WO2021151356A1 - 分诊数据处理方法、装置、计算机设备及存储介质 - Google Patents

分诊数据处理方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2021151356A1
WO2021151356A1 PCT/CN2020/135341 CN2020135341W WO2021151356A1 WO 2021151356 A1 WO2021151356 A1 WO 2021151356A1 CN 2020135341 W CN2020135341 W CN 2020135341W WO 2021151356 A1 WO2021151356 A1 WO 2021151356A1
Authority
WO
WIPO (PCT)
Prior art keywords
symptom
patient
learning model
information
reinforcement learning
Prior art date
Application number
PCT/CN2020/135341
Other languages
English (en)
French (fr)
Inventor
刘卓
李彦轩
孙行智
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021151356A1 publication Critical patent/WO2021151356A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, and storage medium for processing triage data.
  • the inventor found that when a patient goes to the hospital for treatment, he first needs to go to the triage table for manual triage. In this process, the patient needs to spend a lot of time in queuing, and the depth and breadth of the professional knowledge of the guides at the triage table is limited. Higher requirements, if the guides give the patient a wrong diagnosis and need to perform the triage again, it will greatly waste the patient’s time and seriously affect the patient’s experience. Therefore, in the prior art, the manual triage of the patient takes a long time. , It is difficult to give a reasonable medical department or doctor, resulting in poor patient experience and low medical accuracy.
  • This application provides a triage data processing method, device, computer equipment and storage medium, which realizes that the hierarchical reinforcement learning model can be used to inquire about the relevant symptoms of the patient, and useful patient symptom information can be extracted, and then the department triage model can be used for useful patients Symptom information is used for symptom feature identification.
  • This application is suitable for smart medical and other fields, which can further promote the construction of smart cities. It can quickly and accurately determine the departments that patients need to see, improve the accuracy of medical treatment and improve the patient experience.
  • a method for processing triage data including:
  • the patient's symptom information is input into a hierarchical reinforcement learning model, the patient's symptom information is recognized by the human body system through the upper learning model, and the first human body system category corresponding to the patient's symptom information is identified;
  • the hierarchical reinforcement learning model includes The upper-level learning model and multiple lower-level reinforcement learning models;
  • a lower-level reinforcement learning model is associated with a human body system category;
  • the lower level associated with the identified first body system category is obtained from the hierarchical reinforcement learning model Reinforcement learning model;
  • the action result is an optimal scheduling action determined for the patient's symptom information
  • the patient's symptom information is input into the department triage model, and the patient's symptom information is identified through the department triage model to identify the symptoms corresponding to the patient Triage results.
  • a triage data processing device including:
  • the receiving module is used to receive the patient's request from the patient and obtain the patient's symptom information in the patient's request;
  • the recognition module is used to input the patient's symptom information into a hierarchical reinforcement learning model, recognize the patient's symptom information through the upper-layer learning model, and identify the first body system category corresponding to the patient's symptom information;
  • the hierarchical reinforcement learning model includes the upper learning model and multiple lower reinforcement learning models; a lower reinforcement learning model is associated with a human body system category;
  • An acquiring module configured to acquire the lower-level reinforcement learning model associated with the identified first human body system category from the hierarchical reinforcement learning model
  • a prediction module configured to predict the patient's symptom information and obtain an action result through the acquired lower-level reinforcement learning model; the action result is an optimal scheduling action determined for the patient's symptom information;
  • the triage module is used to input the patient's symptom information into the department triage model when the action result is a recommended department action, and perform symptom feature recognition on the patient's symptom information through the department triage model to identify The triage result corresponding to the patient.
  • a computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, and the processor implements the following steps when the processor executes the computer-readable instructions:
  • the patient's symptom information is input into a hierarchical reinforcement learning model, the patient's symptom information is recognized by the human body system through the upper learning model, and the first human body system category corresponding to the patient's symptom information is identified;
  • the hierarchical reinforcement learning model includes The upper-layer learning model and multiple lower-layer reinforcement learning models;
  • a lower-layer reinforcement learning model is associated with a human body system category;
  • the action result is an optimal scheduling action determined for the patient's symptom information
  • the patient's symptom information is input into the department triage model, and the patient's symptom information is identified through the department triage model to identify the symptoms corresponding to the patient Triage results.
  • One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps:
  • the patient's symptom information is input into a hierarchical reinforcement learning model, the patient's symptom information is recognized by the human body system through the upper learning model, and the first human body system category corresponding to the patient's symptom information is identified;
  • the hierarchical reinforcement learning model includes The upper-layer learning model and multiple lower-layer reinforcement learning models;
  • a lower-layer reinforcement learning model is associated with a human body system category;
  • the action result is an optimal scheduling action determined for the patient's symptom information
  • the patient's symptom information is input into the department triage model, and the patient's symptom information is identified through the department triage model to identify the symptoms corresponding to the patient Triage results.
  • the triage data processing method, device, computer equipment, and storage medium provided in this application obtain the patient's symptom information in the patient's request by receiving the patient's request; input the patient's symptom information into the hierarchical reinforcement learning model, and pass The upper learning model performs body system recognition on the patient's symptom information, and identifies the first body system category corresponding to the patient's symptom information;
  • the hierarchical reinforcement learning model includes the upper learning model and multiple lower reinforcement learning models;
  • a lower-level reinforcement learning model is associated with a human body system category; the lower-level reinforcement learning model associated with the identified first body system category is obtained from the hierarchical reinforcement learning model; the lower-level reinforcement learning model is obtained through Predict the patient's symptom information and obtain the action result; the action result is the optimal scheduling action determined for the patient's symptom information; when the action result is a recommended department action, input the patient's symptom information into the department
  • the patient’s symptom information is identified through the department’s triage model to identify the corresponding tri
  • the patient’s symptom information can be obtained through hierarchical enhancement.
  • the upper learning model in the learning model performs body system recognition on the patient's symptom information, and identifies the first body system category corresponding to the patient's symptom information; and obtains the first body system from the hierarchical reinforcement learning model
  • the lower-level reinforcement learning model associated with the category predict the patient’s symptom information through the acquired lower-level reinforcement learning model and obtain the action result; when the action result is a recommended department action, pass the department triage model Perform symptom feature identification on the patient’s symptom information, and identify the triage results corresponding to the patient.
  • the diagnosis model recognizes the symptom characteristics of useful patient symptom information, and can quickly and accurately determine the department that the patient needs to see, which improves the accuracy of the diagnosis and improves the patient experience.
  • Fig. 1 is a schematic diagram of an application environment of a triage data processing method in an embodiment of the present application
  • FIG. 2 is a flowchart of a method for processing triage data in an embodiment of the present application
  • step S20 of the triage data processing method in an embodiment of the present application
  • step S30 of the triage data processing method in an embodiment of the present application is a flowchart of step S30 of the triage data processing method in an embodiment of the present application
  • step S10 of the triage data processing method in an embodiment of the present application
  • Fig. 6 is a functional block diagram of a triage data processing device in an embodiment of the present application.
  • Fig. 7 is a schematic diagram of a computer device in an embodiment of the present application.
  • the triage data processing method provided in this application can be applied in the application environment as shown in Fig. 1, in which the client (computer equipment) communicates with the server through the network.
  • the client includes, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, cameras, and portable wearable devices.
  • the server can be implemented as an independent server or a server cluster composed of multiple servers.
  • a method for processing triage data is provided, and the technical solution mainly includes the following steps S10-S50:
  • S10 Receive a patient request from the patient, and obtain patient symptom information in the patient request.
  • the patient symptom information is information related to the patient's symptoms entered by the patient, and the patient can complete the input of the patient's medical information on the application platform, or all symptom descriptions provided on the application platform are concentrated Select the relevant symptom description as the patient's symptom information, or enter a voice or text description of the current symptom on the application platform, identify the keywords in the voice or text description, and use the recognized keywords as The patient's symptom information and so on, thereby triggering the patient's request.
  • the method before the step S10, before the obtaining the patient symptom information in the patient request, the method includes:
  • S101 Acquire patient input information input by the patient.
  • the patient input information is text information input by the patient on the application platform.
  • the preprocessing model is a neural network model that is based on the BERT model and is trained.
  • the preprocessing model can recognize keywords in the patient input information, and can predict all keywords corresponding to each keyword.
  • the process of keyword recognition is to use the BERT algorithm in the preprocessing model to split the patient input information and convert the word vector. By predicting the converted word vector, the corresponding word vector is predicted.
  • Corresponding to the symptom-related keywords and can calculate the similarity with the keyword, so as to obtain the predicted value of each identified keyword.
  • the BERT (Bidirectional Encoder Representations from Transformers) algorithm is to jointly adjust all the keywords. Algorithm to identify the bidirectional Transformer in the layer.
  • the keyword result is a keyword recognized from the input information of the patient, the keyword result includes a plurality of the recognized keywords, and the keywords are symptom-related and can effectively embody Symptom words.
  • S103 Determine the keyword corresponding to the predicted value reaching a preset threshold as the patient's symptom information, and trigger the patient's request.
  • the preset threshold is a preset probability value, and the preset threshold can be set according to requirements. From the keyword results, the predicted value greater than the preset threshold corresponds to The keywords are marked as the patient's symptom information, so that useful words are extracted from the patient's input information, and the terms of the patient's symptom information are unified, which facilitates the recognition of the follow-up hierarchical reinforcement learning model and the department triage model, and improves The recognition efficiency is improved, and the recognition accuracy and reliability are improved.
  • This application realizes that by obtaining patient input information input by the patient; performing keyword recognition on the patient input information through the preprocessing model to obtain keyword results;
  • the keywords are determined as the patient's symptom information and trigger the patient's request.
  • the keywords in the patient's input information are extracted through the preprocessing model, which can unify the terms of the patient's symptom information and facilitate subsequent level enhancement
  • the recognition of the learning model and the department triage model improves the recognition efficiency, and improves the accuracy and reliability of the recognition.
  • S20 Input the patient symptom information into a hierarchical reinforcement learning model, and perform human body system recognition on the patient symptom information through the upper learning model, and identify the first human body system category corresponding to the patient symptom information; the hierarchical reinforcement learning
  • the model includes the upper-layer learning model and a plurality of lower-layer reinforcement learning models; a lower-layer reinforcement learning model is associated with a human body system category.
  • the hierarchical reinforcement learning model includes the upper learning model and multiple lower reinforcement learning models
  • the upper learning model is a trained network model
  • the upper learning model can perform human body system recognition on the input information , Recognizing which type of human body system category the input information belongs to, the network structure of the upper learning model can be set according to requirements, for example, the network structure of the upper learning model is a deep convolutional neural network model, a reinforcement learning model, etc.
