WO2022041727A1

WO2022041727A1 - Question and answer management method, apparatus, and device for medical inquiry system, and storage medium

Info

Publication number: WO2022041727A1
Application number: PCT/CN2021/084651
Authority: WO
Inventors: 李响; 柳恭; 满晏松
Original assignee: 康键信息技术（深圳）有限公司
Priority date: 2020-08-28
Filing date: 2021-03-31
Publication date: 2022-03-03
Also published as: CN112035636A; CN112035636B

Abstract

A question and answer management method, apparatus, and device for a medical inquiry system, and a storage medium, applied to the field of wisdom medical treatment and capable of reducing the online misdiagnosis rate. The method comprises: obtaining target feature data from a target terminal; calling a preset neural network pre-classification model to pre-classify the target feature data; if the target feature data is a first type of data, calling a preset knowledge graph model and the first type of data to perform medical graph reasoning to generate first diagnosis suggestion data and send same to the target terminal; if the target feature data is a second type of data, performing medical graph query according to a preset knowledge graph decision tree model and the second type of data to generate multiple supplemental problems and send same to the target terminal; generating electronic medical record data; and calling the neural network pre-classification model to reperform pre-classification processing on the electronic medical record data until second diagnosis suggestion data is generated and sent to the target terminal.

Description

Question and answer management method, device, equipment and storage medium for medical consultation system

This application claims the priority of the Chinese patent application filed on August 28, 2020 with the application number 202010884354.2 and the invention titled "Q&A management method, device, equipment and storage medium for medical consultation system", all of which are The contents are incorporated by reference in the application.

technical field

The present application relates to the field of medical data, and in particular, to a question and answer management method, apparatus, device and storage medium for a medical consultation system.

Background technique

Internet medical treatment is an important field of artificial intelligence application. With the rapid development of technology, mobile medical has moved closer to the diagnosis and treatment level and made breakthroughs. In recent years, online consultation has become more and more popular, and the number of online consultations in a single day has already exceeded the average daily scale of one million. The shortage of online doctor resources and the low efficiency of online consultation services have become prominent problems. Because of this, the development of high-quality computer-aided medical systems has become a key area of research and development breakthroughs in major companies and scientific research institutions. Artificial intelligence systems, especially natural language understanding, have gradually become possible to solve this proposition in the context of significant breakthroughs in algorithms and computing power.

The inventor realized that the traditional automated consultation process is to ask preset questions through a defined consultation flow chart and collect user information. This solution will throw out the preset consultation paths in turn. It is determined that the system will not jump and switch according to the specific situation of the user. Because Internet hospitals are limited by the complex nature of medical care and the characteristics of physical space isolation in the Internet, the preset problems do not match the needs of users, resulting in a high rate of online misdiagnosis.

SUMMARY OF THE INVENTION

The present application provides a question and answer management method, device, equipment and storage medium for a medical consultation system, which are used to accurately identify the medical field intention of a user consultation and improve the accuracy of the question and answer management results of the medical consultation system.

In order to achieve the above object, a first aspect of the present application provides a question-and-answer management method for a medical consultation system, including: acquiring target feature data from a target terminal, where the target feature data is used to instruct a target user to contact a medical practitioner through the target terminal. The consultation information input by the consultation system; the preset neural network pre-classification model is called to pre-classify the target feature data, and the pre-classification result corresponding to the target feature data is determined, and the pre-classification result includes the first type of data and the second type of data, the first type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information; if the target feature data is the first type of data, Then call the preset knowledge graph model and the first type of data to perform medical graph reasoning, generate the first diagnosis suggestion data and send it to the target terminal; if the target feature data is the second type of data, according to The preset knowledge graph decision tree model and the second type of data perform medical graph query, generate multiple rounds of supplementary questions and send them to the target terminal; generate electronic data according to the answers to the multiple rounds of supplementary questions and the target feature data. Medical record data; invoking the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

A second aspect of the present application provides a question and answer management device for a medical consultation system, including a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor, and the processor executes The computer-readable instructions implement the following steps: acquiring target feature data from the target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal; The network pre-classification model pre-classifies the target feature data, and determines the pre-classification result corresponding to the target feature data. The pre-classification result includes the first type of data and the second type of data, and the first type of data is the question. If the target feature data is the first type of data, the preset knowledge graph model and the first type of data are called Perform medical graph reasoning on the data, generate first diagnosis suggestion data and send it to the target terminal; if the target feature data is the second type of data, then according to the preset knowledge graph decision tree model, the second type Perform medical map query on the data, generate multiple rounds of supplementary questions and send them to the target terminal; generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data; call the neural network pre-classification model to The electronic medical record data is pre-classified again until the second diagnosis suggestion data is generated and sent to the target terminal.

