WO2022206500A1 - 信息推送方法, 装置和计算机可读存储介质 - Google Patents

信息推送方法, 装置和计算机可读存储介质 Download PDF

Info

Publication number
WO2022206500A1
WO2022206500A1 PCT/CN2022/082428 CN2022082428W WO2022206500A1 WO 2022206500 A1 WO2022206500 A1 WO 2022206500A1 CN 2022082428 W CN2022082428 W CN 2022082428W WO 2022206500 A1 WO2022206500 A1 WO 2022206500A1
Authority
WO
WIPO (PCT)
Prior art keywords
patient
information
doctor
real
historical
Prior art date
Application number
PCT/CN2022/082428
Other languages
English (en)
French (fr)
Inventor
刘宗节
罗杰
Original Assignee
北京京东拓先科技有限公司
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 北京京东拓先科技有限公司 filed Critical 北京京东拓先科技有限公司
Publication of WO2022206500A1 publication Critical patent/WO2022206500A1/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
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
    • 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

Definitions

  • the present application is based on the CN application number 202110331602.5 and the filing date is March 29, 2021, and claims its priority.
  • the disclosure of the CN application is hereby incorporated into the present application as a whole.
  • the present disclosure relates to the field of computer technologies, and in particular, to an information push method, apparatus, and computer-readable storage medium.
  • Internet hospitals realize online medical treatment, diagnosis, and prescription for users, which is convenient for users to seek medical treatment.
  • users in Internet hospitals generally interact with the system to simply describe the condition or select the department to see a doctor, and the system assigns a doctor to the user for consultation.
  • an information push method comprising: generating a fused real-time feature according to real-time information of a patient and real-time information of a doctor, wherein the real-time information of the patient includes: the patient is within a preset first time period
  • the real-time information of the doctor includes: the consultation information of the doctor within the preset first time period; the real-time condition characteristics of the patient are generated according to the real-time condition description information of the patient; according to the pre-acquired historical characteristics of the patient and the doctor's Historical features, as well as fusion of real-time features and real-time condition features of patients, determine the matching degree between patients and doctors; according to the matching between patients and each doctor, at least one doctor is pushed for patients.
  • generating the fused real-time feature according to the real-time information of the patient and the real-time information of the doctor includes: jointly inputting the real-time information of the patient and the real-time information of the doctor into the first fully connected layer of the doctor-patient matching model, and obtaining the output fused real-time information Features; wherein, the real-time information of the patient also includes: the personal feature information of the patient, and the real-time information of the doctor also includes the personal feature information and occupational feature information of the doctor.
  • generating the real-time condition characteristics of the patient according to the real-time condition description information of the patient includes: inputting the real-time condition description information of the patient into the first natural language processing model of the doctor-patient matching model to obtain the output real-time condition characteristics of the patient.
  • determining the matching degree of the patient and the doctor according to the pre-acquired historical characteristics of the patient and the doctor's historical characteristics, and fusing the real-time characteristics and the real-time condition characteristics of the patient includes: combining the pre-acquired historical characteristics of the patient and the doctor After splicing with real-time features and real-time condition features of patients, input the second fully connected layer of the doctor-patient matching model to obtain the matching degree of the output patient and doctor.
  • the method further includes: generating the patient's historical characteristics according to the patient's historical information, wherein the patient's historical information includes the patient's historical condition information and historical medical visit information.
  • generating the patient's history feature according to the patient's history information includes: dividing the patient's history information into a first sub-history information and a second sub-history information in chronological order, wherein the first sub-history information includes a preset The latest historical information in the second time period, the second sub-historical information includes historical information in different time periods divided in chronological order; the first sub-historical characteristics of the patient are determined according to the first sub-historical information; according to the second sub-historical information The history information determines the second sub-historical feature of the patient; according to the first sub-historical feature and the second sub-historical feature of the patient, the historical feature of the patient is determined.
  • determining the first sub-history feature of the patient according to the first sub-history information includes: inputting the first sub-history information into the second natural language processing model of the doctor-patient matching model, and obtaining the first sub-history feature vector as the patient.
  • the first sub-historical feature of The first deep neural network model obtains the second sub-history feature vector as the second sub-history feature of the patient; wherein the length of the first sub-history feature vector is greater than or equal to the length of the second sub-history feature vector.
  • determining the patient's history feature according to the patient's first sub-history feature and the second sub-history feature includes: after splicing the patient's first sub-history feature and the second sub-history feature, inputting a doctor-patient match
  • the third fully-connected layer of the model obtains the output patient's historical features.
  • the method further includes: generating a doctor's history feature according to the doctor's field information and the doctor's historical information, wherein the doctor's historical information includes the doctor's historical consultation information.
  • generating the doctor's history feature according to the doctor's domain information and the doctor's history information includes: determining the doctor's domain feature according to the doctor's domain information; determining the doctor's consultation feature according to the doctor's history information; Characteristics and admissions characteristics determine the physician's historical characteristics.
  • determining the doctor's domain feature according to the doctor's domain information includes: inputting the doctor's domain information into a third natural language processing model of the doctor-patient matching model, and obtaining a domain feature vector as the doctor's domain feature; according to the doctor's history
  • the information to determine the doctor's consultation characteristics includes: inputting the doctor's historical information into the second deep neural network model of the doctor-patient matching model, and obtaining the consultation feature vector as the doctor's consultation characteristics; wherein, the length of the domain feature vector is greater than or equal to the connection. The length of the diagnostic feature vector.
  • determining the doctor's history feature according to the doctor's domain feature and the consultation feature includes: after splicing the doctor's domain feature and the consultation feature, inputting the fourth fully connected layer of the doctor-patient matching model to obtain the output Historical characteristics of physicians.
  • an information push device comprising: a first generation module, configured to generate a real-time fusion feature according to real-time information of a patient and real-time information of a doctor, wherein the real-time information of the patient includes: a patient The online access information within the preset first time period, the real-time information of the doctor includes: the doctor's admission information within the preset first time period; the second generation module is used to generate the patient's medical information according to the real-time condition description information of the patient.
  • Real-time disease characteristics used to determine the degree of matching between patients and doctors based on the pre-obtained historical characteristics of patients and doctors, as well as fusion of real-time characteristics and real-time disease characteristics of patients;
  • the matching degree between the patient and each doctor push at least one doctor for the patient.
  • an information pushing apparatus comprising: a processor; and a memory coupled to the processor for storing instructions, and when the instructions are executed by the processor, the processor executes any of the foregoing The information push method of the embodiment.
  • a non-transitory computer-readable storage medium on which a computer program is stored, wherein when the program is executed by a processor, the information pushing method of any of the foregoing embodiments is implemented.
  • FIG. 1 shows a schematic flowchart of an information push method according to some embodiments of the present disclosure.
  • FIG. 2 shows a schematic structural diagram of a doctor-patient matching model according to some embodiments of the present disclosure.
  • FIG. 3 shows a schematic flowchart of information pushing methods according to other embodiments of the present disclosure.
  • FIG. 4 shows a schematic structural diagram of an information pushing apparatus according to some embodiments of the present disclosure.
  • FIG. 5 shows a schematic structural diagram of an information pushing apparatus according to other embodiments of the present disclosure.
  • FIG. 6 shows a schematic structural diagram of an information pushing apparatus according to further embodiments of the present disclosure.
  • One technical problem to be solved by the present disclosure is how to more accurately match or assign doctors to patients.
  • the present disclosure provides an information push method, which will be described below with reference to FIGS. 1 to 3 .
  • FIG. 1 is a flowchart of some embodiments of the disclosed information pushing method. As shown in FIG. 1 , the method of this embodiment includes steps S102 to S108.
  • step S102 a fused real-time feature is generated according to the real-time information of the patient and the real-time information of the doctor.
  • the real-time information of the patient for example, includes: online access information of the patient within a preset first time period (eg, the day), and the real-time information of the patient may further include: personal characteristic information of the patient.
  • the real-time information of the doctor includes, for example, information about the doctor's consultation within a preset first time period, and the real-time information of the doctor may further include the personal characteristic information and occupational characteristic information of the doctor.
  • the online access information of the patient within the preset first time period includes: the medicines browsed within the preset first time period, the departments browsed within the preset first time period, the browsed medicines within the preset first time period At least one piece of information, such as a doctor, a department for triage within a preset first time period, etc.
  • the patient's personal characteristic information includes, for example, at least one piece of information such as gender and age.
  • the doctor's consultation information within the preset first period of time includes, for example: whether to be online within the preset first period of time, online status within the preset first period of time, the department where the doctor is located within the preset first period of time, Whether there is a refusal within the preset first time period, the number of consultations within the preset first time period, the working hours within the preset first time period, the length of time since the end of the last consultation, and the number of patients in the preset first time period
  • the duration of the most recent order received in the time period, the cumulative consultation time in the preset first time period, the number of times the system urged the system in the preset first time period, and the length of time after the system was urged for the first time in the preset first time period At least one item of information such as order receipt.
  • the personal characteristic information of the doctor includes, for example, at least one item of information such as gender and age.
  • the professional characteristic information of the doctor includes, for example, at least one item of information: rank, whether it is an expert or not.
  • the real-time information of the patient and the real-time information of the doctor can be selected according to actual needs, and are not limited to the examples.
  • FIG. 2 is a schematic diagram of some embodiments of a doctor-patient matching model.
  • the doctor-patient matching model may include an online part, and the fused real-time features may be determined through the online part.
  • the online part includes a first fully connected layer.
