed into information words and user intentions.Therefore,the task is decomposed into two subtasks in the specific implementation process,medical entity recognition and question classification.The work of this paper mainly includes the following aspects:(1)The construction of medical knowledge graph.The data basis of the question and answer system is the knowledge graph.The crawler technology is used to collect raw data on multiple medical information platforms.The three processes of knowledge extraction,knowledge fusion and knowledge storage are used to complete the construction of the medical knowledge graph,which integrates medical treatment from multiple data sources.(2)Study how to identify medical-related entities from question sentences,introduce a pre-training language model into the original Bi LSTM-CRF model to complete this task,and prove through experiments that the accuracy of this method is on the self-built medical question-and-answer data set.The CCKS2017 electronic medical record data set has been improved.(3)Research how to classify question sentences in medical question and answer scenarios to determine user intentions.The BERT-softmax model is established according to the characteristics of the constructed data set.The feature vector is obtained through the BERT model,and the softmax layer is normalized to obtain the corresponding classification.Compared with several other text classification models,the classification effect on the question and answer dataset has been greatly improved.(4)The question and answer system is designed and implemented on the basis of the above three work contents,which is divided into 9functional modules.The question and answer system is divided into two parts: the We Chat applet for users and the question and answer management system for administrators.Users can enter the question through the We Chat applet entrance page and enter the consultation page to ask questions.This realizes human-computer interaction and supports medical-related questions.For consultation,the administrator can manage and configure the related functions of the question and answer system through the management system,and support the update of the existing knowledge graph.