  • the human body system recognition is to extract the characteristics of the patient’s symptoms according to the upper learning model, and to identify the human body system category according to the characteristics of the human body system, and the human body system characteristics are For features related to the human body system category, the patient’s symptom information is recognized by the upper learning model through the human body system, and the first human body system category corresponding to the patient’s symptom information is identified, and the first human body system category is based on The human body system category identified by the patient's symptom information, wherein the human body system category includes the motor system, the nervous system, the digestive system, the urinary system, the reproductive system, the respiratory system, the circulatory system, and the endocrine system.
  • the network structure of the upper learning model is a reinforcement learning model
  • the patient’s symptom information is taken as the current state through the upper learning model
  • the agent in the upper learning model is based on the current state.
  • the reward value is to identify the first human body system category according to the reward value corresponding to the best action.
  • the method before step S20, that is, before inputting the patient's symptom information into a hierarchical reinforcement learning model, the method includes:
  • S201 Obtain a symptom sample set; the symptom sample set includes a plurality of symptom samples, and the symptom samples are associated with a human body system category label.
  • the symptom sample set is a collection of all the symptom samples
  • the symptom samples are symptom words collected in history
  • the symptom samples include several symptom words, one symptom sample and one human system
  • the category label is associated
  • the human body system category label is the human body system category manually labeled for the symptom sample.
  • S202 Input the symptom sample into a triage neural network model containing first initial parameters.
  • the triage neural network model includes the first initial parameters, and the network structure of the triage neural network model can be set according to requirements, for example, the network structure of the triage neural network model is a Word2vec model , Deep convolutional neural network model, etc., the first initial parameter can be directly transferred from other trained neural network models through transfer learning.
  • S203 Perform body system recognition on the symptom sample through the triage neural network model, and obtain a sample recognition result of the body system category corresponding to the symptom sample.
  • the process of recognition by the human body system may be by performing word vector conversion on the symptom sample, that is, converting the symptom sample into a word vector through a word vector dictionary, and concatenating all the converted word vectors into text Feature vector, by convolving the text feature vector, that is, extracting the features of the human body system, and performing full connection recognition on the text feature vector after the convolution, thereby identifying the body system category corresponding to the symptom sample
  • the sample recognition result, the sample recognition result is to identify which type of human body system category the symptom sample belongs to.
  • the sample recognition result and the human body system category label associated with the symptom sample are input into the loss function in the triage neural network model, and the loss value is calculated by the loss function,
  • the loss value indicates the difference between the sample recognition result and the human body system category label, and the loss function can be set according to requirements, such as a cross-entropy loss function.
  • the convergence condition may be a condition that the value of the loss value is very small and will not drop after 5000 calculations, that is, the value of the loss value is very small and will not drop after 5000 calculations.
  • the convergence condition can also be the condition that the loss value is less than the set threshold, that is, when the loss value is less than When the threshold is set, the training is stopped, and the triage neural network model after convergence is recorded as the upper learning model, so that when the loss value does not reach the preset convergence condition, the iterative volume is continuously updated Integrating the initial parameters of the neural network model, and triggering the step of recognizing the symptom sample by the human body system through the triage neural network model to obtain the sample recognition result of the human body system category corresponding to the symptom sample may be Constantly moving closer to accurate results, so that the accuracy of recognition is getting higher and higher.
  • a lower-level reinforcement learning model is associated with a human system category, which includes the motor system, nervous system, digestive system, urinary system, reproductive system, respiratory system, circulatory system, and endocrine system.
  • the lower-level reinforcement Learning models include the lower-level reinforcement learning model of the motor system, the lower-level reinforcement learning model of the nervous system, the lower-level reinforcement learning model of the digestive system, the lower-level reinforcement learning model of the urinary system, the lower-level reinforcement learning model of the reproductive system, the lower-level reinforcement learning model of the respiratory system, and the lower-level reinforcement learning of the circulatory system.
  • each of the upper-level learning models is obtained after reinforcement learning is performed through the samples of the human body system category associated with the upper-level learning model and training is completed
  • the hierarchical reinforcement learning model includes A plurality of the lower reinforcement learning models
  • the lower reinforcement learning model associated with the first human body system category is obtained from the hierarchical reinforcement learning model
  • the human body associated with the lower reinforcement learning model is learned during the training process
  • the mapping of the symptom state of the system category to the action space enables the agent to obtain the maximum reward for the action selected by the agent, and makes the evaluation of the lower-level reinforcement learning model in a certain sense (or the running performance of the entire model) the best. And it can guide the inquiring action for better or greater rewards after the optimal action is taken for the current symptom state.
  • step S30 that is, before the lower-level reinforcement learning model associated with the recognized first body system category is obtained from the hierarchical reinforcement learning model ,include:
  • S301 Obtain a symptom state sample set; the symptom state sample set includes a plurality of symptom state samples, the symptom state samples are associated with a department label, and all the symptom state samples are associated with the same body system category.
  • the symptom state sample set is a set of symptom state samples, and the symptom state samples are historically collected words that reflect the symptom state corresponding to the same human system category, and the symptom state samples It may be the same as the symptom sample or different from the symptom sample.
  • One symptom state sample includes several words of the symptom state, and one symptom state sample is associated with one department label, so The department label is a label that reflects the category of the department, and the department label is a label that is manually labeled according to the symptom state sample associated with it or is determined after a doctor visit, and all the symptom state samples are the same as the same one of the human body system categories.
  • the symptom state samples in the symptom state sample set are all words of the symptom state embodied in a human system category.
  • S302 Input the symptom state sample into an initial reinforcement learning model that is associated with the human body system category and contains a second initial parameter.
  • the initial reinforcement learning model is a network model trained through reinforcement learning, and the initial reinforcement learning model includes the second initial parameters.
  • the initial reinforcement learning model matches the corresponding action space according to the words that reflect the previous symptom state in the symptom state sample, and the action space is a set of actions taken in response to the symptom state sample. Performing the actions in the action space will predict the possibility of the next state of the symptom state sample.
  • S304 Execute the action space to obtain a state transition result; the state transition result includes department results and state results.
  • state prediction is carried out by means of reinforcement learning.
  • the reinforcement learning also known as reinforcement learning, evaluation learning, or reinforcement learning, is used to describe and solve the problem between the agent and the environment.
  • reinforcement learning In the interactive process, learning strategies are used to maximize returns or achieve specific goals.
  • the state prediction is an action that executes the action space. Actions are taken according to the probability of each action, and the action with the highest predicted value is selected, thereby The best probability of the next state is predicted, and the state transition result is determined according to the predicted next state.
  • the state transition result is the next step predicted after the symptom state sample executes the action space Words that reflect the state of symptoms, and the probability of all departments in descending order, the state transition results include the results of the departments and the results of the state, the results of the departments predict the probability of all departments in descending order.
  • the result of small order sorting, and the result of sorting the probability of all departments in descending order of the probability of the last prediction, the state result is a set of words that are predicted to reflect the symptom state in the next step.
  • S305 Determine a reward value according to the symptom state sample, the state transition result, and the department label.
  • the state transition result obtained after executing the action space, and the department label associated with the symptom state sample through the reward function in the initial reinforcement learning model, The reward value corresponding to the state transition result is calculated, and the reward value indicates the reward evaluation given after the execution of the action space.
  • step S305 that is, the determining the reward value according to the state transition result and the department label, includes:
  • R s is the reward value
  • ⁇ 1 is the weight of the return value
  • IF(s t ⁇ S u ) is the return value of whether the state result is in the symptom state sample, the state result is 1 in the symptom state sample, and the state result is not in the symptom state sample, -1 is returned;
  • s t is the state result of the t-th state prediction
  • S u is the symptom state sample
  • ⁇ 2 is the weight of the prediction score
  • p t-1 is the sequence value of the department label associated with the symptom state sample in the department sequence corresponding to the department result obtained by the t-1 state prediction;
  • p t is the sequence value of the department label associated with the symptom state sample in the department sequence corresponding to the department result obtained by the t-th state prediction;
  • ⁇ 3 is the weight of the accurate reward value
  • r t is the exact reward value corresponding to p t.
  • r t is an accurate reward value corresponding to p t .
  • the r t is the preset maximum reward value, and the maximum reward value indicates that the predicted department result reaches the best result, such as 100 or 200, etc. ;
  • r t is zero, indicating that no reward is provided when p t does not reach 1.
  • the reward convergence condition can be that training is stopped when p t is 1, or it can be that after t times of state prediction, the reward value reaches the maximum and does not change again, so that training is stopped.
  • the value does not reach the preset reward convergence condition, iteratively update the second initial parameter of the initial reinforcement learning model, and trigger the step of matching the action space corresponding to the symptom state sample through the initial reinforcement learning model , Until the reward value reaches the preset reward convergence condition, stop training, and record the initial reinforcement learning model after convergence as a lower reinforcement learning model.
  • S40 Predict the patient's symptom information through the acquired lower-level reinforcement learning model and obtain an action result; the action result is an optimal scheduling action determined for the patient's symptom information.
  • the patient's symptom information is predicted through the acquired lower-level reinforcement learning model, that is, state prediction, so as to predict the action result, and the action result is the most determined for the patient's symptom information.
  • the result of the optimal scheduling action that is, the action that produces the best value (the maximum reward) selected in the predicted action space, and the result determined after the action is executed, the lower-level reinforcement learning model completed through training can pass
  • the patient’s symptom information can match a corresponding action that can achieve the maximum reward, and the result after the action is executed.
  • the action result includes the recommended department action and the query action, that is, the next action is determined after the action is performed.
  • the one-step action is a recommended department action or an inquiry action.
  • the action result is the recommended department action
  • the recommended department action it means that the recommended department action is reached, which means that the current patient symptom information can accurately and effectively reflect the current patient’s character, and no further action is required
  • Supplement the patient's symptom information input the patient's symptom information into a department triage model, and perform symptom feature recognition on the patient's symptom information through the department triage model, and the symptom feature is identified as identifying the patient's symptom information Perform word vector conversion, and then splice the converted word vectors, so as to extract symptom features from the spliced vector.
  • the symptom features are the implicit vector characteristics between the symptoms and the departments, and the symptom features are identified
  • the triage result corresponding to the patient, the triage result is the predicted highest probability triage category, the triage result includes the department category, that is, the category of the department provided to the patient for treatment, the triage result It provides an accurate basis for the patient to make an appointment, and it is convenient for the patient to choose an accurate department to make an appointment.