  • the real-time information of the patient and the real-time information of the doctor are jointly input into the first fully connected layer of the doctor-patient matching model to obtain the output fused real-time feature.
  • the fused real-time feature is the fusion feature of the doctor and the patient.
  • the real-time information of the patient and various information in the real-time information of the doctor can be binary encoded in advance, and the encoding of the preset length is obtained and then input to the first fully connected layer.
  • step S104 the real-time condition characteristics of the patient are generated according to the real-time condition description information of the patient.
  • the real-time condition description information of the patient is, for example, text information describing the condition entered by the patient in the system, for example, symptom manifestations, uncomfortable parts, and the like.
  • the real-time condition characteristics of the patient are generated according to the real-time condition description information of the patient and the online access information of the patient within a preset first time period.
  • the patient's real-time condition characteristics can be determined through the online section.
  • the online part includes a first natural language processing model, which is, for example, a BERT model.
  • the real-time condition description information of the patient is input into the first natural language processing model of the doctor-patient matching model to obtain the output real-time condition characteristics of the patient.
  • the length of the vector corresponding to the real-time condition feature of the patient is greater than or equal to the length of the vector corresponding to the fused real-time feature.
  • the patient's real-time condition characteristics are closely related to the patient's current visit. Setting the corresponding vector to a longer dimension is conducive to the expression of richer information.
  • step S106 the degree of matching between the patient and the doctor is determined according to the pre-acquired historical characteristics of the patient and the historical characteristics of the doctor, as well as the fusion of the real-time characteristics and the real-time condition characteristics of the patient.
  • the doctor-patient matching model further includes a second fully connected layer.
  • the patient's historical features, the doctor's historical features, the fused real-time features and the patient's real-time condition features are spliced and input into the doctor-patient Match the second fully connected layer of the model to get the matching degree of the output patient and doctor.
  • the doctor-patient matching model can be pre-trained, selecting training samples corresponding to patients and doctors, and selecting and labeling the matching degree of patients and doctors according to the actual consultation information.
  • the training samples include: historical information, real-time information and real-time condition description information of multiple patients, and domain information, historical information and real-time information of multiple doctors.
  • the information of various patients and doctors in the training samples are respectively input into the corresponding parts of the doctor-patient matching model (as shown in Figure 2), which will not be repeated here.
  • step S108 according to the matching degree between the patient and each doctor, at least one doctor is pushed for the patient.
  • the matching degree between the patient and each doctor can be determined, and at least one doctor whose matching degree exceeds the threshold can be selected and pushed to the patient, and the patient can select it.
  • a doctor pushes to a patient, and the patient chooses.
  • the doctor with the highest matching degree can also be selected to triage the patient.
  • the fusion real-time feature is generated, and the real-time condition feature of the patient is also generated according to the real-time condition description information of the patient. Fusion of real-time characteristics and real-time condition characteristics of patients to determine the matching degree of patients and doctors.
  • the above embodiment refers to the historical information and real-time information of patients and doctors, which can more accurately describe the characteristics of patients and doctors, improve the accuracy of determining the degree of matching between patients and doctors, and then improve the matching or assigning of doctors to patients. accuracy.
  • FIG. 3 is a flowchart of other embodiments of the disclosed information push method. As shown in FIG. 3 , the method of this embodiment includes steps S302-S312.
  • step S302 the patient's historical characteristics are generated according to the patient's historical information.
  • the patient's historical information includes, for example, the patient's historical condition information and historical medical visit information.
  • the patient's historical information is divided into first sub-historical information and second sub-historical information in chronological order, wherein the first sub-historical information includes the latest historical information within a preset second time period,
  • the second sub-historical information includes historical information in different time periods divided by time sequence; the first sub-historical feature of the patient is determined according to the first sub-historical information; the second sub-historical feature of the patient is determined according to the second sub-historical information;
  • the patient's first sub-historical profile and the second sub-historical profile determine the patient's historical profile.
  • the first sub-historical information includes, for example, the latest historical disease information and historical medical visit information such as 2 months, 1 month, or 15 days (within a preset second time period).
  • the second sub-history information includes, for example, historical disease information and historical medical visit information within a continuous preset time period (eg, every three months).
  • the first sub-historical information is short-term historical information, which is closely related to the patient's current medical treatment situation, and the second sub-historical information is long-term historical information, which has time-series characteristics and can reflect periodic, seasonal symptoms, and regular medical treatment information.
  • Undifferentiated learning confuses the distinction between history and current conditions, improves the accuracy of descriptions of patient characteristics, and improves the accuracy of subsequent matching.
  • the first sub-historical information includes, for example, at least one item of discomfort site, discomfort duration, symptom performance, and severity of the latest historical disease information in the second preset time period, and may also include: the last visiting department, the last visiting doctor , the latest prescription information, the latest doctor's diagnosis result, the latest doctor's order, the length of the last doctor's visit, the length of time since the current doctor's visit, and the satisfaction with the doctor who consulted at least one of the most recent history in the second preset time period Consultation information.
  • the first sub-historical information can also be selected according to actual requirements, and is not limited to the examples.
  • the second sub-history information includes, for example: drug allergy history, common diseases in the first preset period (eg, 3-5 months), common diseases in the second preset period (eg, 6-8 months), and the third preset At least one item of historical disease information, such as common symptoms in the stage (for example, September to November), common symptoms in the fourth preset stage (for example, from December to February), and may also include: the first preset stage (for example, 3-2 months) May) frequently visited doctors, the second preset stage (eg June to August) frequently visited doctors, the third preset stage (eg September to November) frequently visited doctors, the fourth preset stage (eg December to February) ) Frequently visiting doctors, frequently visiting departments in the first preset stage (such as 3-5 months), frequently visiting departments in the second preset stage (such as June-August), and frequently visiting departments in the third preset stage (such as September-November) Visiting departments, frequently visited departments in the fourth preset stage (such as December to February), frequently bought medicines in the first preset stage (such as March to May),
  • the doctor-patient matching model can also include an offline part, and the patient's historical characteristics can be determined through the offline patient terminal to improve efficiency.
  • the offline patient side includes: a second natural language processing model, a first deep neural network model (DNN).
  • the second natural language processing model is, for example, a BERT model.
  • the first sub-history information is input into the second natural language processing model of the doctor-patient matching model, and the first sub-history feature vector is obtained as the first sub-history feature of the patient; the second sub-history information or the first sub-history information and the The joint information of the second sub-history information is input into the first deep neural network model of the doctor-patient matching model, and the second sub-history feature vector is obtained as the second sub-history feature of the patient.
  • the input of the first deep neural network model may only include the first sub-history information, or may include the first sub-history information and the second sub-history information.
  • Various information in the first sub-history information and the second sub-history information can be binary-coded in advance, and a preset length of the code is obtained and then input to the offline patient end.
  • the length of the first sub-history feature vector is greater than or equal to the length of the second sub-history feature vector. Since the first sub-history information has a certain correlation with the current matching business, it is expressed as the recent patient's condition and medical treatment status in the business, and the second sub-history information is expressed as the general description of the patient in the business. Therefore, the first sub-history information can be The length of the corresponding first sub-history feature vector is set longer to represent richer information. For example, the first sub-history feature vector is a 10-dimensional feature vector, and the second sub-history feature vector is a 5-dimensional feature vector.
  • the offline patient side further includes: a third fully connected layer. After concatenating the first sub-historical feature and the second sub-historical feature of the patient, input the third fully connected layer of the doctor-patient matching model to obtain the output patient's historical feature.
  • the vector corresponding to the patient's historical features may have the same length as the first sub-historical feature vector.
  • step S304 the doctor's history feature is generated according to the doctor's field information and the doctor's history information.
  • the doctor's field information is, for example, text information describing the doctor's field of expertise.
  • the doctor's historical information includes, for example, the doctor's historical consultation information within a preset time period, and the doctor's historical consultation information within the preset time period includes, for example, the average consultation time within a preset time period (for example, within the past three months).
  • the maximum consultation time in the preset time period, the minimum consultation time in the preset time period, the average consultation time in the preset time period, the average doctor-patient communication rounds in the preset time period, the doctor-patient consultation in the preset time period The maximum number of communication rounds, the minimum number of doctor-patient communication rounds within the preset time period, the average refusal rate of doctors within the preset time period, the average referral rate of doctors and patients within the preset time period, the favorable rate of doctors within the preset time period, The highest praise rate in the preset time period, the total number of orders received in the preset time period, the monthly average number of orders received in the preset time period, the total effective orders received in the preset time period, and the average effective orders received in the preset time period volume, average number of concurrences in a preset time period, average maximum number of concurrences in a preset time period, maximum number of medical assistants of doctors in a preset time period, minimum number of medical assistants of doctors in a preset time period, and number of doctors in a preset
  • At least one piece of historical consultation information such as the department with the longest consultation time and the department with the shortest average doctor consultation time within a preset time period.
  • the doctor's field information and the doctor's history information can be selected according to actual needs, and are not limited to the examples.
  • the doctor's field characteristics are determined according to the doctor's field information; the doctor's consultation characteristics are determined according to the doctor's historical information; and the doctor's historical characteristics are determined according to the doctor's field characteristics and consultation characteristics.