A third aspect of the present application provides a computer-readable storage medium, where computer instructions are stored in the computer-readable storage medium, and when the computer instructions are executed on a computer, the computer is caused to perform the following steps: obtaining a target from a target terminal Feature data, the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal; call the preset neural network pre-classification model to pre-classify the target feature data, and determine the The pre-classification result corresponding to the target feature data, the pre-classification result includes the first type of data and the second type of data, the first type of data is the data with complete types of consultation information, and the second type of data is the consultation Data with missing information types; if the target feature data is the first type of data, call the preset knowledge graph model and the first type of data to perform medical graph inference, generate the first diagnosis suggestion data and send it to the the target terminal; if the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions and send them to the target terminal; generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data; invoke the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

A fourth aspect of the present application provides a question-and-answer management device for a medical consultation system, comprising: a data acquisition module configured to acquire target feature data from a target terminal, where the target feature data is used to instruct a target user to send a request to a user through the target terminal to The consultation information input by the medical consultation system; the discrimination module is used to call the preset neural network pre-classification model to pre-classify the target feature data, and determine the pre-classification result corresponding to the target feature data, the pre-classification The result includes a first type of data and a second type of data, the first type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information; the atlas reasoning module, if the target If the feature data is the first type of data, it is used to call the preset knowledge graph model and the first type of data to perform medical graph inference, generate the first diagnosis suggestion data and send it to the target terminal; the graph tree logic module, if the target feature data is the second type of data, it is used to perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions and send them to the a target terminal; an electronic medical record module for generating electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data; The data is pre-classified again until the second diagnostic suggestion data is generated and sent to the target terminal.

In the technical solution provided by the embodiment of the present application, target feature data is obtained from the target terminal, and the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal; The target feature data is pre-classified, and the pre-classification result corresponding to the target feature data is determined. The pre-classification result includes the first type of data and the second type of data; if the target feature data is the first type of data, the preset knowledge graph model and Perform medical graph inference on the first type of data, generate first diagnostic suggestion data and send it to the target terminal; if the target feature data is the second type of data, perform medical graph query according to the preset knowledge graph decision tree model and the second type of data , generate multiple rounds of supplementary questions and send them to the target terminal; generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and target feature data; call the neural network pre-classification model to pre-classify the electronic medical record data until the second diagnosis suggestion is generated data and sent to the target terminal. The embodiment of the present application reduces the online misdiagnosis rate, saves the consultation time of the Internet hospital, and improves the consultation efficiency of the Internet hospital per unit time.

Description of drawings

1 is a schematic diagram of an embodiment of a question-and-answer management method of a medical consultation system in an embodiment of the application;

2 is a schematic diagram of another embodiment of the question-and-answer management method of the medical consultation system in the embodiment of the application;

3 is a schematic diagram of an embodiment of a question-and-answer management device of a medical consultation system in an embodiment of the present application;

4 is a schematic diagram of another embodiment of the question and answer management device of the medical consultation system in the embodiment of the application;

FIG. 5 is a schematic diagram of an embodiment of a question and answer management device of a medical consultation system in an embodiment of the present application.

detailed description

The present application provides a question and answer management method, device, equipment and storage medium for a medical consultation system, which are used to reduce the online misdiagnosis rate, save the consultation time of Internet hospitals, and improve the efficiency of Internet hospital consultations per unit time.

In order to make those skilled in the art better understand the solutions of the present application, the embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application.

The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It is to be understood that data so used may be interchanged under appropriate circumstances so that the embodiments described herein can be practiced in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those expressly listed steps or units, but may include other steps or units not expressly listed or inherent to these processes, methods, products or devices.

Please refer to FIG. 1 , a flowchart of a question-and-answer management method of a medical consultation system provided by an embodiment of the present application, which specifically includes:

101. Acquire target feature data from a target terminal, where the target feature data is used to indicate consultation information input by a target user to a medical consultation system through the target terminal.

The server receives the target feature data sent by the target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal. Wherein, the consultation information includes the text of the consultation dialogue and the basic information of the target user, and the basic information of the target user includes the age information of the target user, the gender information of the target user, and the main appeal information of the target user.

It can be understood that the execution subject of the present application may be a question-and-answer management device of a medical consultation system, or may be a server, which is not specifically limited here. The embodiments of the present application take the server as an execution subject as an example for description.

It should be noted that when obtaining the target feature data, each sentence needs to be segmented according to the word segmentation method of string matching. The custom word segmentation vocabulary corresponding to different features is not the same, and the text collection can be clicked from left to right. Different characters are separated by spaces, and the text set here is the target feature data.

102. Invoke a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data, and the first type of data is consultation Data with complete information types, and the second type of data is data with missing medical information types.

The server calls the preset neural network pre-classification model to pre-classify the target feature data, and determines the pre-classification result corresponding to the target feature data. The pre-classification result includes the first type of data and the second type of data, and the first type of data is consultation information. Data with complete types, and the second type of data is data with missing types of consultation information.