  • This application realizes that by receiving a patient request from a patient, obtaining patient symptom information in the patient request; inputting the patient symptom information into a hierarchical reinforcement learning model, and performing human body system recognition on the patient symptom information through the upper learning model, Identify the first human body system category corresponding to the patient's symptom information; the hierarchical reinforcement learning model includes the upper learning model and multiple lower reinforcement learning models; a lower reinforcement learning model is associated with a human body system category; Acquiring, from the hierarchical reinforcement learning model, the lower reinforcement learning model associated with the identified first human body system category; predicting the patient's symptom information and acquiring the action result through the acquired lower reinforcement learning model; The action result is the optimal scheduling action determined for the patient's symptom information; when the action result is a recommended department action, the patient's symptom information is input into the department triage model, and the department is used to compare the patient's symptom information.
  • the patient’s symptom information is used for symptom feature identification, and the triage results corresponding to the patient are identified.
  • the patient’s symptom information is acquired;
  • the upper learning model in the hierarchical reinforcement learning model is used to perform the symptom information Human body system recognition, identifying the first body system category corresponding to the patient’s symptom information; acquiring the lower-level reinforcement learning model associated with the first body system category from the hierarchical reinforcement learning model;
  • the lower-level reinforcement learning model predicts the patient's symptom information and obtains the action result; when the action result is a recommended department action, the patient's symptom information is identified through the department triage model to identify the symptoms and
  • the hierarchical reinforcement learning model can be used to inquire about the relevant symptoms of the patient, and useful patient symptom information can be extracted, and then the useful patient symptom information can be identified by the department triage model. Quickly and accurately determine the department that the patient needs to see, which improves the accuracy of medical treatment and improves the patient experience.
  • the method further includes:
  • S60 When the result of the action is an inquiry action, send out a new round of symptom inquiry information in the inquiry action, receive response information of the patient's response to the new round of symptom inquiry information, and update all information according to the response information. State the patient’s symptom information.
  • the inquiry action includes the new round of symptom inquiry information, and the new A round of symptom inquiry information is information that asks questions raised by determining the next action strategy based on the patient's symptom information and the state information after predicting the patient's symptom information.
  • the patient asks questions that help supplement complete patient symptom information, and after receiving the new round of symptom questioning information, the patient makes the response information in response to the new round of symptom questioning information, According to the received response information, it is added to the original patient symptom information, that is, the response information is added after the original patient symptom information, so as to complete the update of the patient symptom information.
  • the response information can be Symptom words used to extract the content of the patient's answer to the new round of symptom inquiry information.
  • the updated patient symptom information is input into the upper learning model, and the human body system is identified as extracting the characteristics of the patient's symptom information according to the upper learning model.
  • the human body system feature is a feature related to the human body system category
  • the updated patient symptom information is recognized by the human body system through the upper learning model to identify and update
  • the second body system category corresponding to the latter patient’s symptom information, the second body system category being the body system category identified according to the updated patient’s symptom information, and the second body system category may follow the
  • the first body system category is the same, or may be different from the first body system category, because some symptom words (that is, symptom information and symptom status in the full text) appear in multiple body system categories.
  • the lower-layer reinforcement learning model associated with the second human body system category is obtained from the hierarchical reinforcement learning model, and the symptom state of the human body system category associated with the lower-layer reinforcement learning model is learned during the training process
  • the mapping to the action space enables the agent to obtain the maximum reward for the action selected by the agent, so that the evaluation of the lower-level reinforcement learning model in a certain sense (or the running performance of the entire model) is the best, and can guide the current After the optimal action is taken in the symptom state, ask for better or more rewarding actions.
  • S90 Predict the updated symptom information of the patient through the acquired lower-level reinforcement learning model, and obtain an action result corresponding to the updated symptom information of the patient.
  • the updated patient symptom information is predicted through the acquired lower-level reinforcement learning model, that is, state prediction, so as to predict the action result, and the action result is for the updated
  • the result of the optimal scheduling action determined by the patient s symptom information, that is, the action that produces the best value (maximum reward) selected in the predicted action space, the result determined after the action is performed, and the training completed
  • the lower-level reinforcement learning model can match a corresponding action that can achieve the maximum reward through the updated patient symptom information, and the result generated after the action is executed, which corresponds to the updated patient symptom information
  • the action results include recommended department actions and inquiry actions.
  • the action result is the recommended department action
  • the recommended department action is reached, which means that the updated patient symptom information can accurately and effectively reflect the current patient’s character, and there is no need to
  • the patient symptom information is supplemented again, the updated patient symptom information is input into the department triage model, and the patient symptom information is identified through the department triage model, and the symptom characteristic is identified as
  • the updated patient symptom information undergoes word vector conversion, and then the converted word vectors are spliced to extract symptom features from the spliced vector.
  • the symptom features are implicit vector characteristics between symptoms and departments , Identifying the triage result corresponding to the patient based on the symptom characteristics, the triage result being the predicted highest probability triage category, the triage result including the department category, that is, the department provided for the patient to see
  • the results of the triage provide an accurate basis for the patient to make an appointment, and it is convenient for the patient to choose an accurate department to make an appointment.
  • This application realizes that by sending out a new round of symptom inquiry information in the inquiry action when the action result is detected as an inquiry action after obtaining the action result, and receiving the patient's response to the new round of symptom inquiry information Information, update the patient symptom information according to the response information; input the updated patient symptom information into the upper learning model, and perform body system recognition on the updated patient symptom information through the upper learning model, Identify the second human body system recognition category corresponding to the updated patient symptom information; obtain the lower reinforcement learning model associated with the recognized second human system category from the hierarchical reinforcement learning model; pass The acquired lower-level reinforcement learning model predicts the updated patient symptom information, and obtains the action result corresponding to the updated patient symptom information; when the action result is a recommended department action, the updated The patient symptom information is input into the department triage model, and the updated patient symptom information is identified through the department triage model to identify the triage results corresponding to the patient.
  • the action result is an inquiry action
  • a new round of symptom inquiry information in the inquiry action is issued, and the patient symptom information is updated according to the response information of the patient's answer.
  • Perform human body system recognition on patient symptom information and identify the first human body system category corresponding to the patient symptom information; obtain the lower reinforcement learning model associated with the second human body system category from the hierarchical reinforcement learning model; Predict the updated patient symptom information through the acquired lower-level reinforcement learning model and obtain the action result; when the action result is a recommended department action, the updated patient is analyzed through the department triage model Symptom information is used for symptom feature identification, and the corresponding triage results of the patient are identified.
  • a new round of symptom inquiry information for inquiring patients is output through the hierarchical reinforcement learning model, and related symptoms are supplemented by the new round of symptom inquiry information. It can supplement useful patient symptom information, and then use the department triage model to identify the symptom characteristics of the useful supplemented patient symptom information, which can more accurately determine the department that the patient needs to see, improve the accuracy of medical treatment, and improve the patient experience.
  • step S90 that is, after obtaining the action result corresponding to the updated patient symptom information, the method includes:
  • a new round of symptom inquiry information in the inquiry action is sent out, and after multiple rounds of new symptom inquiry information interacted with the patient, the information is continuously updated. State the patient’s symptom information until it is detected that the action result is a recommended department action, indicating that the continuously updated patient symptom information can accurately reflect the patient’s symptom information. At this time, enter the last updated patient symptom information into the In the department triage model, the last updated patient symptom information is identified by the department triage model to obtain the final triage result.
  • This application realizes that by continuously interacting with the patient through a new round of symptom inquiry information when it is detected that the action result is an inquiring action, and continuously updating the patient's symptom information, until it is detected that the action result is a recommended department action , Complete the patient's symptom information, perform symptom feature recognition on the last updated patient's symptom information through the department triage model, and identify the triage results corresponding to the patient.
  • a hierarchical reinforcement learning model is realized Output the new round of symptom inquiry information of the patient, and constantly supplement the relevant symptoms with the new round of symptom inquiry information, which can continuously supplement useful patient symptom information, and then use the department triage model to symptom the useful supplemented patient symptom information Feature recognition can more accurately determine the department where the patient needs to see a doctor, improve the accuracy of medical treatment, and improve the patient experience.
  • a triage data processing device is provided, and the triage data processing device corresponds to the triage data processing method in the foregoing embodiment one-to-one.
  • the triage data processing device includes a receiving module 11, an identification module 12, an acquisition module 13, a prediction module 14 and a triage module 15.
  • the detailed description of each functional module is as follows:
  • the receiving module 11 is configured to receive a patient request from a patient, and obtain patient symptom information in the patient request;
  • the recognition module 12 is configured to input the patient's symptom information into a hierarchical reinforcement learning model, and perform body system recognition on the patient's symptom information through the upper learning model, and identify the first body system category corresponding to the patient's symptom information;
  • the hierarchical reinforcement learning model includes the upper learning model and multiple lower reinforcement learning models; a lower reinforcement learning model is associated with a human body system category;
  • the acquiring module 13 is configured to acquire the lower-level reinforcement learning model associated with the identified first human body system category from the hierarchical reinforcement learning model;
  • the prediction module 14 is configured to predict the patient's symptom information through the acquired lower-level reinforcement learning model and obtain an action result; the action result is an optimal scheduling action determined for the patient's symptom information;
  • the triage module 15 is used to input the patient's symptom information into the department triage model when the action result is a recommended department action, and perform symptom feature identification on the patient's symptom information through the department triage model, and identify A triage result corresponding to the patient is output.
  • Each module in the above-mentioned triage data processing device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 7.
  • the computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a readable storage medium and an internal memory.
  • the readable storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer readable instructions in the readable storage medium.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by the processor to realize a triage data processing method.
  • the readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.
  • a computer device including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor. Diagnostic data processing method.
  • one or more readable storage media storing computer readable instructions are provided.
  • the readable storage media provided in this embodiment include non-volatile readable storage media and volatile readable storage. Medium; the readable storage medium stores computer readable instructions, and when the computer readable instructions are executed by one or more processors, the one or more processors implement the triage data processing method in the above-mentioned embodiment.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种分诊数据处理方法、装置、计算机设备及存储介质，涉及人工智能技术领域，所述方法包括：通过获取患者的患者症状信息；通过层级强化学习模型中的上层学习模型对患者症状信息进行人体系统识别，识别出第一人体系统类别；自层级强化学习模型中获取与第一人体系统类别关联的下层强化学习模型；通过获取的下层强化学习模型对患者症状信息进行预测并获取动作结果；在动作结果为推荐科室动作时，通过科室分诊模型对患者症状信息进行症状特征识别，识别出与患者对应的分诊结果。该方法通过层级强化学习模型询问患者相关症状，能够快速地、准确地确定患者就诊的科室。该方法适用于智慧医疗等领域，可进一步推动智慧城市的建设。