  • the doctor's historical characteristics can be determined through the offline doctor's terminal, which improves the efficiency.
  • the offline doctor terminal includes: a third natural language processing model and a second deep neural network model (DNN).
  • the third natural language processing model is, for example, the BERT model.
  • the doctor's domain information is input into the third natural language processing model of the doctor-patient matching model, and the domain feature vector is obtained as the doctor's domain feature; the doctor's historical information is input into the second deep neural network model of the doctor-patient matching model, and the connected
  • the consultation feature vector is used as the doctor's consultation feature.
  • Various information in the doctor's field information and the doctor's historical information can be binary coded in advance, and the code with a preset length can be input to the offline doctor terminal.
  • the length of the domain feature vector is greater than or equal to the length of the admissions feature vector. Because the doctor's field information has a certain correlation with the current matching business, it is expressed as the doctor's professional field in the business, and the doctor's historical information is expressed as the doctor's historical summary description in the business. Therefore, the doctor's field information can be corresponding to The length of the domain feature vector is set longer to represent richer information.
  • the domain feature vector is a 10-dimensional feature vector
  • the admission feature vector is a 5-dimensional feature vector.
  • the vector corresponding to the aforementioned fused real-time feature may be equal to the length of the second sub-history feature vector or the length of the consultation feature vector.
  • the aforementioned vector corresponding to the real-time condition feature of the patient may have the same length as the first sub-history feature vector and the domain feature vector.
  • the offline doctor terminal further includes: a fourth fully connected layer. After splicing the doctor's domain features and the admission features, input the fourth fully connected layer of the doctor-patient matching model to obtain the output doctor's historical features.
  • the length of the vector corresponding to the doctor's history feature may be equal to the length of the domain feature vector.
  • Steps S302 and 304 may be performed in parallel, in no particular order. Using steps S302 and 304, the historical characteristics of each patient and the historical characteristics of each doctor can be obtained.
  • step S306 a fused real-time feature is generated according to the real-time information of the patient and the real-time information of the doctor.
  • step S308 the real-time condition characteristics of the patient are generated according to the real-time condition description information of the patient.
  • Steps S306 and S308 may be performed in parallel, in no particular order.
  • step S310 the matching degree of the patient and the doctor is determined according to the historical characteristics of the patient and the historical characteristics of the doctor, as well as the fusion of the real-time characteristics and the real-time condition characteristics of the patient.
  • step S312 at least one doctor is pushed for the patient according to the degree of matching between the patient and each doctor.
  • the historical characteristics of patients and doctors can be determined offline and stored in the database (for example, online Redis), and the real-time characteristics of patients and the real-time condition characteristics of patients can be obtained online, and imported into the historical characteristics of patients and doctors stored in the database. After splicing, input the second fully connected layer to obtain the matching degree of the output patient and doctor.
  • the method of the above embodiment uses natural language processing to perform embedded coding on the description text information of the patient's real-time condition, the first sub-history information and the doctor's field information, which improves the matching accuracy between the patient and the doctor; Information and doctors' historical information are processed offline and feature compression, thereby reducing the feature dimension and improving the online model prediction performance; increasing the time-series related features of patients and improving the accuracy of push.
  • the present disclosure also provides an information push device, which will be described below with reference to FIG. 4 .
  • FIG. 4 is a structural diagram of some embodiments of the disclosed information pushing apparatus.
  • the apparatus 40 of this embodiment includes: a first generation module 410 , a second generation module 420 , a matching degree determination module 430 , and a push module 440 .
  • the first generation module 410 is configured to generate a fusion real-time feature according to the real-time information of the patient and the real-time information of the doctor, wherein the real-time information of the patient includes: the online access information of the patient in the preset first time period, and the real-time information of the doctor includes: The doctor's admission information within the preset first time period.
  • the first generation module 410 is configured to jointly input the real-time information of the patient and the real-time information of the doctor into the first fully connected layer of the doctor-patient matching model to obtain the output fused real-time feature; wherein the real-time information of the patient is also Including: the patient's personal characteristic information, the doctor's real-time information also includes the doctor's personal characteristic information and occupational characteristic information.
  • the second generating module 420 is configured to generate real-time condition characteristics of the patient according to the real-time condition description information of the patient.
  • the second generation module 420 is configured to input the real-time condition description information of the patient into the first natural language processing model of the doctor-patient matching model, and obtain the output real-time condition characteristics of the patient.
  • the matching degree determination module 430 is configured to determine the matching degree between the patient and the doctor according to the pre-acquired historical characteristics of the patient and the historical characteristics of the doctor, as well as combining the real-time characteristics and the real-time condition characteristics of the patient.
  • the matching degree determination module 430 is configured to splicing the pre-acquired historical characteristics of the patient and the historical characteristics of the doctor, as well as the fusion real-time characteristics and the real-time condition characteristics of the patient, and then inputting the second whole of the doctor-patient matching model. Connect the layer to get the matching degree of the output patient and doctor.
  • the push module 440 is configured to push at least one doctor for the patient according to the matching degree between the patient and each doctor.
  • the apparatus 40 further includes: a third generating module 450, configured to generate the patient's historical characteristics according to the patient's historical information, wherein the patient's historical information includes the patient's historical condition information and historical medical visit information.
  • a third generating module 450 configured to generate the patient's historical characteristics according to the patient's historical information, wherein the patient's historical information includes the patient's historical condition information and historical medical visit information.
  • the third generation module 450 is configured to divide the patient's history information into first sub-history information and second sub-history information in chronological order, wherein the first sub-history information includes a preset second time period The most recent historical information, the second sub-historical information includes historical information in different time periods divided in chronological order; the first sub-historical characteristics of the patient are determined according to the first sub-historical information; the patient’s The second sub-historical feature; according to the first sub-historical feature and the second sub-historical feature of the patient, the historical feature of the patient is determined.
  • the third generation module 450 is configured to input the first sub-history information into the second natural language processing model of the doctor-patient matching model, and obtain the first sub-history feature vector as the first sub-history feature of the patient;
  • the second sub-history information or the joint information of the first sub-history information and the second sub-history information is input into the first deep neural network model of the doctor-patient matching model, and the second sub-history feature vector is obtained as the second sub-history feature of the patient; wherein, The length of the first sub-history feature vector is greater than or equal to the length of the second sub-history feature vector.
  • the third generation module 450 is configured to input the third fully connected layer of the doctor-patient matching model after splicing the first sub-historical feature and the second sub-historical feature of the patient to obtain the output historical feature of the patient .
  • the apparatus 40 further includes: a fourth generating module 460, configured to generate a doctor's history feature according to the doctor's field information and the doctor's historical information, wherein the doctor's historical information includes the doctor's historical consultation information.
  • a fourth generating module 460 configured to generate a doctor's history feature according to the doctor's field information and the doctor's historical information, wherein the doctor's historical information includes the doctor's historical consultation information.
  • the fourth generation module 460 is configured to determine the doctor's field characteristics according to the doctor's field information; determine the doctor's consultation characteristics according to the doctor's historical information; determine the doctor's historical characteristics according to the doctor's field characteristics and consultation characteristics .
  • the fourth generation module 460 is configured to input the doctor's domain information into the third natural language processing model of the doctor-patient matching model, and obtain a domain feature vector as the doctor's domain feature; input the doctor's historical information into the doctor-patient matching model
  • the second deep neural network model of the model obtains the consultation feature vector as the doctor's consultation feature; wherein, the length of the domain feature vector is greater than or equal to the length of the consultation feature vector.
  • the fourth generation module 460 is configured to input the doctor-patient matching model to the fourth fully-connected layer after splicing the doctor's domain feature and the consultation feature to obtain the output doctor's historical feature.
  • the information pushing apparatuses in the embodiments of the present disclosure may each be implemented by various computing devices or computer systems, which will be described below with reference to FIG. 5 and FIG. 6 .
  • FIG. 5 is a structural diagram of some embodiments of the disclosed information pushing apparatus.
  • the apparatus 50 of this embodiment includes a memory 510 and a processor 520 coupled to the memory 510, the processor 520 being configured to execute any of the implementations of the present disclosure based on instructions stored in the memory 510 The information push method in the example.
  • the memory 510 may include, for example, a system memory, a fixed non-volatile storage medium, and the like.
  • the system memory stores, for example, an operating system, an application program, a boot loader (Boot Loader), a database, and other programs.
  • FIG. 6 is a structural diagram of other embodiments of the disclosed information pushing apparatus.
  • the apparatus 60 of this embodiment includes: a memory 610 and a processor 620 , which are similar to the memory 510 and the processor 520 respectively. It may also include an input and output interface 630, a network interface 640, a storage interface 650, and the like. These interfaces 630 , 640 , 650 and the memory 610 and the processor 620 can be connected, for example, through a bus 660 .
  • the input and output interface 630 provides a connection interface for input and output devices such as a display, a mouse, a keyboard, and a touch screen.
  • the network interface 640 provides a connection interface for various networked devices, for example, it can be connected to a database server or a cloud storage server.
  • the storage interface 650 provides a connection interface for external storage devices such as SD cards and U disks.
  • embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein .
  • computer-usable non-transitory storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions
  • the apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.
  • These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps configured to implement the functions specified in a flow or flows of the flowcharts and/or a block or blocks of the block diagrams.

