[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

WO2022068160A1 - 基于人工智能的重症问诊数据识别方法、装置、设备及介质 - Google Patents

基于人工智能的重症问诊数据识别方法、装置、设备及介质 Download PDF

Info

Publication number
WO2022068160A1
WO2022068160A1 PCT/CN2021/084349 CN2021084349W WO2022068160A1 WO 2022068160 A1 WO2022068160 A1 WO 2022068160A1 CN 2021084349 W CN2021084349 W CN 2021084349W WO 2022068160 A1 WO2022068160 A1 WO 2022068160A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
consultation
feature data
historical
entity
Prior art date
Application number
PCT/CN2021/084349
Other languages
English (en)
French (fr)
Inventor
满晏松
柳恭
李响
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 WO2022068160A1 publication Critical patent/WO2022068160A1/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
    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • 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

  • the present application relates to the technical field of artificial intelligence, and in particular, to a method, device, equipment and medium for identifying data of critical care consultation based on artificial intelligence.
  • An artificial intelligence-based method for identifying data of critical care inquiries comprising:
  • the prediction model is based on the multi-dimensional feature data and the training labels corresponding to the multi-dimensional feature data Obtained through training, the multi-dimensional feature data is extracted and obtained according to historical consultation data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data;
  • An artificial intelligence-based critical care data identification device comprising:
  • an acquisition module used to acquire the consultation session data corresponding to the target user ID
  • the first identification module is used to input the data of the consultation session into a prediction model, and output a model identification result corresponding to the data of the consultation session through the prediction model; wherein, the prediction model is based on the multi-dimensional feature data and the The training labels corresponding to the multi-dimensional feature data are obtained by training, the multi-dimensional feature data is extracted and obtained according to the historical consultation data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data;
  • a second identification module configured to determine an expert identification result corresponding to the consultation session data according to the tags hit by the consultation session data in the expert knowledge base;
  • a decision-making module configured to combine the model identification result and the expert identification result to obtain a target identification result of whether the consultation session data is critical consultation data.
  • a computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
  • the prediction model is based on the multi-dimensional feature data and the training labels corresponding to the multi-dimensional feature data Obtained through training, the multi-dimensional feature data is extracted and obtained according to historical consultation data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data;
  • the prediction model is based on the multi-dimensional feature data and the training labels corresponding to the multi-dimensional feature data Obtained through training, the multi-dimensional feature data is extracted and obtained according to historical consultation data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data;
  • FIG. 1 is an application scenario diagram of an artificial intelligence-based critical care data identification method in one embodiment
  • Fig. 2 is a schematic flowchart of an artificial intelligence-based critical care data identification method in one embodiment
  • Fig. 3 is a flow diagram of using a predictive model in one embodiment
  • Fig. 4 is a flow chart of training a prediction model in one embodiment
  • Fig. 5 is a structural block diagram of an artificial intelligence-based critical care data identification device in one embodiment
  • FIG. 6 is a structural block diagram of an artificial intelligence-based critical interrogation data identification device in another embodiment
  • Fig. 7 is the internal structure diagram of the computer device in one embodiment
  • FIG. 8 is an internal structure diagram of a computer device in another embodiment.
  • the artificial intelligence-based critical care data identification method provided in this application can be applied to the application environment shown in FIG. 1 .
  • the terminal 102 communicates with the server 104 through the network through the network. Specifically, the terminal 102 can obtain the consultation session data corresponding to the target user identifier, send the consultation session data to the server 104, and the server 104 inputs the consultation session data into the prediction model, and outputs the data corresponding to the consultation session through the prediction model.
  • the model identification results of the 2000 wherein, the prediction model is trained according to the multi-dimensional feature data and the corresponding training labels of the multi-dimensional feature data, the multi-dimensional feature data is extracted according to the historical consultation data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data; server; 104 then determine the expert identification result corresponding to the consultation session data according to the tag hit in the expert knowledge base by the consultation session data; the server 104 then combines the model identification result and the expert identification result to obtain whether the consultation session data is a critical consultation
  • the target recognition result of the data can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices, and the server 104 can be implemented by an independent server or a server cluster composed of multiple servers.