Description

分诊数据处理方法、装置、计算机设备及存储介质
本申请要求于2020年9月8日提交中国专利局、申请号为202010935263.7，发明名称为“分诊数据处理方法、装置、计算机设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种分诊数据处理方法、装置、计算机设备及存储介质。
背景技术
随着医学的进步和发展，医院对于科室的设置也更专业化，随之带来的问题是用户选择科室带来一定的困难，为了解决这个问题各大医院都增加了导诊环节，包括导诊人员和自主导诊服务，主要是帮助患者推荐诊疗科室。
目前，发明人发现患者去医院就诊时，首先需要去分诊台进行人工分诊，在该过程中患者需要消耗大量排队时间，而且对分诊台的导诊人员的专业知识深度及广度上有较高的要求，如果导诊人员给患者分诊错误，又需要重新进行分诊，大大浪费患者的时间，严重影响患者体验，因此，在现有技术上，患者进行人工分诊过程中耗时长、很难给出合理的就诊科室或者就诊医生，从而导致患者体验差，以及就诊准确率低。
发明内容
本申请提供一种分诊数据处理方法、装置、计算机设备及存储介质，实现了通过层级强化学习模型询问患者相关症状，能够提取出有用的患者症状信息，再通过科室分诊模型对有用的患者症状信息进行症状特征识别，本申请适用于智慧医疗等领域，可进一步推动智慧城市的建设，能够快速地、准确地确定患者需要就诊的科室，提升了就诊准确率，提升了患者体验。
一种分诊数据处理方法，包括：
接收到患者的患者请求，获取所述患者请求中的患者症状信息；
将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
一种分诊数据处理装置，包括：
接收模块，用于接收到患者的患者请求，获取所述患者请求中的患者症状信息；
识别模块，用于将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学 习模型与一个人体系统类别关联；
获取模块，用于自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
预测模块，用于通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
分诊模块，用于在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
接收到患者的患者请求，获取所述患者请求中的患者症状信息；
将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
一个或多个存储有计算机可读指令的可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
接收到患者的患者请求，获取所述患者请求中的患者症状信息；
将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
本申请提供的分诊数据处理方法、装置、计算机设备及存储介质，通过接收到患者的患者请求，获取所述患者请求中的患者症状信息；将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征 识别，识别出与所述患者对应的分诊结果，如此，实现了通过获取患者的患者症状信息；通过层级强化学习模型中的上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；自所述层级强化学习模型中获取与所述第一人体系统类别关联的所述下层强化学习模型；通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；在所述动作结果为推荐科室动作时，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果，因此，实现了通过层级强化学习模型询问患者相关症状，能够提取出有用的患者症状信息，再通过科室分诊模型对有用的患者症状信息进行症状特征识别，能够快速地、准确地确定患者需要就诊的科室，提升了就诊准确率，提升了患者体验。
本申请的一个或多个实施例的细节在下面的附图和描述中提出，本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例中分诊数据处理方法的应用环境示意图；
图2是本申请一实施例中分诊数据处理方法的流程图；
图3是本申请一实施例中分诊数据处理方法的步骤S20的流程图；
图4是本申请一实施例中分诊数据处理方法的步骤S30的流程图；
图5是本申请一实施例中分诊数据处理方法的步骤S10的流程图；
图6是本申请一实施例中分诊数据处理装置的原理框图；
图7是本申请一实施例中计算机设备的示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请提供的分诊数据处理方法，可应用在如图1的应用环境中，其中，客户端(计算机设备)通过网络与服务器进行通信。其中，客户端(计算机设备)包括但不限于为各种个人计算机、笔记本电脑、智能手机、平板电脑、摄像头和便携式可穿戴设备。服务器可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一实施例中，如图2所示，提供一种分诊数据处理方法，其技术方案主要包括以下步骤S10-S50：
S10，接收到患者的患者请求，获取所述患者请求中的患者症状信息。
可理解地，所述患者症状信息为患者输入的与患者的症状相关的信息，所述患者可以在应用程序平台上输入完所述患者就诊信息，或者在应用程序平台上提供的所有症状描述集中选取相关的症状描述作为所述患者症状信息，或者在应用程序平台上输入完一段当前症状的语音或者文本描述，从该语音或者文本描述中识别出其中的关键词，将识别到的关键词作为所述患者症状信息等等，从而触发所述患者请求。
在一实施例中，如图5所示，所述步骤S10之前，所述获取所述患者请求中的患者症状信息之前，包括：
S101，获取患者输入的患者输入信息。
可理解地，所述患者输入信息为患者在应用程序平台上输入的文本信息。
S102，将所述患者输入信息输入预设的预处理模型，通过所述预处理模型对所述患者输入信息进行关键词识别，得到关键词结果；所述关键词结果包括关键词及与关键词关联的预测值。
可理解地，所述预处理模型为基于BERT模型且训练完成的神经网络模型，所述预处理模型能够实现识别所述患者输入信息中的关键词，并且能够预测出与各关键词对应的所述预测值，所述关键词识别的过程为运用所述预处理模型中的BERT算法对所述患者输入信息进行拆分及词向量转换，通过对转换后的词向量进行预测，预测出与其相对应的与症状相关的关键词，并且能够计算出与该关键词的相似度，从而得出各识别出的关键词的预测值，所述BERT(Bidirectional Encoder Representations from Transformers)算法为通过联合调节所有层中的双向Transformer来识别的算法。
其中，所述关键词结果为从所述患者输入信息中识别到的关键词，所述关键词结果包括多个识别到的所述关键词，所述关键词为与症状相关的且能够有效体现症状的词语。
S103，将与达到预设阈值的所述预测值对应的所述关键词确定为所述患者症状信息，并触发所述患者请求。
可理解地，所述预设阈值为预先设定的概率值，所述预设阈值可以根据需求设定，从所述关键词结果中，将大于所述预设阈值的所述预测值对应的所述关键词标记为所述患者症状信息，从而从患者输入信息中提取出有用的词语，对所述患者症状信息的用语进行统一，便于后续层级强化学习模型和科室分诊模型的识别，提高了识别效率，并提高了识别准确率和可靠性。
本申请实现了通过获取患者输入的患者输入信息；通过所述预处理模型对所述患者输入信息进行关键词识别，得到关键词结果；将与达到预设阈值的所述预测值对应的所述关键词确定为所述患者症状信息，并触发所述患者请求，如此，实现了通过预处理模型提取所述患者输入信息中的关键词，能够对患者症状信息的用语进行统一，便于后续层级强化学习模型和科室分诊模型的识别，提高了识别效率，并提高了识别准确率和可靠性。
S20，将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联。
可理解地，所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型，所述上层学习模型为训练完成的网络模型，所述上层学习模型能够对输入的信息进行人体系统识别，识别出输入的信息属于哪一类的人体系统类别，所述上层学习模型的网络结构可以根据需求设定，比如所述上层学习模型的网络结构为深度卷积神经网络模型、强化学习模型等等，所述人体系统识别为根据所述上层学习模型对所述患者症状信息进行人体系统特征提取，根据所述人体系统特征进行识别出属于哪一类的人体系统类别，所述人体系统特征为与人体系统类别相关的特征，通过所述上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别，所述第一人体系统类别为根据所述患者症状信息识别出的人体系统类别，其中，所述人体系统类别包括运动系统、神经系统、消化系统、泌尿系统、生殖系统、呼吸系统、循环系统和内分泌系统。
在一实施例中，所述上层学习模型的网络结构为强化学习模型，通过上层学习模型将所述患者症状信息作为当前状态，所述上层学习模型中的智能体(agent)根据所述当前状态确定出动作空间，并抽象出各动作空间的价值，从而得到最佳的动作，通过执行最佳的动作向与所述患者症状信息对应的人体系统类别靠拢，得到与该最佳的动作对应的奖励值，根据与该最佳的动作对应的奖励值识别出所述第一人体系统类别。
在一实施例中，如图3所示，所述步骤S20之前，即所述将所述患者症状信息输入层级强化学习模型之前，包括：
S201，获取症状样本集；所述症状样本集包括多个症状样本，所述症状样本与一个人体系统类别标签关联。
可理解地，所述症状样本集为所有所述症状样本的集合，所述症状样本为历史收集的症状词语，所述症状样本包括若干个症状词语，一个所述症状样本与一个所述人体系统类别标签关联，所述人体系统类别标签为人工对该症状样本标注的哪一类人体系统类别。
S202，将所述症状样本输入含有第一初始参数的分诊神经网络模型。
可理解地，所述分诊神经网络模型包括所述第一初始参数，所述分诊神经网络模型的网络结构可以根据需求进行设定，比如所述分诊神经网络模型的网络结构为Word2vec模型、深度卷积神经网络模型等等，所述第一初始参数可以通过迁移学习的方式直接将其他训练完成的神经网络模型中迁移过来。
S203，通过所述分诊神经网络模型对所述症状样本进行人体系统识别，获取与所述症状样本对应的所述人体系统类别的样本识别结果。
可理解地，所述人体系统识别的过程可以为通过对所述症状样本进行词向量转换，即通过词向量字典，将所述症状样本转换成词向量，将所有转换后的词向量拼接成文本特征向量，通过对文本特征向量进行卷积，即提取所述人体系统特征，通过对卷积后的文本特征向量进行全连接识别，从而识别出与所述症状样本对应的所述人体系统类别的样本识别结果，所述样本识别结果为识别出所述症状样本属于哪一类人体系统类别。
S204，根据所述样本识别结果与所述人体系统类别标签，确定出损失值。
可理解地，将所述样本识别结果和与所述症状样本关联的所述人体系统类别标签输入所述分诊神经网络模型中的损失函数中，通过所述损失函数计算出所述损失值，所述损失值表明了所述样本识别结果与所述人体系统类别标签之间的差异，所述损失函数可以根据需求设定，比如交叉熵损失函数。
S205，在所述损失值未达到预设的收敛条件时，迭代更新所述分诊神经网络模型的初始参数，直至所述损失值达到所述预设的收敛条件时，将收敛之后的所述分诊神经网络模型记录为上层学习模型。
可理解地，所述收敛条件可以为所述损失值经过了5000次计算后值为很小且不会再下降的条件，即在所述损失值经过5000次计算后值为很小且不会再下降时，停止训练，并将收敛之后的所述分诊神经网络模型记录为上层学习模型；所述收敛条件也可以为所述损失值小于设定阈值的条件，即在所述损失值小于设定阈值时，停止训练，并将收敛之后的所述所述分诊神经网络模型记录为上层学习模型，如此，在所述损失值未达到预设的收敛条件时，不断更新迭代所述卷积神经网络模型的初始参数，并触发通过通过所述分诊神经网络模型对所述症状样本进行人体系统识别，获取与所述症状样本对应的所述人体系统类别的样本识别结果的步骤，可以不断向准确的结果靠拢，让识别的准确率越来越高。
S30，自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型。