Landscapes

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

Abstract

本公开涉及一种信息推送方法, 装置和计算机可读存储介质, 涉及计算机技术领域. 本公开的方法包括: 根据患者的实时信息和医生的实时信息生成融合实时特征, 其中, 患者的实时信息包括: 患者在预设第一时间段内的在线访问信息, 医生的实时信息包括: 医生在预设第一时间段内的接诊信息; 根据患者的实时病情描述信息生成患者的实时病情特征; 根据预先获取的患者的历史特征和医生的历史特征, 以及融合实时特征和患者的实时病情特征, 确定患者和医生的匹配度; 根据患者与每个医生的匹配度, 为患者推送至少一个医生.

Description

信息推送方法、装置和计算机可读存储介质
相关申请的交叉引用
本申请是以CN申请号为202110331602.5，申请日为2021年3月29日的申请为基础，并主张其优先权，该CN申请的公开内容在此作为整体引入本申请中。
技术领域
本公开涉及计算机技术领域，特别涉及一种信息推送方法、装置和计算机可读存储介质。
背景技术
随着互联网技术的发展，用户的生活变得越来越方便。用户可以通过互联网社交、购物、观看视频等。
互联网医院作为新兴的互联网应用平台，实现了为用户网上看病、诊断、开药等，方便了用户的就诊。目前，用户在互联网医院一般通过与系统交互，简单的描述病情或者选择就诊的科室，系统为用户分配医生进行接诊。
发明内容
根据本公开的一些实施例，提供的一种信息推送方法，包括：根据患者的实时信息和医生的实时信息生成融合实时特征，其中，患者的实时信息包括：患者在预设第一时间段内的在线访问信息，医生的实时信息包括：医生在预设第一时间段内的接诊信息；根据患者的实时病情描述信息生成患者的实时病情特征；根据预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度；根据患者与每个医生的匹配度，为患者推送至少一个医生。
在一些实施例中，根据患者的实时信息和医生的实时信息生成融合实时特征包括：将患者的实时信息和医生的实时信息共同输入医患匹配模型的第一全连接层，得到输出的融合实时特征；其中，患者的实时信息还包括：患者的个人特征信息，医生的实时信息还包括医生的个人特征信息和职业特征信息。
在一些实施例中，根据患者的实时病情描述信息生成患者的实时病情特征包括：将患者的实时病情描述信息输入医患匹配模型的第一自然语言处理模型，得到输出的 患者的实时病情特征。
在一些实施例中，根据预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度包括：将预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征进行拼接后，输入医患匹配模型的第二全连接层，得到输出的患者和医生的匹配度。
在一些实施例中，该方法还包括：根据患者的历史信息生成患者的历史特征，其中，患者的历史信息包括患者的历史病症信息和历史就诊信息。
在一些实施例中，根据患者的历史信息生成患者的历史特征包括：按时间顺序将患者的历史信息划分为第一子历史信息和第二子历史信息，其中，第一子历史信息包括预设第二时间段内的最近一次的历史信息，第二子历史信息包括按时间顺序划分的不同时间段内的历史信息；根据第一子历史信息确定患者的第一子历史特征；根据第二子历史信息确定患者的第二子历史特征；根据患者的第一子历史特征和第二子历史特征，确定患者的历史特征。
在一些实施例中，根据第一子历史信息确定患者的第一子历史特征包括：将第一子历史信息输入医患匹配模型的第二自然语言处理模型，得到第一子历史特征向量作为患者的第一子历史特征；根据第二子历史信息确定患者的第二子历史特征包括：将第二子历史信息或者第一子历史信息和第二子历史信息的联合信息输入医患匹配模型的第一深度神经网络模型，得到第二子历史特征向量作为患者的第二子历史特征；其中，第一子历史特征向量的长度大于或等于第二子历史特征向量的长度。
在一些实施例中，根据患者的第一子历史特征和第二子历史特征，确定患者的历史特征包括：将患者的第一子历史特征和第二子历史特征进行拼接后，输入医患匹配模型的第三全连接层，得到输出的患者的历史特征。
在一些实施例中，该方法还包括：根据医生的领域信息和医生的历史信息生成医生的历史特征，其中，医生的历史信息包括医生的历史接诊信息。
在一些实施例中，根据医生的领域信息和医生的历史信息生成医生的历史特征包括：根据医生的领域信息确定医生的领域特征；根据医生的历史信息确定医生的接诊特征；根据医生的领域特征和接诊特征确定医生的历史特征。
在一些实施例中，根据医生的领域信息确定医生的领域特征包括：将医生的领域信息输入医患匹配模型的第三自然语言处理模型，得到领域特征向量作为医生的领域特征；根据医生的历史信息确定医生的接诊特征包括：将医生的历史信息输入医患匹 配模型的第二深度神经网络模型，得到接诊特征向量作为医生的接诊特征；其中，领域特征向量的长度大于或等于接诊特征向量的长度。
在一些实施例中，根据医生的领域特征和接诊特征确定医生的历史特征包括：将医生的领域特征和接诊特征进行拼接后，输入医患匹配模型的第四全连接层，得到输出的医生的历史特征。
根据本公开的另一些实施例，提供的一种信息推送装置，包括：第一生成模块，用于根据患者的实时信息和医生的实时信息生成融合实时特征，其中，患者的实时信息包括：患者在预设第一时间段内的在线访问信息，医生的实时信息包括：医生在预设第一时间段内的接诊信息；第二生成模块，用于根据患者的实时病情描述信息生成患者的实时病情特征；匹配度确定模块，用于根据预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度；推送模块，用于根据患者与每个医生的匹配度，为患者推送至少一个医生。
根据本公开的再一些实施例，提供的一种信息推送装置，包括：处理器；以及耦接至处理器的存储器，用于存储指令，指令被处理器执行时，使处理器执行如前述任意实施例的信息推送方法。
根据本公开的又一些实施例，提供的一种非瞬时性计算机可读存储介质，其上存储有计算机程序，其中，该程序被处理器执行时实现前述任意实施例的信息推送方法。
通过以下参照附图对本公开的示例性实施例的详细描述，本公开的其它特征及其优点将会变得清楚。
附图说明
此处所说明的附图用来提供对本公开的进一步理解，构成本申请的一部分，本公开的示意性实施例及其说明被配置为解释本公开，并不构成对本公开的不当限定。
图1示出本公开的一些实施例的信息推送方法的流程示意图。
图2示出本公开的一些实施例的医患匹配模型的结构示意图。
图3示出本公开的另一些实施例的信息推送方法的流程示意图。
图4示出本公开的一些实施例的信息推送装置的结构示意图。
图5示出本公开的另一些实施例的信息推送装置的结构示意图。
图6示出本公开的又一些实施例的信息推送装置的结构示意图。
具体实施方式
下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。以下对至少一个示例性实施例的描述实际上仅仅是说明性的，决不作为对本公开及其应用或使用的任何限制。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本公开保护的范围。
发明人发现：互联网医院中为用户分配医生接诊，仅参考用户简单的描述病情或者选择就诊的科室，信息很少且方法简单，在用户不能清楚描述病情或者无法准确选择就诊科室时，不能准确的确定用户和医生的匹配度，从而不能准确的为用户匹配和分配医生。
本公开所要解决的一个技术问题是：如何为患者更加准确地匹配或分配医生。
本公开提供一种信息推送方法，下面结合图1～3进行描述。
图1为本公开信息推送方法一些实施例的流程图。如图1所示，该实施例的方法包括：步骤S102～S108。
在步骤S102中，根据患者的实时信息和医生的实时信息生成融合实时特征。
患者的实时信息例如包括：患者在预设第一时间段内(例如，当天)的在线访问信息，患者的实时信息还可以还包括：患者的个人特征信息。医生的实时信息例如包括：医生在预设第一时间段内的接诊信息，医生的实时信息还可以包括医生的个人特征信息和职业特征信息。
患者在预设第一时间段内的在线访问信息例如包括：在预设第一时间段内浏览的药品、在预设第一时间段内浏览的科室、在预设第一时间段内浏览的医生、在预设第一时间段内分诊的科室等至少一项信息。患者的个人特征信息例如包括：性别、年龄等至少一项信息。医生在预设第一时间段内的接诊信息例如包括：在预设第一时间段内是否在线、在预设第一时间段内在线状态、在预设第一时间段内医生所在科室、在预设第一时间段内是否存在拒诊，在预设第一时间段内接诊量、在预设第一时间段内已上班时间、距上次接诊结束时长、在预设第一时间段内最近一单接单时长、在预设第一时间段内累计接诊时长、在预设第一时间段内系统催促次数、在预设第一时间段内系统第一次催促后多久接单等至少一项信息。医生的个人特征信息例如包括：性别、年龄等至少一项信息。医生的职业特征信息例如包括：职级、是否专家等至少一项信息。患者的实时信息和医生的实时信息可以根据实际需求进行选取，不限于所举示例。