The pre-classification result not only includes the data type, that is, the pre-classification result is the first type of data or the second type of data, but also includes a pre-classification value (classification prediction value). The server can judge the size of the target data according to the size of the pre-classification value. completeness. When the pre-classification value is greater than a certain threshold (such as the first threshold), it will prompt that the collected information is complete (that is, the data types are complete), the diagnosis can be completed, and the consultation is ended. Similarly, when the pre-classification value is less than a certain threshold (also the first threshold), the server will prompt that the collected information is not complete (that is, the data type is incomplete or missing), which means that the knowledge graph decision tree model needs to be called to generate supplementary questions and Return to the target terminal of the target user, so that the target user can describe the supplementary questions and obtain more consultation information, wherein there are multiple supplementary questions, so as to obtain as much required consultation information as possible.

It should be noted that the standard of whether the collected information is complete (that is, whether the data type is complete) refers to the existence of necessary parameters in the target feature data, and these necessary parameters include the text of the consultation dialogue, the age information of the target user, and the gender information of the target user. , the target user's main appeal information, case history information, keyword information, and category label information, where the category label information mainly refers to the disease information that the target user may belong to. For example, if the target user mentions keywords in the text dialogue " Liver", then the category labels corresponding to the target user may include labels such as "abnormal liver function", "abnormal visceral function", "alcohol", "abnormal metabolism", etc. The same user can correspond to multiple category labels. Repeat.

103. If the target feature data is the first type of data, call the preset knowledge graph model and the first type of data to perform medical graph inference, generate the first diagnosis suggestion data, and send it to the target terminal.

If the target feature data is the first type of data, the server invokes the preset knowledge graph model and the first type of data to perform medical graph inference, generates the first diagnosis suggestion data and sends it to the target terminal.

104. If the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions, and send them to the target terminal.

Specifically, if the target feature data is the second type of data, the server performs a medical map query according to the preset knowledge map decision tree model and the second type of data, generates multiple rounds of supplementary questions, and sends them to the target terminal.

Specifically, when it is judged that the collected information is incomplete (the second type of data), the server will input the relevant user information, output results and pre-diagnosis and diagnosis results as mixed fields into the medical graph query, and the graph will query the most likely relevant information in the knowledge base. Questions (that is, generating multiple rounds of supplementary questions) are pushed.

Multiple rounds of supplementary questions are organized into im messages and displayed on the target terminal. The supplementary information answered by the target user is extracted and integrated into the electronic medical record module of the medical consultation system through entity information extraction and integration to obtain updated data. The updated data includes basic information such as symptom information, medical history information, and age. There are no restrictions.

105. Generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data.

The server generates electronic medical record data based on answers to multiple rounds of supplementary questions and target feature data.

106. Invoke the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

The server invokes the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

In this embodiment of the present application, the preset attention neural network model is used to judge the completeness of the target feature data, pre-classify the target feature data, and return the pre-classification result to the preset knowledge graph model for analysis, and return it to the target terminal The key questions with the highest ranking are asked, which reduces the online misdiagnosis rate, saves the time of Internet hospital admissions, and improves the efficiency of Internet hospital admissions per unit time. And this solution can be applied in the field of smart medical care, thereby promoting the construction of smart cities.

Please refer to FIG. 2, another flowchart of the question and answer management method of the medical consultation system provided by the embodiment of the present application, specifically including:

201. Acquire target feature data from a target terminal, where the target feature data is used to indicate consultation information input by a target user to a medical consultation system through the target terminal.

Optionally, before step 201, the training process of the neural network pre-classification model may also be included:

The server obtains multiple initial historical medical questionnaires, and performs desensitization processing on the multiple initial historical medical questionnaires to obtain desensitized candidate historical medical questionnaires; the server performs feature extraction on the desensitized candidate historical medical questionnaires, Obtain a plurality of candidate features, the candidate features at least include the dialogue text of the medical questionnaire, the user's age information, the user's gender information, the user's main appeal information, the doctor's diagnosis and treatment prescription information and the diagnosis information; the server determines the multiple candidate features as the preset template model the input data, the marked diagnostic label is determined as the output label of the preset template model, and the preset template model is trained; the server generates a preset neural network pre-classification model, and the neural network pre-classification model is used to perform two Classification.

202. Invoke a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data, and the first type of data is consultation Data with complete information types, and the second type of data is data with missing medical information types.

Specifically, the server invokes a preset neural network pre-classification model to pre-classify the target data, and determines the pre-classification value; the server determines whether the pre-classification value is greater than or equal to the first threshold; if the pre-classification value is greater than or equal to the first threshold, then The server determines that the pre-classification result corresponding to the target feature data is the first type of data, and the first type of data is data with complete types of consultation information; if the pre-classification value is less than the first threshold, the server determines that the pre-classification result corresponding to the target feature data is The second type of data, the second type of data is the data missing the type of consultation information.