可理解地，一个下层强化学习模型与一个人体系统类别关联，所述人体系统类别包括运动系统、神经系统、消化系统、泌尿系统、生殖系统、呼吸系统、循环系统和内分泌系统，所述下层强化学习模型包括运动系统下层强化学习模型、神经系统下层强化学习模型、消化系统下层强化学习模型、泌尿系统下层强化学习模型、生殖系统下层强化学习模型、呼吸系统下层强化学习模型、循环系统下层强化学习模型和内分泌系统下层强化学习模型，各个所述上层学习模型都是通过与所述上层学习模型关联的所述人体系统类别的样本进行强化学习并训练完成后获得，所述层级强化学习模型中包括多个所述下层强化学习模型，从所述层级强化学习模型中获取与所述第一人体系统类别关联的所述下层强化学习模型， 所述下层强化学习模型的训练过程中学习与其关联的人体系统类别的症状状态到动作空间的映射，使得智能体(agent)选择的动作能够获得最大的奖励，使得下层强化学习模型在某种意义下的评价(或整个模型的运行性能)为最佳，并且能够引导针对当前症状状态下做出最优动作之后向更佳或者更大奖励的询问动作。
在一实施例中，如图4所示，所述步骤S30之前，即所述自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型之前，包括：
S301，获取症状状态样本集；所述症状状态样本集包括多个症状状态样本，所述症状状态样本与一个科室标签关联，所有所述症状状态样本均与一个相同的人体系统类别关联。
可理解地，所述症状状态样本集为所述症状状态样本的集合，所述症状状态样本为历史收集的与相同的一个所述人体系统类别对应的体现症状状态的词语，所述症状状态样本可以与所述症状样本相同，也可以与所述症状样本不相同，一个所述症状状态样本包括了若干个所述症状状态的词语，一个所述症状状态样本与一个所述科室标签关联，所述科室标签为体现科室类别的标签，并且所述科室标签是根据与其关联的所述症状状态样本人工标注或者就诊之后确定的标签，所有所述症状状态样本都与相同的一个所述人体系统类别关联，所述症状状态样本集中的所述症状状态样本均是在一个人体系统类别下体现的症状状态的词语。
S302，将所述症状状态样本输入与所述人体系统类别关联且含有第二初始参数的初始强化学习模型。
可理解地，所述初始强化学习模型为通过强化学习进行训练的网络模型，所述初始强化学习模型包括所述第二初始参数。
S303，通过所述初始强化学习模型匹配出与所述症状状态样本对应的动作空间。
可理解地，所述初始强化学习模型根据所述症状状态样本中体现先的症状状态的词语，匹配出与其对应的动作空间，所述动作空间为针对所述症状状态样本而采取动作的集合，执行所述动作空间中的动作会预测出所述症状状态样本的下一步状态的可能性。
S304，执行所述动作空间，得到状态转移结果；所述状态转移结果包括科室结果和状态结果。
可理解地，通过强化学习的方式进行状态预测，所述强化学习(Reinforcement Learning,RL)，又称再励学习、评价学习或增强学习，用于描述和解决智能体(agent)在与环境的交互过程中通过学习策略以达成回报最大化或实现特定目标的过程，所述状态预测为执行所述动作空间的动作，根据每一个动作出现的概率来采取行动，选择预测价值最高的动作，从而预测出下一步状态的最佳的可能性，根据预测出的下一步状态确定出所述状态转移结果，所述状态转移结果即为所述症状状态样本执行所述动作空间后预测出的下一步体现症状状态的词语，以及所有科室的可能性由大到小的排序，所述状态转移结果包括所述科室结果和所述状态结果，所述科室结果为预测出所有科室的可能性由大到小的顺序排序的结果，以及上一次预测出所有科室的可能性由大到小的顺序排序的结果，所述状态结果为预测出的下一步体现症状状态的词语的集合。
S305，根据所述症状状态样本、所述状态转移结果和所述科室标签，确定出奖励值。
可理解地，根据所述症状状态样本、执行完所述动作空间得到的所述状态转移结果和与所述症状状态样本关联的所述科室标签，通过所述初始强化学习模型中的奖励函数，计算出与所述状态转移结果对应的奖励值，所述奖励值表明了执行完所述动作空间后给出的奖励评价。
在一实施例中，所述步骤S305中，即所述根据所述状态转移结果和所述科室标签，确定出奖励值，包括：
S3051，将所述症状状态样本、所述状态转移结果和所述科室标签输入奖励函数中，通过所述奖励函数计算出所述奖励值；所述奖励函数为：
R s＝α 1·IF(s t∈S u)+α 2·tanh(τ·(p t-1-p t))+α 3·r t
其中，
R s为奖励值；
α 1为返回值的权重；
IF(s t∈S u)为所述状态结果是否在症状状态样本中的返回值，所述状态结果在症状状态样本中则返回1，所述状态结果不在症状状态样本中则返回-1；
s t为第t次状态预测的状态结果；
S u为症状状态样本；
α 2为预测分数指的权重；
p t-1为在第t-1次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
p t为在第t次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
α 3为准确奖励值的权重；
r t为与p t对应的准确奖励值。
可理解地，r t为与p t对应的准确奖励值，在一实施例中，所述r t与所述p t可以存在一对一的对应关系，即随着所述p t的变化而确定出与所述p t对应的准确奖励值，比如p t＝4，则r t为50；p t＝3，则r t为60等等。
在一实施例中，在所述p t为1时，所述r t为预设的最大奖励值，所述最大奖励值表明了预测出的科室结果达到最佳的结果，比如100或者200等；在p t不为1时，r t为零，表明了在p t未达到1时，不提供奖励。
S306，在所述奖励值未达到预设的奖励收敛条件时，迭代更新所述初始强化学习模型的第二初始参数，直至所述奖励值达到所述预设的奖励收敛条件时，将收敛之后的所述初始强化学习模型记录为下层强化学习模型。
可理解地，所述奖励收敛条件可以为所述p t为1时，停止训练，也可以为经过t次状态预测之后，奖励值达到最大不会再发生变化，从而停止训练，在所述奖励值未达到预设的所述奖励收敛条件时，迭代更新所述初始强化学习模型的第二初始参数，并触发通过所述初始强化学习模型匹配出与所述症状状态样本对应的动作空间的步骤，直到所述奖励值达到所述预设的奖励收敛条件，停止训练，将收敛之后的所述初始强化学习模型记录为下层强化学习模型。
如此，通过强化学习的方式进行训练，可以不断向准确的结果靠拢，让预测的准确率越来越高。
S40，通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作。
可理解地，通过获取到的所述下层强化学习模型对所述患者症状信息进行预测，也即状态预测，从而预测出所述动作结果，所述动作结果为针对所述患者症状信息确定的最优调度动作产生的结果，也即在预测出的动作空间中选择的产生最佳价值(最大奖励)的动作，执行完动作之后确定出的结果，通过训练完成的所述下层强化学习模型能够通过所述患者症状信息可以匹配出一个与其对应的能够达到最大奖励的动作，并执行完该动作后产生的结果，所述动作结果包括推荐科室动作和询问动作，即在执行完动作后确定出下一步的动作是推荐科室动作还是询问动作。
S50，在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的 分诊结果。
可理解地，在所述动作结果为所述推荐科室动作时，即达到了推荐科室动作的情况，即表明当前的所述患者症状信息能够准确地、有效地体现当前患者的表征，无需再进行补充所述患者症状信息，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，所述症状特征识别为将所述患者症状信息进行词向量转换，然后将转换后的词向量进行拼接，从而对拼接后的向量进行提取症状特征，所述症状特征为症状与科室之间的隐含的向量特性，根据所述症状特征识别出与所述患者对应的分诊结果，所述分诊结果为预测出的最高概率的分诊类别，所述分诊结果包括科室类别，即提供给患者就诊的科室的类别，所述分诊结果给患者进行预约提供了准确的依据，便于患者选择准确的科室进行预约就诊。
本申请实现了通过接收到患者的患者请求，获取所述患者请求中的患者症状信息；将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果，如此，实现了通过获取患者的患者症状信息；通过层级强化学习模型中的上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；自所述层级强化学习模型中获取与所述第一人体系统类别关联的所述下层强化学习模型；通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；在所述动作结果为推荐科室动作时，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果，因此，实现了通过层级强化学习模型询问患者相关症状，能够提取出有用的患者症状信息，再通过科室分诊模型对有用的患者症状信息进行症状特征识别，能够快速地、准确地确定患者需要就诊的科室，提升了就诊准确率，提升了患者体验。
在一实施例中，所述步骤S40之后，即所述通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果之后，还包括：
S60，在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，接收到所述患者针对新一轮症状询问信息回答的应答信息，根据所述应答信息更新所述患者症状信息。
可理解地，在所述动作结果为询问动作时，即表明需要对所述患者症状信息进行补充，增加有用的症状信息，所述询问动作中包括所述新一轮症状询问信息，所述新一轮症状询问信息为根据所述患者症状信息和预测所述患者症状信息后的状态信息，确定出下一步的动作策略而提出的询问问题的信息，通过所述新一轮症状询问信息能够向所述患者询问有助于补充完整的患者症状信息的问题，所述患者接收到所述新一轮症状询问信息后，做出针对所述新一轮症状询问信息而回答的所述应答信息，根据接收到的所述应答信息补充至原来的所述患者症状信息中，即在原来的所述患者症状信息后增加所述应答信息，从而完成所述患者症状信息的更新，所述应答信息可以为对所述患者回答所述新一轮症状询问信息的内容进行提取的症状词语。
S70，将更新后的所述患者症状信息输入所述上层学习模型，通过所述上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与更新后的所述患者症状信息对应的第二人体系统识别类别。
可理解地，将更新后的所述患者症状信息输入所述上层学习模型，所述人体系统识别为根据所述上层学习模型对所述患者症状信息进行人体系统特征提取，根据所述人体系统特征进行识别出属于哪一类的人体系统类别，所述人体系统特征为与人体系统类别相关的特征，通过所述上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与更新后的所述患者症状信息对应的第二人体系统类别，所述第二人体系统类别为根据更新后的所述患者症状信息识别出的人体系统类别，所述第二人体系统类别可能跟所述第一人体系统类别相同，也可能跟所述第一人体系统类别不相同，因为有些症状词语(也即全文中的症状信息、症状状态)会出现在多个人体系统类别中。
S80，自所述层级强化学习模型中获取与识别出的所述第二人体系统类别关联的所述下层强化学习模型。
可理解地，从所述层级强化学习模型中获取与所述第二人体系统类别关联的所述下层强化学习模型，所述下层强化学习模型的训练过程中学习与其关联的人体系统类别的症状状态到动作空间的映射，使得智能体(agent)选择的动作能够获得最大的奖励，使得下层强化学习模型在某种意义下的评价(或整个模型的运行性能)为最佳，并且能够引导针对当前症状状态下做出最优动作之后向更佳或者更大奖励的询问动作。
S90，通过获取的所述下层强化学习模型对更新后的所述患者症状信息进行预测，获取与更新后的所述患者症状信息对应的动作结果。
可理解地，通过获取到的所述下层强化学习模型对更新后的所述患者症状信息进行预测，也即状态预测，从而预测出所述动作结果，所述动作结果为针对更新后的所述患者症状信息确定的最优调度动作产生的结果，也即在预测出的动作空间中选择的产生最佳价值(最大奖励)的动作，执行完动作之后确定出的结果，通过训练完成的所述下层强化学习模型能够通过更新后的所述患者症状信息可以匹配出一个与其对应的能够达到最大奖励的动作，并执行完该动作后产生的结果，所述与更新后的所述患者症状信息对应的动作结果包括推荐科室动作和询问动作。
S100，在所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对更新后的所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