图2为医患匹配模型的一些实施例的示意图。如图2所示，医患匹配模型可以包括在线部分，融合实时特征可以通过在线部分确定。在线部分包括第一全连接层。在一些实施例中，将患者的实时信息和医生的实时信息共同输入医患匹配模型的第一全连接层，得到输出的融合实时特征。融合实时特征是医生和患者的融合特征，患者的实时信息和医生的实时信息中各种信息可以预先进行二进制编码，得到预设长度的编码后输入第一全连接层。
在步骤S104中，根据患者的实时病情描述信息生成患者的实时病情特征。
患者的实时病情描述信息例如为患者在系统输入的描述病情的文本信息，例如，症状表现、不适部位等。在一些实施例中，根据患者的实时病情描述信息和患者在预设第一时间段内在线访问信息生成患者的实时病情特征。
如图2所示，患者的实时病情特征可以通过在线部分确定。在线部分包括第一自然语言处理模型，第一自然语言处理模型例如为BERT模型。在一些实施例中，将患者的实时病情描述信息输入医患匹配模型的第一自然语言处理模型，得到输出的患者的实时病情特征。在一些实施例中，患者的实时病情特征对应的向量的长度大于或等于融合实时特征对应的向量的长度。患者的实时病情特征与患者本次就诊密切相关，将对应的向量设置更长的维度，有利于更丰富的信息的表达。
在步骤S106中，根据预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度。
如图2所示，医患匹配模型还包括第二全连接层，在一些实施例中，将患者的历史特征、医生的历史特征、融合实时特征和患者的实时病情特征进行拼接后输入医患匹配模型的第二全连接层，得到输出的患者和医生的匹配度。
医患匹配模型可以预先训练，选取患者和医生对应的训练样本，并根据实际的就诊信息选取确定患者和医生匹配度进行标注。训练样本包括：多个患者的历史信息、实时信息和实时病情描述信息，多个医生的领域信息、历史信息和实时信息。将训练样本输入医患匹配模型，得到输出的患者和医生匹配度，根据标注的患者和医生匹配度与输出患者和医生匹配度确定损失函数，根据损失函数调整医患匹配模型的参数，直至达到预设收敛条件。训练样本中各种患者的和医生的信息，分别输入医患匹配模型中对应的部分(如图2所示)，在此不再赘述。
在步骤S108，根据患者与每个医生的匹配度，为患者推送至少一个医生。
根据前述的方法可以确定患者和每个医生的匹配度，可以选取匹配度超过阈值的 至少一个医生推送给患者，由患者选择，也可以按照匹配度由高到低排序，选择排序在先的至少一个医生推送给患者，由患者选择。也可以选取匹配度最高的医生，对患者进行分诊。
上述实施例中根据患者的实时信息和医生的实时信息生成融合实时特征，还根据患者的实时病情描述信息生成患者的实时病情特征，进一步根据预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度。上述实施例参考患者和医生的历史信息和实时信息多方面信息，这些信息能够更加准确的描述患者和医生的特征，提高确定患者和医生的匹配度的准确性，进而提高为患者匹配或分配医生的准确性。
下面结合图3描述本公开的信息推送方法的另一些实施例。
图3为本公开信息推送方法另一些实施例的流程图。如图3所示，该实施例的方法包括：步骤S302～S312。
在步骤S302中，根据患者的历史信息生成患者的历史特征。
患者的历史信息例如包括患者的历史病症信息和历史就诊信息。在一些实施例中，按时间顺序将患者的历史信息划分为第一子历史信息和第二子历史信息，其中，第一子历史信息包括预设第二时间段内的最近一次的历史信息，第二子历史信息包括按时间顺序划分的不同时间段内的历史信息；根据第一子历史信息确定患者的第一子历史特征；根据第二子历史信息确定患者的第二子历史特征；根据患者的第一子历史特征和第二子历史特征，确定患者的历史特征。
第一子历史信息例如包括2个月、1个月或15天(预设第二时间段内)等最近一次的历史病症信息和历史就诊信息。第二子历史信息例如包括连续每隔预设时长内(例如每三个月内)的历史病症信息和历史就诊信息。第一子历史信息为短期历史信息，与患者当前就诊的情况密切相关，第二子历史信息为长期历史信息，具有时序性特征，可以反映周期性、季节性病症、规律性的就诊信息等。对第一子历史信息和第二子历史信息进行区分，分别确定第一子历史特征和第二子历史特征，可以更加准确的提取患者的短期特征和长期特征，避免对患者短期与长期历史信息无差别学习会混淆历史与当前病情的区分，提高对患者特征的描述准确性，提高后续匹配的准确性。
第一子历史信息例如包括：不适部位、不适时长、症状表现、严重程度中至少一项预设第二时间段内最近一次的历史病症信息，还可以包括：最近一次就诊科室、最近一次就诊医生、最近一次开药信息、最近一次医生诊断结果、最近一次医嘱、最近 一次就诊时长、距离本次就诊时长、对接诊医生的满意度中至少一项预设第二时间段内最近一次的历史就诊信息。第一子历史信息还可以根据实际需求进行选取，不限于所举示例。
第二子历史信息例如包括：药物过敏史、第一预设阶段(例如3～5月)常患病症、第二预设阶段(例如6～8月)常患病症、第三预设阶段(例如9～11月)常患病症、第四预设阶段(例如12～2月)常患病症等至少一项历史病症信息，还可以包括：第一预设阶段(例如3～5月)常就诊医生、第二预设阶段(例如6～8月)常就诊医生、第三预设阶段(例如9～11月)常就诊医生、第四预设阶段(例如12～2月)常就诊医生、第一预设阶段(例如3～5月)常就诊科室、第二预设阶段(例如6～8月)常就诊科室、第三预设阶段(例如9～11月)常就诊科室、第四预设阶段(例如12～2月)常就诊科室、第一预设阶段(例如3～5月)常买的药、第二预设阶段(例如6～8月)常买的药、第三预设阶段(例如9～11月)常买的药、第四预设阶段(例如12～2月)常买的药、月就诊时长等至少一项历史就诊信息。第二子历史信息还可以根据实际需求进行选取，不限于所举示例。
如图2所示，医患匹配模型还可以包括离线部分，患者的历史特征可以通过离线的患者端确定，提高效率。在一些实施例中，离线的患者端包括：第二自然语言处理模型、第一深度神经网络模型(DNN)。第二自然语言处理模型例如为BERT模型。例如，将第一子历史信息输入医患匹配模型的第二自然语言处理模型，得到第一子历史特征向量作为患者的第一子历史特征；将第二子历史信息或者第一子历史信息和第二子历史信息的联合信息输入医患匹配模型的第一深度神经网络模型，得到第二子历史特征向量作为患者的第二子历史特征。第一深度神经网络模型的输入可以只包括第一子历史信息，也可以包括第一子历史信息和第二子历史信息。第一子历史信息和第二子历史信息中各种信息可以预先进行二进制编码，得到预设长度的编码后输入离线的患者端。
在一些实施例中，第一子历史特征向量的长度大于或等于第二子历史特征向量的长度。由于第一子历史信息和当前匹配业务有一定相关性，在业务上表示为近期患者病情和就诊状态，第二子历史信息在业务上表示为患者的概括描述，因此可以将第一子历史信息对应的第一子历史特征向量的长度设置的更长，用来表示更丰富的信息。例如，第一子历史特征向量为10维特征向量，第二子历史特征向量为5维特征向量。
进一步，在一些实施例中，离线的患者端还包括：第三全连接层。将患者的第一 子历史特征和第二子历史特征进行拼接(concat)后，输入医患匹配模型的第三全连接层，得到输出的患者的历史特征。患者的历史特征对应的向量可以与第一子历史特征向量的长度相等。
在步骤S304中，根据医生的领域信息和医生的历史信息生成医生的历史特征。
医生的领域信息例如为描述医生的擅长领域等的文本信息。医生的历史信息例如包括医生在预设时间段内的历史接诊信息，医生在预设时间段内的历史接诊信息例如包括预设时间段内(例如近三个月内)平均接诊时长、预设时间段内最大接诊时长、预设时间段内最小接诊时长、预设时间段内平均问诊时长、预设时间段内医患平均沟通轮次、预设时间段内医患最大沟通轮次、预设时间段内医患最小沟通轮次、预设时间段内医生平均拒诊率、预设时间段内医患平均转诊率、预设时间段内医生的好评率、预设时间段内最高好评率、预设时间段内总接单量、预设时间段内月平均接单量、预设时间段内总有效接单量、预设时间段内平均有效接单量、预设时间段内平均并发数、预设时间段内平均最大并发数、预设时间段内医生最多医助数、预设时间段内医生最少医助数、预设时间段内医生的活跃天数、预设时间段内医生的活跃总时长、预设时间段内医生接诊单量最多的科室、预设时间段内医生接诊单量最少的科室、预设时间段内医生平均接诊时长最长的科室、预设时间段内医生平均接诊时长最短的科室等至少一项历史接诊信息。医生的领域信息、医生的历史信息可以根据实际需求进行选取，不限于所举示例。