Optionally, calling a preset neural network pre-classification model to pre-classify the target data and determining the pre-classification value includes: the server invoking multiple preset encoders to perform fixed-length encoding on the target data, and generating multiple fixed-dimensional vectors , wherein the multiple preset encoders include a plain text encoder, a case history encoder, a user information encoder, a key keyword encoder and a category label encoder, and the fixed-dimensional vectors include user basic information vector, historical information vector and The current inquiry mainly requires information vectors; the server inputs multiple fixed-dimensional vectors into the preset neural network pre-classification model to generate prediction vectors; the server scores the prediction vectors to obtain the pre-classification value of the target data.

203. If the target feature data is the first type of data, perform a pruning operation on the medical knowledge graph in the preset knowledge graph model according to the first type of data, to obtain a pruned medical knowledge graph.

Specifically, if the target feature data is the first type of data, the server determines the question and answer keywords involved in the first type of data; the server determines the corresponding graph node in the medical knowledge graph of the preset knowledge graph model according to the question and answer keywords; The server performs a pruning operation on the medical knowledge graph, and obtains a pruned medical knowledge graph. The pruned medical knowledge graph does not contain corresponding graph nodes.

It should be noted that the pruning here is to delete the question and answer keywords that have been obtained in the target feature data from the medical knowledge graph, and then perform reasoning based on the deleted medical knowledge graph to avoid duplicate data in the medical graph reasoning process. .

204. Perform decision tree analysis on the pruned medical knowledge graph to obtain an analysis result.

The server performs decision tree parsing on the pruned medical knowledge graph to obtain parsing results.

205. Determine the first diagnosis suggestion data based on the analysis result and the preset recommendation relationship table, and send the first diagnosis suggestion data to the target terminal.

The server determines the first diagnosis suggestion data based on the analysis result and the preset recommendation relationship table, and sends the first diagnosis suggestion data to the target terminal. For example, when the analysis result is diabetes, the server calls the preset recommendation relation table, and queries the recommendation relation table to obtain multiple diabetes treatment plans; the server scores the multiple diabetes treatment plans according to the preset scoring rules, and obtains the corresponding multiple treatment plans. The scores are sorted in descending order according to the scores, and a treatment recommendation list is obtained. The treatment recommendation list includes multiple diabetes treatment plans; the server sends the top two diabetes treatment plans in the treatment recommendation list. to the target terminal.

206. If the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions, and send them to the target terminal.

Optionally, before step 206, the construction process of the knowledge graph decision tree model is also included, and the specific process is as follows:

The server divides the preset knowledge graph training data into multiple sample data sets; the server calls the first preset formula to calculate the purity of each data sample set,

Among them, H(X) represents the information entropy of the data sample set, p(x)=pi represents the occurrence probability of random variable X; the server according to the preset formula _g (D,A)=H(D)-H(D|A ) calculate the information gain to obtain multiple feature information gains, where g(D,A) represents the information gain of feature A to sample data set D, H(D) represents the uncertainty of sample data set D, H(D| A) represents the uncertainty of the sample data set D under the given condition A; the server selects the feature information gain with the largest value among multiple feature information gains, and determines it as the target feature information gain; the server uses the ID3 algorithm to generate the target feature information gain according to the target feature information gain. Knowledge graph decision tree model.

It should be noted that H(D|A) can be understood as the degree to which the uncertainty of the classification of the sample data set D is reduced due to the feature A, that is, the feature with large information gain has a stronger classification ability.

It can be understood that, the construction process of the knowledge graph decision tree model can also be performed before step 201 .

207. Generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data.

208. Invoke the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

The question and answer management method of the medical consultation system in the embodiment of the present application has been described above, and the question and answer management device of the medical consultation system in the embodiment of the present application is described below. Please refer to FIG. 3 . An embodiment of the question and answer management device includes:

A data acquisition module 301, configured to acquire target feature data from a target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal;

The discrimination module 302 is used to call a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data. Type data, the first type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information;

The graph reasoning module 303, if the target feature data is the first type of data, is used to call the preset knowledge graph model and the first type of data to perform medical graph reasoning, generate the first diagnosis suggestion data and send it to the target terminal;

The map tree logic module 304, if the target feature data is the second type of data, is used to perform medical map query according to the preset knowledge map decision tree model and the second type of data, and generate multiple rounds of supplementary questions and sent to the target terminal;

an electronic medical record module 305, configured to generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data;

The discriminating module 302 is further configured to call the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

Referring to FIG. 4, another embodiment of the question and answer management device of the medical consultation system in the embodiment of the present application includes:

Optionally, the discriminating module 302 includes:

The pre-classification unit 3021 is used to call the preset neural network pre-classification model to pre-classify the target data, and determine the pre-classification value;

Judging unit 3022, for judging whether the pre-classification value is greater than or equal to the first threshold;

The first determining unit 3023, if the pre-classification value is greater than or equal to the first threshold value, is used to determine that the pre-classification result corresponding to the target feature data is the first type of data, and the first type of data is consultation Data with complete information types;

The second determining unit 3024 is configured to, if the pre-classification value is less than the first threshold, determine that the pre-classification result corresponding to the target feature data is the second type of data, and the second type of data is consultation Information type missing data.