可理解地，在所述动作结果为所述推荐科室动作时，即达到了推荐科室动作的情况，即表明更新后的所述患者症状信息能够准确地、有效地体现当前患者的表征，无需再进行再次补充所述患者症状信息，将更新后的所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，所述症状特征识别为将更新后的所述患者症状信息进行词向量转换，然后将转换后的词向量进行拼接，从而对拼接后的向量进行提取症状特征，所述症状特征为症状与科室之间的隐含的向量特性，根据所述症状特征识别出与所述患者对应的分诊结果，所述分诊结果为预测出的最高概率的分诊类别，所述分诊结果包括科室类别，即提供给患者就诊的科室的类别，所述分诊结果给患者进行预约提供了准确的依据，便于患者选择准确的科室进行预约就诊。
本申请实现了通过在获取动作结果之后检测到所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，接收到所述患者针对新一轮症状询问信息回答的应答信息，根据所述应答信息更新所述患者症状信息；将更新后的所述患者症状信息输入所述上层学习模型，通过所述上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与更新后的所述患者症状信息对应的第二人体系统识别类别；自所述层级强化学习模型中获取与识别出的所述第二人体系统类别关联的所述下层强化学习模型；通过获取的所述下层强化学习模型对更新后的所述患者症状信息进行预测，获取与更新后的所述患者症状信息对应的动作结果；在所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对更新后的所述患者症状信息进行症状 特征识别，识别出与所述患者对应的分诊结果，如此，实现了通过动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，根据患者回答的应答信息更新所述患者症状信息，通过层级强化学习模型中的上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；自所述层级强化学习模型中获取与所述第二人体系统类别关联的所述下层强化学习模型；通过获取的所述下层强化学习模型对更新后的所述患者症状信息进行预测并获取动作结果；在所述动作结果为推荐科室动作时，通过所述科室分诊模型对更新后的所述患者症状信息进行症状特征识别，识别出所述患者对应的分诊结果，因此，实现了通过层级强化学习模型输出询问患者的新一轮症状询问信息，通过该新一轮症状询问信息补充相关症状，能够补充有用的患者症状信息，再通过科室分诊模型对有用的补充后的患者症状信息进行症状特征识别，能够更加准确地确定患者需要就诊的科室，提升了就诊准确率，提升了患者体验。
在一实施例中，所述步骤S90之后，即所述获取与更新后的所述患者症状信息对应的动作结果之后，包括：
S110，在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，经过多轮与患者交互的新一轮症状询问信息之后，对应更新所述患者症状信息，直至检测到所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入所述科室分诊模型中，通过所述科室分诊模型识别出与所述患者对应的分诊结果。
可理解地，在检测到所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，经过多轮与患者之间交互的新一轮症状询问信息之后，不断更新所述患者症状信息，直到检测到所述动作结果为推荐科室动作时，说明不断更新后的所述患者症状信息已经能够准确体现患者症状的信息，此时将最后更新后的患者症状信息输入所述科室分诊模型中，通过所述科室分诊模型对最后更新的患者症状信息的识别，得到最后的所述分诊结果。
本申请实现了通过在检测到所述动作结果为询问动作时，不断通过新一轮症状询问信息与患者进行交互，不断更新所述患者症状信息，直到检测到所述动作结果为推荐科室动作时，补充完整所述患者症状信息，通过所述科室分诊模型对最后更新后的患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果，如此，实现了通过层级强化学习模型输出询问患者的新一轮症状询问信息，不断通过新一轮症状询问信息进行补充相关症状，能够不断补充有用的患者症状信息，再通过科室分诊模型对有用的补充后的患者症状信息进行症状特征识别，能够更加准确地确定患者需要就诊的科室，提升了就诊准确率，提升了患者体验。
在一实施例中，提供一种分诊数据处理装置，该分诊数据处理装置与上述实施例中分诊数据处理方法一一对应。如图6所示，该分诊数据处理装置包括接收模块11、识别模块12、获取模块13、预测模块14和分诊模块15。各功能模块详细说明如下：
接收模块11，用于接收到患者的患者请求，获取所述患者请求中的患者症状信息；
识别模块12，用于将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
获取模块13，用于自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
预测模块14，用于通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
分诊模块15，用于在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述 患者对应的分诊结果。
关于分诊数据处理装置的具体限定可以参见上文中对于分诊数据处理方法的限定，在此不再赘述。上述分诊数据处理装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图7所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括可读存储介质、内存储器。该可读存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为可读存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种分诊数据处理方法。本实施例所提供的可读存储介质包括非易失性可读存储介质和易失性可读存储介质。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机可读指令，处理器执行计算机可读指令时实现上述实施例中分诊数据处理方法。
在一个实施例中，提供了一个或多个存储有计算机可读指令的可读存储介质，本实施例所提供的可读存储介质包括非易失性可读存储介质和易失性可读存储介质；该可读存储介质上存储有计算机可读指令，该计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器实现上述实施例中分诊数据处理方法。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质或易失性可读存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种分诊数据处理方法，其中，包括：
    接收到患者的患者请求，获取所述患者请求中的患者症状信息；
    将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
    自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
    通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
    在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  2. 如权利要求1所述的分诊数据处理方法，其中，所述通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果之后，还包括：
    在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，接收到所述患者针对新一轮症状询问信息回答的应答信息，根据所述应答信息更新所述患者症状信息；
    将更新后的所述患者症状信息输入所述上层学习模型，通过所述上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与更新后的所述患者症状信息对应的第二人体系统识别类别；
    自所述层级强化学习模型中获取与识别出的所述第二人体系统类别关联的所述下层强化学习模型；
    通过获取的所述下层强化学习模型对更新后的所述患者症状信息进行预测，获取与更新后的所述患者症状信息对应的动作结果；
    在所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对更新后的所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  3. 如权利要求2所述的分诊数据处理方法，其中，所述获取与更新后的所述患者症状信息对应的动作结果之后，包括：
    在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，经过多轮与患者交互的新一轮症状询问信息之后，对应更新所述患者症状信息，直至检测到所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入所述科室分诊模型中，通过所述科室分诊模型识别出与所述患者对应的分诊结果。
  4. 如权利要求1所述的分诊数据处理方法，其中，所述将所述患者症状信息输入层级强化学习模型之前，包括：
    获取症状样本集；所述症状样本集包括多个症状样本，所述症状样本与一个人体系统类别标签关联；
    将所述症状样本输入含有第一初始参数的分诊神经网络模型；
    通过所述分诊神经网络模型对所述症状样本进行人体系统识别，获取与所述症状样本对应的所述人体系统类别的样本识别结果；
    根据所述样本识别结果与所述人体系统类别标签，确定出损失值；
    在所述损失值未达到预设的收敛条件时，迭代更新所述分诊神经网络模型的初始参数， 直至所述损失值达到所述预设的收敛条件时，将收敛之后的所述分诊神经网络模型记录为上层学习模型。
  5. 如权利要求1所述的分诊数据处理方法，其中，所述自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型之前，包括：
    获取症状状态样本集；所述症状状态样本集包括多个症状状态样本，所述症状状态样本与一个科室标签关联，所有所述症状状态样本均与一个相同的人体系统类别关联；
    将所述症状状态样本输入与所述人体系统类别关联且含有第二初始参数的初始强化学习模型；
    通过所述初始强化学习模型匹配出与所述症状状态样本对应的动作空间；
    执行各所述动作空间，得到状态转移结果；所述状态转移结果包括科室结果和状态结果；
    根据所述症状状态样本、所述状态转移结果和所述科室标签，确定出奖励值；
    在所述奖励值未达到预设的奖励收敛条件时，迭代更新所述初始强化学习模型的第二初始参数，直至所述奖励值达到所述预设的奖励收敛条件时，将收敛之后的所述初始强化学习模型记录为下层强化学习模型。
  6. 如权利要求5所述的分诊数据处理方法，其中，所述根据所述状态转移结果和所述科室标签，确定出奖励值，包括：
    将所述症状状态样本、所述状态转移结果和所述科室标签输入奖励函数中，通过所述奖励函数计算出所述奖励值；所述奖励函数为：
    R s＝α 1·IF(s t∈S u)+α 2·tanh(τ·(p t-1-p t))+α 3·r t
    其中，
    R s为奖励值；
    α 1为返回值的权重；
    IF(s t∈S u)为所述状态结果是否在症状状态样本中的返回值，所述状态结果在症状状态样本中则返回1，所述状态结果不在症状状态样本中则返回-1；
    s t为第t次状态预测的状态结果；
    S u为症状状态样本；
    α 2为预测分数指的权重；
    p t-1为在第t-1次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
    p t为在第t次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
    α 3为准确奖励值的权重；
    r t为与p t对应的准确奖励值。
  7. 如权利要求1所述的分诊数据处理方法，其中，所述获取所述患者请求中的患者症状信息之前，包括：
    获取患者输入的患者输入信息；
    将所述患者输入信息输入预设的预处理模型，通过所述预处理模型对所述患者输入信息进行关键词识别，得到关键词结果；所述关键词结果包括关键词及与关键词关联的预测值；