在一些实施例中，根据医生的领域信息确定医生的领域特征；根据医生的历史信息确定医生的接诊特征；根据医生的领域特征和接诊特征确定医生的历史特征。
如图2所示，医生的历史特征可以通过离线的医生端确定，提高效率。在一些实施例中，离线的医生端包括：第三自然语言处理模型、第二深度神经网络模型(DNN)。第三自然语言处理模型例如为BERT模型。例如，将医生的领域信息输入医患匹配模型的第三自然语言处理模型，得到领域特征向量作为医生的领域特征；将医生的历史信息输入医患匹配模型的第二深度神经网络模型，得到接诊特征向量作为医生的接诊特征。医生的领域信息和医生的历史信息中各种信息可以预先进行二进制编码，得到预设长度的编码后输入离线的医生端。
在一些实施例中，领域特征向量的长度大于或等于接诊特征向量的长度。由于医生的领域信息和当前匹配业务有一定相关性，在业务上表示为医生的专业领域，医生的历史信息在业务上表示为医生的历史性的概括描述，因此可以将医生的领域信息对 应的领域特征向量的长度设置的更长，用来表示更丰富的信息。例如，领域特征向量为10维特征向量，接诊特征向量为5维特征向量。前述融合实时特征对应的向量可以与第二子历史特征向量的长度或接诊特征向量的长度相等。前述的患者的实时病情特征对应的向量可以与第一子历史特征向量、领域特征向量的长度相等。
进一步，在一些实施例中，离线的医生端还包括：第四全连接层。将医生的领域特征和接诊特征进行拼接后，输入医患匹配模型的第四全连接层，得到输出的医生的历史特征。医生的历史特征对应的向量可以与领域特征向量的长度相等。
步骤S302和304可以并列执行，不分先后顺序。利用步骤S302和304可以获得每个患者的历史特征和每个医生的历史特征。
在步骤S306中，根据患者的实时信息和医生的实时信息生成融合实时特征。
在步骤S308中，根据患者的实时病情描述信息生成患者的实时病情特征。
步骤S306和S308可以并列执行，不分先后顺序。
在步骤S310中，根据患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度。
在步骤S312中，根据患者与每个医生的匹配度，为患者推送至少一个医生。
患者的历史特征、医生的历史特征可以离线确定后存入数据库(例如，线上Redis)，在线得到融合实时特征和患者的实时病情特征，导入数据库中存储的患者的历史特征、医生的历史特征进行拼接后输入第二全连接层，得到输出的患者和医生的匹配度。
上述实施例的方法，利用自然语言处理将患者实时病情的描述文本信息、第一子历史信息和医生领域信息进行嵌入式编码，提高了患者与医生的匹配准确性；将患者的第二子历史信息和医生的历史信息进行离线处理和特征压缩，从而降低了特征维度，提高线上模型预测性能；增加患者时序性相关特征，提高了推送的准确性。
本公开还提供一种信息推送装置，下面结合图4进行描述。
图4为本公开信息推送装置的一些实施例的结构图。如图4所示，该实施例的装置40包括：第一生成模块410，第二生成模块420，匹配度确定模块430，推送模块440。
第一生成模块410用于根据患者的实时信息和医生的实时信息生成融合实时特征，其中，患者的实时信息包括：患者在预设第一时间段内的在线访问信息，医生的实时信息包括：医生在预设第一时间段内的接诊信息。
在一些实施例中，第一生成模块410用于将患者的实时信息和医生的实时信息共 同输入医患匹配模型的第一全连接层，得到输出的融合实时特征；其中，患者的实时信息还包括：患者的个人特征信息，医生的实时信息还包括医生的个人特征信息和职业特征信息。
第二生成模块420用于根据患者的实时病情描述信息生成患者的实时病情特征。
在一些实施例中，第二生成模块420用于将患者的实时病情描述信息输入医患匹配模型的第一自然语言处理模型，得到输出的患者的实时病情特征。
匹配度确定模块430用于根据预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征，确定患者和医生的匹配度。
在一些实施例中，匹配度确定模块430用于将预先获取的患者的历史特征和医生的历史特征，以及融合实时特征和患者的实时病情特征进行拼接后，输入医患匹配模型的第二全连接层，得到输出的患者和医生的匹配度。
推送模块440用于根据患者与每个医生的匹配度，为患者推送至少一个医生。
在一些实施例中，装置40还包括：第三生成模块450，用于根据患者的历史信息生成患者的历史特征，其中，患者的历史信息包括患者的历史病症信息和历史就诊信息。
在一些实施例中，第三生成模块450用于按时间顺序将患者的历史信息划分为第一子历史信息和第二子历史信息，其中，第一子历史信息包括预设第二时间段内的最近一次的历史信息，第二子历史信息包括按时间顺序划分的不同时间段内的历史信息；根据第一子历史信息确定患者的第一子历史特征；根据第二子历史信息确定患者的第二子历史特征；根据患者的第一子历史特征和第二子历史特征，确定患者的历史特征。
在一些实施例中，第三生成模块450用于将第一子历史信息输入医患匹配模型的第二自然语言处理模型，得到第一子历史特征向量作为患者的第一子历史特征；将第二子历史信息或者第一子历史信息和第二子历史信息的联合信息输入医患匹配模型的第一深度神经网络模型，得到第二子历史特征向量作为患者的第二子历史特征；其中，第一子历史特征向量的长度大于或等于第二子历史特征向量的长度。
在一些实施例中，第三生成模块450用于将患者的第一子历史特征和第二子历史特征进行拼接后，输入医患匹配模型的第三全连接层，得到输出的患者的历史特征。
在一些实施例中，装置40还包括：第四生成模块460，用于根据医生的领域信息和医生的历史信息生成医生的历史特征，其中，医生的历史信息包括医生的历史接诊信息。
在一些实施例中，第四生成模块460用于根据医生的领域信息确定医生的领域特征；根据医生的历史信息确定医生的接诊特征；根据医生的领域特征和接诊特征确定医生的历史特征。
在一些实施例中，第四生成模块460用于将医生的领域信息输入医患匹配模型的第三自然语言处理模型，得到领域特征向量作为医生的领域特征；将医生的历史信息输入医患匹配模型的第二深度神经网络模型，得到接诊特征向量作为医生的接诊特征；其中，领域特征向量的长度大于或等于接诊特征向量的长度。
在一些实施例中，第四生成模块460用于将医生的领域特征和接诊特征进行拼接后，输入医患匹配模型的第四全连接层，得到输出的医生的历史特征。
本公开的实施例中的信息推送装置可各由各种计算设备或计算机系统来实现，下面结合图5以及图6进行描述。
图5为本公开信息推送装置的一些实施例的结构图。如图5所示，该实施例的装置50包括：存储器510以及耦接至该存储器510的处理器520，处理器520被配置为基于存储在存储器510中的指令，执行本公开中任意一些实施例中的信息推送方法。
其中，存储器510例如可以包括系统存储器、固定非易失性存储介质等。系统存储器例如存储有操作系统、应用程序、引导装载程序(Boot Loader)、数据库以及其他程序等。
图6为本公开信息推送装置的另一些实施例的结构图。如图6所示，该实施例的装置60包括：存储器610以及处理器620，分别与存储器510以及处理器520类似。还可以包括输入输出接口630、网络接口640、存储接口650等。这些接口630，640，650以及存储器610和处理器620之间例如可以通过总线660连接。其中，输入输出接口630为显示器、鼠标、键盘、触摸屏等输入输出设备提供连接接口。网络接口640为各种联网设备提供连接接口，例如可以连接到数据库服务器或者云端存储服务器等。存储接口650为SD卡、U盘等外置存储设备提供连接接口。
本领域内的技术人员应当明白，本公开的实施例可提供为方法、系统、或计算机程序产品。因此，本公开可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用非瞬时性存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本公开是参照根据本公开实施例的方法、设备(系统)、和计算机程序产品的流 程图和/或方框图来描述的。应理解为可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生被配置为实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供被配置为实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
以上所述仅为本公开的较佳实施例，并不用以限制本公开，凡在本公开的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。