Optionally, the pre-classification unit 3021 is specifically used for:

Invoke multiple preset encoders to perform fixed-length encoding on target data, and generate multiple fixed-dimensional vectors, where multiple preset encoders include plain text encoders, case history encoders, user information encoders, and key keywords An encoder and a category label encoder, the fixed-dimensional vectors include user basic information vectors, historical information vectors, and current consultation main appeal information vectors; input the multiple fixed-dimensional vectors into the preset neural network In the pre-classification model, a prediction vector is generated; the prediction vector is scored to obtain the pre-classification value of the target data.

Optionally, the graph reasoning module 303 includes:

The pruning unit 3031 is configured to perform a pruning operation on the medical knowledge map in the preset knowledge map model according to the first type of data, if the target feature data is the first type of data, to obtain a pruning operation Post-medical knowledge graph;

The parsing unit 3032 is configured to perform decision tree parsing on the pruned medical knowledge graph to obtain parsing results;

A determination sending unit 3033, configured to determine first diagnosis suggestion data based on the analysis result and the preset recommendation relationship table, and send the first diagnosis suggestion data to the target terminal.

Optionally, the pruning unit 3031 is specifically used for:

If the target feature data is the first type of data, determine the question and answer keywords involved in the first type of data; determine the corresponding question and answer keywords in the medical knowledge graph of the preset knowledge graph model according to the question and answer keywords A graph node; perform a pruning operation on the medical knowledge graph to obtain a pruned medical knowledge graph, where the pruned medical knowledge graph does not include the corresponding graph nodes.

Optionally, determine that the sending unit 3033 is specifically used for:

When the analysis result is diabetes, the preset recommendation relationship table is called, and multiple diabetes treatment plans are obtained by querying the recommendation relationship table; the multiple diabetes treatment plans are scored according to the preset scoring rules, and the corresponding multiple treatment plans are obtained. score, and sort according to the multiple scores in descending order to obtain a treatment recommendation list, where the treatment recommendation list includes the multiple diabetes treatment plans; before sorting the treatment recommendation list Two diabetes treatment plans are sent to the target terminal.

Optionally, the question and answer management device of the medical consultation system further includes:

A medical order obtaining module 306, configured to obtain a plurality of initial historical medical questions, and perform desensitization processing on the plurality of initial historical medical surveys to obtain desensitized candidate historical medical surveys;

The feature extraction module 307 is configured to perform feature extraction on the desensitized candidate historical medical questionnaires to obtain a plurality of candidate features, and the candidate features at least include the dialogue text of the medical questionnaires, user age information, user gender information, user The main appeal information, doctor's diagnosis and treatment prescription information and diagnosis information;

A training module 308, configured to determine the multiple candidate features as the input data of the preset template model, determine the marked diagnostic label as the output label of the preset template model, and train the preset template model;

The generating module 309 is configured to generate a preset neural network pre-classification model, where the neural network pre-classification model is used to perform binary classification on the data.

In this embodiment of the present application, the preset attention neural network model is used to judge the completeness of the target feature data, pre-classify the target feature data, and return the pre-classification result to the preset knowledge graph model for analysis, and return it to the target terminal The key questions with the highest ranking are asked, which reduces the online misdiagnosis rate, saves the time for Internet hospital admissions, and improves the efficiency of Internet hospital admissions per unit time. And this solution can be applied in the field of smart medical care, so as to promote the construction of smart city.

Figures 3 to 4 above describe in detail the question and answer management device of the medical consultation system in the embodiment of the present application from the perspective of modular functional entities. The following is a description of the question and answer management device of the medical consultation system in the embodiment of the present application from the perspective of hardware processing. Describe in detail.

5 is a schematic structural diagram of a question-and-answer management device of a medical consultation system provided by an embodiment of the present application. The question-and-answer management device 500 of the medical consultation system may vary greatly due to different configurations or performances, and may include one or more One or more central processing units (CPUs) 510 (eg, one or more processors) and memory 520, one or more storage media 530 (eg, one or more mass storage devices) that store applications 533 or data 532 ). Among them, the memory 520 and the storage medium 530 may be short-term storage or persistent storage. The program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the question-and-answer management device 500 of the medical consultation system. Furthermore, the processor 510 may be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the question-and-answer management device 500 of the medical consultation system.