    将与达到预设阈值的所述预测值对应的所述关键词确定为所述患者症状信息，并触发所述患者请求。
  8. 一种分诊数据处理装置，其中，包括：
    接收模块，用于接收到患者的患者请求，获取所述患者请求中的患者症状信息；
    识别模块，用于将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
    获取模块，用于自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
    预测模块，用于通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
    分诊模块，用于在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，其中，所述处理器执行所述计算机可读指令时实现如下步骤：接收到患者的患者请求，获取所述患者请求中的患者症状信息；
    将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
    自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
    通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
    在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  10. 如权利要求9所述的计算机设备，其中，所述通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果之后，所述处理器执行所述计算机可读指令时还实现如下步骤：
    在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，接收到所述患者针对新一轮症状询问信息回答的应答信息，根据所述应答信息更新所述患者症状信息；
    将更新后的所述患者症状信息输入所述上层学习模型，通过所述上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与更新后的所述患者症状信息对应的第二人体系统识别类别；
    自所述层级强化学习模型中获取与识别出的所述第二人体系统类别关联的所述下层强化学习模型；
    通过获取的所述下层强化学习模型对更新后的所述患者症状信息进行预测，获取与更新后的所述患者症状信息对应的动作结果；
    在所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对更新后的所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  11. 如权利要求10所述的计算机设备，其中，所述获取与更新后的所述患者症状信息对应的动作结果之后，包括：
    在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，经过多 轮与患者交互的新一轮症状询问信息之后，对应更新所述患者症状信息，直至检测到所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入所述科室分诊模型中，通过所述科室分诊模型识别出与所述患者对应的分诊结果。
  12. 如权利要求9所述的计算机设备，其中，所述将所述患者症状信息输入层级强化学习模型之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取症状样本集；所述症状样本集包括多个症状样本，所述症状样本与一个人体系统类别标签关联；
    将所述症状样本输入含有第一初始参数的分诊神经网络模型；
    通过所述分诊神经网络模型对所述症状样本进行人体系统识别，获取与所述症状样本对应的所述人体系统类别的样本识别结果；
    根据所述样本识别结果与所述人体系统类别标签，确定出损失值；
    在所述损失值未达到预设的收敛条件时，迭代更新所述分诊神经网络模型的初始参数，直至所述损失值达到所述预设的收敛条件时，将收敛之后的所述分诊神经网络模型记录为上层学习模型。
  13. 如权利要求9所述的计算机设备，其中，所述自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取症状状态样本集；所述症状状态样本集包括多个症状状态样本，所述症状状态样本与一个科室标签关联，所有所述症状状态样本均与一个相同的人体系统类别关联；
    将所述症状状态样本输入与所述人体系统类别关联且含有第二初始参数的初始强化学习模型；
    通过所述初始强化学习模型匹配出与所述症状状态样本对应的动作空间；
    执行各所述动作空间，得到状态转移结果；所述状态转移结果包括科室结果和状态结果；
    根据所述症状状态样本、所述状态转移结果和所述科室标签，确定出奖励值；
    在所述奖励值未达到预设的奖励收敛条件时，迭代更新所述初始强化学习模型的第二初始参数，直至所述奖励值达到所述预设的奖励收敛条件时，将收敛之后的所述初始强化学习模型记录为下层强化学习模型。
  14. 如权利要求13所述的计算机设备，其中，所述根据所述状态转移结果和所述科室标签，确定出奖励值，包括：
    将所述症状状态样本、所述状态转移结果和所述科室标签输入奖励函数中，通过所述奖励函数计算出所述奖励值；所述奖励函数为：
    R s＝α 1·IF(s t∈S u)+α 2·tanh(τ·(p t-1-p t))+α 3·r t
    其中，
    R s为奖励值；
    α 1为返回值的权重；
    IF(s t∈S u)为所述状态结果是否在症状状态样本中的返回值，所述状态结果在症状状态样本中则返回1，所述状态结果不在症状状态样本中则返回-1；
    s t为第t次状态预测的状态结果；
    S u为症状状态样本；
    α 2为预测分数指的权重；
    p t-1为在第t-1次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
    p t为在第t次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
    α 3为准确奖励值的权重；
    r t为与p t对应的准确奖励值。
  15. 一个或多个存储有计算机可读指令的可读存储介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
    接收到患者的患者请求，获取所述患者请求中的患者症状信息；
    将所述患者症状信息输入层级强化学习模型，通过上层学习模型对所述患者症状信息进行人体系统识别，识别出与所述患者症状信息对应的第一人体系统类别；所述层级强化学习模型包括所述上层学习模型和多个下层强化学习模型；一个下层强化学习模型与一个人体系统类别关联；
    自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型；
    通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果；所述动作结果为针对所述患者症状信息确定的最优调度动作；
    在所述动作结果为推荐科室动作时，将所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  16. 如权利要求15所述的可读存储介质，其中，所述通过获取的所述下层强化学习模型对所述患者症状信息进行预测并获取动作结果之后，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，接收到所述患者针对新一轮症状询问信息回答的应答信息，根据所述应答信息更新所述患者症状信息；
    将更新后的所述患者症状信息输入所述上层学习模型，通过所述上层学习模型对更新后的所述患者症状信息进行人体系统识别，识别出与更新后的所述患者症状信息对应的第二人体系统识别类别；
    自所述层级强化学习模型中获取与识别出的所述第二人体系统类别关联的所述下层强化学习模型；
    通过获取的所述下层强化学习模型对更新后的所述患者症状信息进行预测，获取与更新后的所述患者症状信息对应的动作结果；
    在所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入科室分诊模型中，通过所述科室分诊模型对更新后的所述患者症状信息进行症状特征识别，识别出与所述患者对应的分诊结果。
  17. 如权利要求16所述的可读存储介质，其中，所述获取与更新后的所述患者症状信息对应的动作结果之后，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    在所述动作结果为询问动作时，发出所述询问动作中的新一轮症状询问信息，经过多轮与患者交互的新一轮症状询问信息之后，对应更新所述患者症状信息，直至检测到所述动作结果为推荐科室动作时，将更新后的所述患者症状信息输入所述科室分诊模型中，通过所述科室分诊模型识别出与所述患者对应的分诊结果。
  18. 如权利要求15所述的可读存储介质，其中，所述将所述患者症状信息输入层级强化学习模型之前，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    获取症状样本集；所述症状样本集包括多个症状样本，所述症状样本与一个人体系统 类别标签关联；
    将所述症状样本输入含有第一初始参数的分诊神经网络模型；
    通过所述分诊神经网络模型对所述症状样本进行人体系统识别，获取与所述症状样本对应的所述人体系统类别的样本识别结果；
    根据所述样本识别结果与所述人体系统类别标签，确定出损失值；
    在所述损失值未达到预设的收敛条件时，迭代更新所述分诊神经网络模型的初始参数，直至所述损失值达到所述预设的收敛条件时，将收敛之后的所述分诊神经网络模型记录为上层学习模型。
  19. 如权利要求15所述的可读存储介质，其中，所述自所述层级强化学习模型中获取与识别出的所述第一人体系统类别关联的所述下层强化学习模型之前，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    获取症状状态样本集；所述症状状态样本集包括多个症状状态样本，所述症状状态样本与一个科室标签关联，所有所述症状状态样本均与一个相同的人体系统类别关联；
    将所述症状状态样本输入与所述人体系统类别关联且含有第二初始参数的初始强化学习模型；
    通过所述初始强化学习模型匹配出与所述症状状态样本对应的动作空间；
    执行各所述动作空间，得到状态转移结果；所述状态转移结果包括科室结果和状态结果；
    根据所述症状状态样本、所述状态转移结果和所述科室标签，确定出奖励值；
    在所述奖励值未达到预设的奖励收敛条件时，迭代更新所述初始强化学习模型的第二初始参数，直至所述奖励值达到所述预设的奖励收敛条件时，将收敛之后的所述初始强化学习模型记录为下层强化学习模型。
  20. 如权利要求19所述的可读存储介质，其中，所述根据所述状态转移结果和所述科室标签，确定出奖励值，包括：
    将所述症状状态样本、所述状态转移结果和所述科室标签输入奖励函数中，通过所述奖励函数计算出所述奖励值；所述奖励函数为：
    R s＝α 1·IF(s t∈S u)+α 2·tanh(τ·(p t-1-p t))+α 3·r t
    其中，
    R s为奖励值；
    α 1为返回值的权重；
    IF(s t∈S u)为所述状态结果是否在症状状态样本中的返回值，所述状态结果在症状状态样本中则返回1，所述状态结果不在症状状态样本中则返回-1；
    s t为第t次状态预测的状态结果；
    S u为症状状态样本；
    α 2为预测分数指的权重；
    p t-1为在第t-1次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
    p t为在第t次状态预测得到的所述科室结果对应的科室序列中，与所述症状状态样本关联的科室标签的序列值；
    α 3为准确奖励值的权重；
    r t为与p t对应的准确奖励值。
PCT/CN2020/135341 2020-09-08 2020-12-10 分诊数据处理方法、装置、计算机设备及存储介质 WO2021151356A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010935263.7A CN112017742B (zh) 2020-09-08 2020-09-08 分诊数据处理方法、装置、计算机设备及存储介质
CN202010935263.7 2020-09-08