Claims (15)

  1. 一种信息推送方法，包括：
    根据患者的实时信息和医生的实时信息生成融合实时特征，其中，所述患者的实时信息包括：所述患者在预设第一时间段内的在线访问信息，所述医生的实时信息包括：所述医生在预设第一时间段内的接诊信息；
    根据所述患者的实时病情描述信息生成所述患者的实时病情特征；
    根据预先获取的所述患者的历史特征和所述医生的历史特征，以及所述融合实时特征和所述患者的实时病情特征，确定所述患者和所述医生的匹配度；
    根据所述患者与每个医生的匹配度，为所述患者推送至少一个医生。
  2. 根据权利要求1所述的信息推送方法，其中，所述根据所述患者的实时信息和所述医生的实时信息生成融合实时特征包括：
    将所述患者的实时信息和所述医生的实时信息共同输入医患匹配模型的第一全连接层，得到输出的所述融合实时特征；
    其中，所述患者的实时信息还包括：所述患者的个人特征信息，所述医生的实时信息还包括所述医生的个人特征信息和职业特征信息。
  3. 根据权利要求1所述的信息推送方法，其中，所述根据所述患者的实时病情描述信息生成所述患者的实时病情特征包括：
    将所述患者的实时病情描述信息输入医患匹配模型的第一自然语言处理模型，得到输出的所述患者的实时病情特征。
  4. 根据权利要求1所述的信息推送方法，其中，所述根据预先获取的所述患者的历史特征和所述医生的历史特征，以及所述融合实时特征和所述患者的实时病情特征，确定所述患者和所述医生的匹配度包括：
    将预先获取的所述患者的历史特征和所述医生的历史特征，以及所述融合实时特征和所述患者的实时病情特征进行拼接后，输入医患匹配模型的第二全连接层，得到输出的所述患者和所述医生的匹配度。
  5. 根据权利要求1所述的信息推送方法，还包括：
    根据所述患者的历史信息生成所述患者的历史特征，其中，所述患者的历史信息包括所述患者的历史病症信息和历史就诊信息。
  6. 根据权利要求5所述的信息推送方法，其中，所述根据患者的历史信息生成所述患者的历史特征包括：
    按时间顺序将所述患者的历史信息划分为第一子历史信息和第二子历史信息，其中，所述第一子历史信息包括预设第二时间段内的最近一次的历史信息，所述第二子历史信息包括按时间顺序划分的不同时间段内的历史信息；
    根据所述第一子历史信息确定所述患者的第一子历史特征；
    根据所述第二子历史信息确定所述患者的第二子历史特征；
    根据所述患者的第一子历史特征和第二子历史特征，确定所述患者的历史特征。
  7. 根据权利要求6所述的信息推送方法，其中，所述根据所述第一子历史信息确定所述患者的第一子历史特征包括：
    将所述第一子历史信息输入医患匹配模型的第二自然语言处理模型，得到第一子历史特征向量作为所述患者的第一子历史特征；
    所述根据所述第二子历史信息确定所述患者的第二子历史特征包括：
    将所述第二子历史信息或者所述第一子历史信息和第二子历史信息的联合信息输入所述医患匹配模型的第一深度神经网络模型，得到第二子历史特征向量作为所述患者的第二子历史特征；
    其中，所述第一子历史特征向量的长度大于或等于所述第二子历史特征向量的长度。
  8. 根据权利要求6所述的信息推送方法，其中，所述根据所述患者的第一子历史特征和第二子历史特征，确定所述患者的历史特征包括：
    将所述患者的第一子历史特征和第二子历史特征进行拼接后，输入医患匹配模型的第三全连接层，得到输出的所述患者的历史特征。
  9. 根据权利要求1所述的信息推送方法，还包括：
    根据所述医生的领域信息和所述医生的历史信息生成所述医生的历史特征，其中，所述医生的历史信息包括所述医生的历史接诊信息。
  10. 根据权利要求9所述的信息推送方法，其中，所述根据医生的领域信息和所述医生的历史信息生成所述医生的历史特征包括：
    根据所述医生的领域信息确定所述医生的领域特征；
    根据所述医生的历史信息确定所述医生的接诊特征；
    根据所述医生的领域特征和接诊特征确定所述医生的历史特征。
  11. 根据权利要求10所述的信息推送方法，其中，所述根据所述医生的领域信息确定所述医生的领域特征包括：
    将所述医生的领域信息输入医患匹配模型的第三自然语言处理模型，得到领域特征向量作为所述医生的领域特征；
    所述根据所述医生的历史信息确定所述医生的接诊特征包括：
    将所述医生的历史信息输入医患匹配模型的第二深度神经网络模型，得到接诊特征向量作为所述医生的接诊特征；
    其中，所述领域特征向量的长度大于或等于所述接诊特征向量的长度。
  12. 根据权利要求10所述的信息推送方法，其中，所述根据所述医生的领域特征和接诊特征确定所述医生的历史特征包括：
    将所述医生的领域特征和接诊特征进行拼接后，输入医患匹配模型的第四全连接层，得到输出的所述医生的历史特征。
  13. 一种信息推送装置，包括：
    第一生成模块，用于根据患者的实时信息和医生的实时信息生成融合实时特征，其中，所述患者的实时信息包括：所述患者在预设第一时间段内的在线访问信息，所述医生的实时信息包括：所述医生在预设第一时间段内的接诊信息；
    第二生成模块，用于根据所述患者的实时病情描述信息生成所述患者的实时病情特征；
    匹配度确定模块，用于根据预先获取的所述患者的历史特征和所述医生的历史特 征，以及所述融合实时特征和所述患者的实时病情特征，确定所述患者和所述医生的匹配度；
    推送模块，用于根据所述患者与每个医生的匹配度，为所述患者推送至少一个医生。
  14. 一种信息推送装置，包括：
    处理器；以及
    耦接至所述处理器的存储器，用于存储指令，所述指令被所述处理器执行时，使所述处理器执行如权利要求1-12任一项所述的信息推送方法。
  15. 一种非瞬时性计算机可读存储介质，其上存储有计算机程序，其中，该程序被处理器执行时实现权利要求1-12任一项所述方法的步骤。
PCT/CN2022/082428 2021-03-29 2022-03-23 信息推送方法, 装置和计算机可读存储介质 WO2022206500A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110331602.5A CN113782163A (zh) 2021-03-29 2021-03-29 信息推送方法、装置和计算机可读存储介质
CN202110331602.5 2021-03-29