The question and answer management device 500 of the medical consultation system may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and/or, one or more operating systems 531 , such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that the structure of the question and answer management device of the medical consultation system shown in FIG. 5 does not constitute a limitation on the question and answer management device of the medical consultation system, and may include more or less components than those shown in the figure, or Combining certain components, or different component arrangements. The processor 510 can execute the data acquisition module 301, the discrimination module 302, the atlas inference module 303, the atlas tree logic module 304, the electronic medical record module 305, the medical order acquisition module 306, the feature extraction module 307, the training module 308 and The function of the generation module 309 .

The present application also provides a question-and-answer management device for a medical consultation system, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected through a line; the at least one processor A processor invokes the instructions in the memory, so that the question and answer management device of the medical consultation system executes the steps in the question and answer management method of the medical consultation system.

The present application also provides a computer-readable storage medium, and the computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores computer instructions, which, when executed on the computer, cause the computer to perform the following steps:

Obtain target feature data from the target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal;

Invoke a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data. One type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information;

If the target feature data is the first type of data, call the preset knowledge graph model and the first type of data to perform medical graph inference, generate first diagnosis suggestion data, and send it to the target terminal;

If the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions, and send them to the target terminal;

generating electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data;

The neural network pre-classification model is invoked to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above may refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, and the computer software products are stored in a storage medium , including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes .

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand: The technical solutions recorded in the embodiments are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions in the embodiments of the present application.

Claims

A question and answer management method for a medical consultation system, comprising:

Obtain target feature data from the target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal;

Invoke a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data. One type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information;

If the target feature data is the first type of data, call the preset knowledge graph model and the first type of data to perform medical graph inference, generate first diagnosis suggestion data, and send it to the target terminal;

If the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions, and send them to the target terminal;

generating electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data;

The neural network pre-classification model is invoked to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.
The question and answer management method of a medical consultation system according to claim 1, wherein the target feature data is pre-classified by invoking a preset neural network pre-classification model, and a pre-classification result corresponding to the target feature data is determined , the pre-classification result includes a first type of data and a second type of data, the first type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information, including:

Calling the preset neural network pre-classification model to pre-classify the target data, and determine the pre-classification value;

judging whether the pre-classification value is greater than or equal to a first threshold;

If the pre-classification value is greater than or equal to the first threshold, determining that the pre-classification result corresponding to the target feature data is the first type of data, and the first type of data is data with complete types of consultation information;

If the pre-classification value is less than the first threshold, it is determined that the pre-classification result corresponding to the target feature data is the second type of data, and the second type of data is data missing the type of consultation information.
The question-and-answer management method of a medical consultation system according to claim 2, wherein the pre-classification of the target data by invoking a preset neural network pre-classification model to determine a pre-classification value comprises:

Invoke multiple preset encoders to perform fixed-length encoding on target data, and generate multiple fixed-dimensional vectors, where multiple preset encoders include plain text encoders, case history encoders, user information encoders, and key keywords An encoder and a category label encoder, wherein the fixed-dimensional vector includes a user basic information vector, a historical information vector, and an information vector of current main appeals for consultation;

Inputting the vectors of the plurality of fixed dimensions into the preset neural network pre-classification model to generate a prediction vector;

The prediction vector is scored to obtain the pre-classification value of the target data.
The question-and-answer management method of a medical consultation system according to claim 1, wherein, if the target feature data is the first type of data, a preset knowledge graph model and the first type of data are called to carry out Medical graph reasoning, generating first diagnosis suggestion data and sending it to the target terminal, including:

If the target feature data is the first type of data, perform a pruning operation on the medical knowledge graph in the preset knowledge graph model according to the first type of data, to obtain a pruned medical knowledge graph;

Performing decision tree analysis on the pruned medical knowledge graph to obtain an analysis result;

Determine the first diagnosis suggestion data based on the analysis result and the preset recommendation relationship table, and send the first diagnosis suggestion data to the target terminal.
The question-and-answer management method of a medical consultation system according to claim 4, wherein, if the target feature data is the first type of data, the preset knowledge graph is analyzed according to the first type of data. The medical knowledge graph in the model is pruned, and the pruned medical knowledge graph is obtained, including:

If the target feature data is the first type of data, determine the question and answer keywords involved in the first type of data;

Determine the corresponding graph node in the medical knowledge graph of the preset knowledge graph model according to the question and answer keywords;

A pruning operation is performed on the medical knowledge graph to obtain a pruned medical knowledge graph, where the pruned medical knowledge graph does not include the corresponding graph nodes.
The question and answer management method of a medical consultation system according to claim 4, wherein the first diagnosis suggestion data is determined based on the analysis result and the preset recommendation relation table, and the first diagnosis suggestion data is sent to The target terminal includes:

When the analysis result is diabetes, the preset recommendation relationship table is called, and a plurality of diabetes treatment plans are obtained by querying the recommendation relationship table;