Publications (1)

Publication Number Publication Date
WO2021151356A1 true WO2021151356A1 (zh) 2021-08-05

Family

ID=73516132

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/135341 WO2021151356A1 (zh) 2020-09-08 2020-12-10 分诊数据处理方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN112017742B (zh)
WO (1) WO2021151356A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112017742B (zh) * 2020-09-08 2024-05-24 平安科技（深圳）有限公司 分诊数据处理方法、装置、计算机设备及存储介质
CN113421638B (zh) * 2021-06-22 2022-07-15 平安科技（深圳）有限公司 基于迁移学习的模型生成方法、装置和计算机设备
CN114743665A (zh) * 2022-03-21 2022-07-12 平安科技（深圳）有限公司 患病风险估计网络的优化方法、装置、介质及设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109192300A (zh) * 2018-08-17 2019-01-11 百度在线网络技术（北京）有限公司 智能问诊方法、系统、计算机设备和存储介质
CN109887587A (zh) * 2019-01-22 2019-06-14 平安科技（深圳）有限公司 智能分诊方法、系统、装置及存储介质
US20190355471A1 (en) * 2018-05-18 2019-11-21 Htc Corporation Control method and reinforcement learning for medical system
CN110928936A (zh) * 2019-10-18 2020-03-27 平安科技（深圳）有限公司 基于强化学习的信息处理方法、装置、设备和存储介质
CN111477310A (zh) * 2020-03-04 2020-07-31 平安国际智慧城市科技股份有限公司 分诊数据处理方法、装置、计算机设备及存储介质
CN112017742A (zh) * 2020-09-08 2020-12-01 平安科技（深圳）有限公司 分诊数据处理方法、装置、计算机设备及存储介质

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102708377B (zh) * 2012-04-25 2014-06-25 中国科学院计算技术研究所 虚拟人组合任务规划方法
US10013416B1 (en) * 2015-12-18 2018-07-03 Amazon Technologies, Inc. Language based solution agent
CN113610240A (zh) * 2017-05-05 2021-11-05 第四范式（北京）技术有限公司 利用嵌套机器学习模型来执行预测的方法及系统
CN109147930A (zh) * 2017-06-28 2019-01-04 京东方科技集团股份有限公司 分诊对话方法、分诊对话设备及系统
CN107944915B (zh) * 2017-11-21 2022-01-18 北京字节跳动网络技术有限公司 一种游戏用户行为分析方法及计算机可读存储介质
CN108763504B (zh) * 2018-05-30 2020-07-24 浙江大学 一种基于强化双通道序列学习的对话回复生成方法及系统
CN108922608A (zh) * 2018-06-13 2018-11-30 平安医疗科技有限公司 智能导诊方法、装置、计算机设备和存储介质
US11468998B2 (en) * 2018-10-09 2022-10-11 Radect Inc. Methods and systems for software clinical guidance
CN109635122A (zh) * 2018-11-28 2019-04-16 平安科技（深圳）有限公司 智能疾病问询方法、装置、设备及存储介质
CN109817329B (zh) * 2019-01-21 2021-06-29 暗物智能科技(广州)有限公司 一种医疗问诊对话系统以及应用于该系统的强化学习方法
CN110136839B (zh) * 2019-05-14 2021-10-08 北京百度网讯科技有限公司 一种症状信息处理方法、装置和电子设备
CN110364251B (zh) * 2019-06-14 2022-08-16 南京理工大学 一种基于机器阅读理解的智能交互导诊咨询系统
CN111241265A (zh) * 2020-01-09 2020-06-05 平安国际智慧城市科技股份有限公司 信息推荐方法、设备、存储介质及装置
CN111415740B (zh) * 2020-02-12 2024-04-19 东北大学 问诊信息的处理方法、装置、存储介质及计算机设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190355471A1 (en) * 2018-05-18 2019-11-21 Htc Corporation Control method and reinforcement learning for medical system
CN109192300A (zh) * 2018-08-17 2019-01-11 百度在线网络技术（北京）有限公司 智能问诊方法、系统、计算机设备和存储介质
CN109887587A (zh) * 2019-01-22 2019-06-14 平安科技（深圳）有限公司 智能分诊方法、系统、装置及存储介质
CN110928936A (zh) * 2019-10-18 2020-03-27 平安科技（深圳）有限公司 基于强化学习的信息处理方法、装置、设备和存储介质
CN111477310A (zh) * 2020-03-04 2020-07-31 平安国际智慧城市科技股份有限公司 分诊数据处理方法、装置、计算机设备及存储介质
CN112017742A (zh) * 2020-09-08 2020-12-01 平安科技（深圳）有限公司 分诊数据处理方法、装置、计算机设备及存储介质

Also Published As

Publication number Publication date
CN112017742A (zh) 2020-12-01
CN112017742B (zh) 2024-05-24

Similar Documents

Publication Publication Date Title
WO2021151356A1 (zh) 分诊数据处理方法、装置、计算机设备及存储介质
US11810671B2 (en) System and method for providing health information
CN108986908B (zh) 问诊数据处理方法、装置、计算机设备和存储介质
CN110364251B (zh) 一种基于机器阅读理解的智能交互导诊咨询系统
US20190311814A1 (en) Systems and methods for responding to healthcare inquiries
CN112016295B (zh) 症状数据处理方法、装置、计算机设备及存储介质
WO2021164388A1 (zh) 分诊融合模型训练方法、分诊方法、装置、设备及介质
CN112287089B (zh) 用于自动问答系统的分类模型训练、自动问答方法及装置
US10984024B2 (en) Automatic processing of ambiguously labeled data
US20200227147A1 (en) Automated generation of codes
CN112036154A (zh) 基于问诊对话的电子病历生成方法、装置和计算机设备
CN112017789B (zh) 分诊数据处理方法、装置、设备及介质
CN112287068B (zh) 基于人工智能的问诊对话数据处理方法及装置
WO2022227203A1 (zh) 基于对话表征的分诊方法、装置、设备及存储介质
WO2021139231A1 (zh) 基于神经网络模型的分诊方法、装置和计算机设备
CN112201359A (zh) 基于人工智能的重症问诊数据识别方法及装置
CN111696661A (zh) 患者分群模型构建方法、患者分群方法及相关设备
WO2022134357A1 (zh) 分诊数据处理方法、装置、计算机设备及存储介质
CN111651579B (zh) 信息查询方法、装置、计算机设备和存储介质
CN114628001B (zh) 基于神经网络的处方推荐方法、系统、设备及存储介质
CN116975218A (zh) 文本处理方法、装置、计算机设备和存储介质
CN112397197A (zh) 基于人工智能的问诊数据处理方法及装置
CN117238437A (zh) 基于知识图谱的病情诊断辅助方法及系统
CN111783473B (zh) 医疗问答中最佳答案的识别方法、装置和计算机设备
CN117238527B (zh) 基于大数据的医疗信息管理方法及系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20916878

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20916878

Country of ref document: EP

Kind code of ref document: A1