Publications (1)

Publication Number Publication Date
WO2022206500A1 true WO2022206500A1 (zh) 2022-10-06

Family

ID=78835548

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/082428 WO2022206500A1 (zh) 2021-03-29 2022-03-23 信息推送方法, 装置和计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN113782163A (zh)
WO (1) WO2022206500A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113782163A (zh) * 2021-03-29 2021-12-10 北京京东拓先科技有限公司 信息推送方法、装置和计算机可读存储介质
CN117171407A (zh) * 2023-11-03 2023-12-05 山东中联佳裕软件股份有限公司 一种基于大数据分析的智慧平台管理系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160147943A1 (en) * 2014-11-21 2016-05-26 Argo Data Resource Corporation Semantic Address Parsing Using a Graphical Discriminative Probabilistic Model
CN110752017A (zh) * 2019-09-04 2020-02-04 重庆特斯联智慧科技股份有限公司 一种基于深度学习的社区医生调度方法及系统
CN112071425A (zh) * 2020-09-04 2020-12-11 平安科技（深圳）有限公司 一种数据处理方法、装置、计算机设备及存储介质
CN112530576A (zh) * 2020-11-30 2021-03-19 百度健康(北京)科技有限公司 一种线上医患匹配方法、装置、电子设备及存储介质
CN113782163A (zh) * 2021-03-29 2021-12-10 北京京东拓先科技有限公司 信息推送方法、装置和计算机可读存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101323699B1 (ko) * 2012-11-14 2013-10-30 나인성 온라인 상에서의 환자와 치료자간의 매칭시스템
CN107563120B (zh) * 2017-09-13 2021-06-01 青岛海信医疗设备股份有限公司 针对患者的医生推荐方法及装置
CN110232971B (zh) * 2019-05-24 2022-04-12 深圳市翩翩科技有限公司 一种医生推荐方法及装置
CN110993081B (zh) * 2019-12-03 2023-08-11 济南大学 一种医生在线推荐方法及系统
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
CN112071420A (zh) * 2020-08-12 2020-12-11 福建中榕数据科技有限公司 基于实时数据的临床辅助决策方法、系统、设备和介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160147943A1 (en) * 2014-11-21 2016-05-26 Argo Data Resource Corporation Semantic Address Parsing Using a Graphical Discriminative Probabilistic Model
CN110752017A (zh) * 2019-09-04 2020-02-04 重庆特斯联智慧科技股份有限公司 一种基于深度学习的社区医生调度方法及系统
CN112071425A (zh) * 2020-09-04 2020-12-11 平安科技（深圳）有限公司 一种数据处理方法、装置、计算机设备及存储介质
CN112530576A (zh) * 2020-11-30 2021-03-19 百度健康(北京)科技有限公司 一种线上医患匹配方法、装置、电子设备及存储介质
CN113782163A (zh) * 2021-03-29 2021-12-10 北京京东拓先科技有限公司 信息推送方法、装置和计算机可读存储介质

Also Published As

Publication number Publication date
CN113782163A (zh) 2021-12-10

Similar Documents

Publication Publication Date Title
CN111191020B (zh) 基于机器学习和知识图谱的处方推荐方法和系统
WO2022206500A1 (zh) 信息推送方法, 装置和计算机可读存储介质
US11256876B2 (en) Gap in care determination using a generic repository for healthcare
AU2012245343B2 (en) Predictive modeling
US20180089385A1 (en) Personalized treatment management system
US20170061091A1 (en) Indication of Outreach Options for Healthcare Facility to Facilitate Patient Actions
US20080103369A1 (en) System and method for a lifestyle management system
US20150254754A1 (en) Methods and apparatuses for consumer evaluation of insurance options
US20210350910A1 (en) System and method for supporting healthcare cost and quality management
US20230052573A1 (en) System and method for autonomously generating personalized care plans
JP2014533860A (ja) 患者の縦断的なエピソードを管理するためのグラフィックツール
US8666761B2 (en) System and method for an integrated disease management system
EP2973104A2 (en) Method and apparatus for transmitting healthcare messages to an automatically identified set of patients
US20190318829A1 (en) Adaptive medical documentation system
US20220384001A1 (en) System and method for a clinic viewer generated using artificial-intelligence
EP4131279A1 (en) Experience engine-method and apparatus of learning from similar patients
US20180121607A1 (en) Electronic health record quality enhancement
US11417417B2 (en) Generating clinical forms
Lakshmi et al. Prediction of diabetes patient stage using ontology based machine learning system
Miloski Opportunities for artificial intelligence in healthcare and in vitro fertilization
CN110473636B (zh) 基于深度学习的智能医嘱推荐方法及系统
CN116860935A (zh) 基于提示词问答交互的内容管理方法、装置、设备及介质
US20230047253A1 (en) System and Method for Dynamic Goal Management in Care Plans
US20170177813A1 (en) System and method for recommending analytic modules based on leading factors contributing to a category of concern
CN114661985A (zh) 患者健康管理平台的信息推送方法、平台、设备及介质

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: 22778683

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13.02.2024)