Score the multiple diabetes treatment plans according to the preset scoring rules, obtain multiple corresponding scores, and sort them in descending order according to the multiple scores to obtain a treatment recommendation list. The recommended list includes the plurality of diabetes treatment options;

Send the top two diabetes treatment plans in the treatment recommendation list to the target terminal.
The question and answer management method of the medical consultation system according to any one of claims 1-6, wherein, before the acquisition of the target feature data from the target terminal, the question and answer management method of the medical consultation system further comprises:

Obtaining a plurality of initial historical medical questionnaires, and performing desensitization processing on the plurality of initial historical medical questionnaires to obtain desensitized candidate historical medical questionnaires;

Perform feature extraction on the desensitized candidate historical medical questionnaire to obtain multiple candidate features, the candidate features at least include the dialogue text of the medical questionnaire, the user's age information, the user's gender information, the user's main appeal information, and the doctor's diagnosis and treatment. Prescribing and diagnostic information;

Determining the plurality of candidate features as input data of the preset template model, determining the marked diagnostic label as the output label of the preset template model, and training the preset template model;

A preset neural network pre-classification model is generated, and the neural network pre-classification model is used for binary classification of data.
A question and answer management device for a medical consultation system, comprising a memory, a processor, and computer-readable instructions stored on the memory and executable on the processor, when the processor executes the computer-readable instructions Implement the following steps:

Obtain target feature data from the target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal;

Invoke a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data. One type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information;

If the target feature data is the first type of data, call the preset knowledge graph model and the first type of data to perform medical graph inference, generate first diagnosis suggestion data, and send it to the target terminal;

If the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions, and send them to the target terminal;

generating electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data;

The neural network pre-classification model is invoked to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.
The question-and-answer management device of a medical consultation system according to claim 8, wherein the processor further implements the following steps when executing the computer program:

Calling the preset neural network pre-classification model to pre-classify the target data, and determine the pre-classification value;

judging whether the pre-classification value is greater than or equal to a first threshold;

If the pre-classification value is greater than or equal to the first threshold, determining that the pre-classification result corresponding to the target feature data is the first type of data, and the first type of data is data with complete types of consultation information;

If the pre-classification value is less than the first threshold, it is determined that the pre-classification result corresponding to the target feature data is the second type of data, and the second type of data is data with missing medical information types.
The question-and-answer management device of a medical consultation system according to claim 9, wherein the processor further implements the following steps when executing the computer program:

Invoke multiple preset encoders to perform fixed-length encoding on target data, and generate multiple fixed-dimensional vectors, where multiple preset encoders include plain text encoders, case history encoders, user information encoders, and key keywords An encoder and a category label encoder, wherein the fixed-dimensional vector includes a user basic information vector, a historical information vector, and an information vector of current main appeals for consultation;

Inputting the vectors of the plurality of fixed dimensions into the preset neural network pre-classification model to generate a prediction vector;

The prediction vector is scored to obtain the pre-classification value of the target data.
The question-and-answer management device of a medical consultation system according to claim 8, wherein the processor further implements the following steps when executing the computer program:

If the target feature data is the first type of data, perform a pruning operation on the medical knowledge graph in the preset knowledge graph model according to the first type of data, to obtain a pruned medical knowledge graph;

Performing decision tree analysis on the pruned medical knowledge graph to obtain an analysis result;

Determine the first diagnosis suggestion data based on the analysis result and the preset recommendation relationship table, and send the first diagnosis suggestion data to the target terminal.
The question and answer management device of the medical consultation system according to claim 11, wherein the processor further implements the following steps when executing the computer program:

If the target feature data is the first type of data, determine the question and answer keywords involved in the first type of data;

Determine the corresponding graph node in the medical knowledge graph of the preset knowledge graph model according to the question and answer keywords;

A pruning operation is performed on the medical knowledge graph to obtain a pruned medical knowledge graph, where the pruned medical knowledge graph does not include the corresponding graph nodes.
The question and answer management device of the medical consultation system according to claim 11, wherein the processor further implements the following steps when executing the computer program:

When the analysis result is diabetes, the preset recommendation relationship table is called, and a plurality of diabetes treatment plans are obtained by querying the recommendation relationship table;

Score the multiple diabetes treatment plans according to the preset scoring rules, obtain multiple corresponding scores, and sort them in descending order according to the multiple scores to obtain a treatment recommendation list. The recommended list includes the plurality of diabetes treatment options;

Send the top two diabetes treatment plans in the treatment recommendation list to the target terminal.
According to the question and answer management device of the medical consultation system according to any one of claims 8-13, the processor further implements the following steps when executing the computer program:

Obtaining a plurality of initial historical medical questionnaires, and performing desensitization processing on the plurality of initial historical medical questionnaires to obtain desensitized candidate historical medical questionnaires;