  • the terminal 102 or the server 104 may also be independently used to execute the artificial intelligence-based method for identifying data of critical care inquiries. This application is not limited here.
  • an artificial intelligence-based method for recognizing data for critical care consultation is provided, and the method is applied to a computer device as an example for description, and the computer device may specifically be the terminal in FIG. 1 . or server.
  • the artificial intelligence-based method for identifying data of critically ill inquiries specifically includes the following steps:
  • Step 202 Obtain the consultation session data corresponding to the target user identifier.
  • a consultation session is a process of consultation interaction between at least two users.
  • the at least two users include a user corresponding to a patient role and a user corresponding to a doctor role.
  • the consultation session data is the data generated during the consultation interaction.
  • the user corresponding to the doctor role can be the doctor himself or an artificial intelligence robot.
  • an online consultation application or an online consultation website may run on the terminal, and the online consultation application or the online consultation website may provide a consultation entrance.
  • the user inputs the consultation session data based on the consultation portal through the terminal to conduct an online consultation.
  • the consultation session data may be voice data, text data, or image data, or the like.
  • the consultation session data may include consultation session data corresponding to the patient role and consultation session data corresponding to the doctor role.
  • the consultation session data corresponding to the patient's role such as user basic information, symptom description information, symptom photos, medical examination reports or past history information, etc. It can be understood that the inquiring user may or may not be the patient himself, such as in the scene of inquiring for children or the elderly.
  • Consultation session data corresponding to the doctor's role such as disease description information, symptom analysis information, etiology analysis information, or reply information for user inquiry data, etc.
  • the consultation session data may be dialogue data of one or more rounds of question and answer in a consultation process.
  • the consultation session data provided by the user for the first time can identify whether the consultation session data is critical consultation data, it can be identified as soon as possible for corresponding processing; when the amount of information is insufficient, the user can be guided to provide more information to combine these The information is more accurate in the identification of critical care data.
  • the critically ill consultation data is the consultation session data involving acute and critical illnesses.
  • "Emergency and critical illness” is a medical term, which usually means that the patient's disease is an urgent and endangered disease, and medical treatment should be carried out as soon as possible, otherwise it may cause serious harm to the patient's body or cause death.
  • the consultation session data includes the clinical symptoms of acute and critical illness.
  • the clinical symptoms of acute and critical illness such as "fainting", "difficulty breathing", etc.
  • Step 204 input the data of the consultation session into the prediction model, and output the model identification result corresponding to the data of the consultation session through the prediction model; wherein, the prediction model is obtained by training according to the multi-dimensional feature data and the corresponding training labels of the multi-dimensional feature data, and the multi-dimensional feature data is obtained according to The historical consultation data is extracted, and the multi-dimensional feature data includes entity feature data and entity relationship feature data.
  • the prediction model is a pre-trained machine learning model used to identify whether the consultation session data is critical consultation data.
  • the machine learning model can be a neural network model, a support vector machine, or a logistic regression model.
  • Neural network models such as Convolutional Neural Networks, Backpropagation Neural Networks, Feedback Neural Networks, Radial Basis Neural Networks, or Self-Organizing Neural Networks.
  • Entity characteristics are data reflecting the characteristics of the entity itself.
  • the consultation session data “drinking during dinner, abdominal pain is unbearable now” includes two entities, the first entity is “drinking”, and the second entity is “abdominal pain”.
  • An entity-relationship feature refers to data that reflects the relationship between at least two entities. For example, the entity relationship between "drinking alcohol” and “abdominal pain” in the consultation session data "drinking during dinner, abdominal pain is unbearable” is “causal relationship", that is, the cause of abdominal pain is drinking.
  • the computer equipment can integrate various aspects of information when designing the input data of the prediction model, for example, the two feature dimensions of the data, the entity dimension and the entity relationship dimension, can be used as the input data of the prediction model.
  • the prediction model can learn the effective information of these two feature dimensions during training, and improve the model's ability to recognize critical care data.
  • the computer equipment can input the consultation session data into the prediction model, process the consultation session data through a plurality of neurons included in the prediction model, obtain a model identification result corresponding to the consultation session data, and then output the model identification through the prediction model. result.