Perform feature extraction on the desensitized candidate historical medical questionnaire to obtain multiple candidate features, the candidate features at least include the dialogue text of the medical questionnaire, the user's age information, the user's gender information, the user's main appeal information, and the doctor's diagnosis and treatment. Prescribing and diagnostic information;

Determining the plurality of candidate features as input data of the preset template model, determining the marked diagnostic label as the output label of the preset template model, and training the preset template model;

A preset neural network pre-classification model is generated, and the neural network pre-classification model is used for binary classification of data.
A computer-readable storage medium, storing computer instructions in the computer-readable storage medium, when the computer instructions are executed on a computer, the computer is made to perform the following steps:

Obtain target characteristic data from the target terminal, and the target characteristic data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal;

Invoke a preset neural network pre-classification model to pre-classify the target feature data, and determine a pre-classification result corresponding to the target feature data, where the pre-classification result includes the first type of data and the second type of data. One type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information;

If the target feature data is the first type of data, call the preset knowledge graph model and the first type of data to perform medical graph inference, generate first diagnosis suggestion data, and send it to the target terminal;

If the target feature data is the second type of data, perform medical map query according to the preset knowledge map decision tree model and the second type of data, generate multiple rounds of supplementary questions, and send them to the target terminal;

generating electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data;

The neural network pre-classification model is invoked to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.
The computer-readable storage medium of claim 15, when the computer instructions are executed on a computer, causing the computer to further perform the following steps:

Calling the preset neural network pre-classification model to pre-classify the target data, and determine the pre-classification value;

judging whether the pre-classification value is greater than or equal to a first threshold;

If the pre-classification value is greater than or equal to the first threshold, determining that the pre-classification result corresponding to the target feature data is the first type of data, and the first type of data is data with complete types of consultation information;

If the pre-classification value is less than the first threshold, it is determined that the pre-classification result corresponding to the target feature data is the second type of data, and the second type of data is data missing the type of consultation information.
The computer-readable storage medium of claim 16, when the computer instructions are executed on a computer, causing the computer to further perform the following steps:

Invoke multiple preset encoders to perform fixed-length encoding on target data, and generate multiple fixed-dimensional vectors, where multiple preset encoders include plain text encoders, case history encoders, user information encoders, and key keywords An encoder and a category label encoder, wherein the fixed-dimensional vector includes a user basic information vector, a historical information vector, and an information vector of current main appeals for consultation;

Inputting the vectors of the plurality of fixed dimensions into the preset neural network pre-classification model to generate a prediction vector;

The prediction vector is scored to obtain the pre-classification value of the target data.
The computer-readable storage medium of claim 15, when the computer instructions are executed on a computer, causing the computer to further perform the following steps:

If the target feature data is the first type of data, perform a pruning operation on the medical knowledge graph in the preset knowledge graph model according to the first type of data, to obtain a pruned medical knowledge graph;

Performing decision tree analysis on the pruned medical knowledge graph to obtain an analysis result;

Determine the first diagnosis suggestion data based on the analysis result and the preset recommendation relationship table, and send the first diagnosis suggestion data to the target terminal.
The computer-readable storage medium of claim 18, the computer instructions, when executed on a computer, cause the computer to further perform the following steps:

If the target feature data is the first type of data, determine the question and answer keywords involved in the first type of data;

Determine the corresponding graph node in the medical knowledge graph of the preset knowledge graph model according to the question and answer keywords;

A pruning operation is performed on the medical knowledge graph to obtain a pruned medical knowledge graph, where the pruned medical knowledge graph does not include the corresponding graph nodes.
A question and answer management device of a medical consultation system, the question and answer management device of the medical consultation system includes:

a data acquisition module, configured to acquire target feature data from the target terminal, where the target feature data is used to indicate the consultation information input by the target user to the medical consultation system through the target terminal;

The discrimination module is used to call the preset neural network pre-classification model to pre-classify the target feature data, and determine the pre-classification result corresponding to the target feature data, and the pre-classification result includes the first type of data and the second type of data Data, the first type of data is data with complete types of consultation information, and the second type of data is data with missing types of consultation information;

The graph reasoning module, if the target feature data is the first type of data, is used to call the preset knowledge graph model and the first type of data to perform medical graph reasoning, generate the first diagnosis suggestion data and send it to the the target terminal;

The map tree logic module, if the target feature data is the second type of data, is used to perform medical map query according to the preset knowledge map decision tree model and the second type of data, and generate multiple rounds of supplementary questions. sent to the target terminal;

an electronic medical record module, configured to generate electronic medical record data according to the answers to the multiple rounds of supplementary questions and the target feature data;

The discrimination module is further configured to call the neural network pre-classification model to pre-classify the electronic medical record data again until the second diagnosis suggestion data is generated and sent to the target terminal.