  • the neuron is the most basic structure in the neural network. Under normal circumstances, most neurons are in an inhibitory state, but when the neuron receives input information, causing its potential to exceed a threshold, then the neuron will It is activated and is in an "excited" state, and then the output information is propagated to other neurons.
  • a connection line connecting neurons corresponds to a weight (the value of which is called a weight), and usually different connection lines correspond to different weights.
  • the threshold value of each neuron and the weight of the connection relationship between each neuron are determined when the prediction model is trained by the multi-dimensional feature data and the training labels corresponding to the multi-dimensional feature data.
  • Neurons include input neurons, output neurons and hidden neurons.
  • the artificial intelligence-based method for identifying data of critical care consultation further includes a training step of the prediction model, and the training step specifically includes: collecting historical consultation data and training labels corresponding to the historical consultation data; historical consultation data Corresponding training labels are used to indicate whether the historical consultation data is critical consultation data; entity feature data and entity relationship feature data are extracted from the historical consultation data to generate multi-dimensional feature data corresponding to the historical consultation data; The multi-dimensional feature data corresponding to the data and the historical consultation data are jointly input to the prediction model to be trained to obtain the prediction recognition result; the prediction model is trained based on the prediction recognition result of the prediction model and the training label.
  • the multidimensional feature data also includes intent feature data.
  • the training steps of the prediction model specifically include: collecting historical consultation data and training labels corresponding to the historical consultation data; training labels corresponding to the historical consultation data are used to indicate whether the historical consultation data is critical consultation data; Extract entity feature data, entity relationship feature data and intent feature data from the diagnostic data to generate multi-dimensional feature data corresponding to the historical consultation data; input the historical consultation data and the multi-dimensional feature data corresponding to the historical consultation data together to be trained
  • the prediction model is obtained, and the prediction recognition result is obtained; the prediction model is trained based on the prediction recognition result of the prediction model and the training label.
  • the intent feature is the data reflecting the expressed intent.
  • the intent feature includes feature data reflecting the doctor's inquiring or answering intent and the user's inquiring or responding intent. For example, the consultation session data "Do you have unbearable abdominal pain right now” expresses the symptom confirmation intention; the consultation session data "Yes, I have unbearable abdominal pain now” expresses the symptom confirmation intention, and so on.
  • the computer device may convert the consultation session data into a data format that can be processed by the prediction model, and then input the converted data into the prediction model.
  • the data format that the prediction model can handle such as vector format or matrix format, etc.
  • Step 206 Determine the expert identification result corresponding to the consultation session data according to the tags hit by the consultation session data in the expert knowledge base.
  • the knowledge base refers to the set of rules applied in the design of the expert system, including the facts and data related to the rules, and the whole of them constitutes the knowledge base.
  • the knowledge base is related to a specific expert system, and in this application, the expert knowledge base is related to the medical consultation expert system.
  • the rules in the expert knowledge base are obtained by extracting the frequently occurring data in the critical consultation data in the historical consultation process.
  • the facts linked by the rules include whether the consultation session data in the historical consultation process is the critical consultation data.
  • the data contacted by the rule includes labels such as frequently occurring disease symptoms in the consultation session data during the historical consultation process.
  • the computer device may be driven by a rule engine, and based on the rule set of the expert knowledge base, determine the tags hit by the consultation session data in the expert knowledge base.
  • the hit label is the label corresponding to the critical care consultation data
  • the expert identification result corresponding to the consultation session data is obtained as the consultation session data is the critical consultation data
  • the hit label is not the label corresponding to the critical consultation data
  • the expert identification result corresponding to the consultation session data is obtained as that the consultation session data is not critical consultation data.
  • Step 208 combining the model identification result and the expert identification result, obtain the target identification result of whether the consultation session data is critical consultation data.
  • the model recognition result includes two types of recognition results, that the consultation session data is critical-care consultation data and the consultation session data is not critical-care consultation data.
  • the expert identification result also includes two kinds of identification results, which are that the consultation session data is critically ill consultation data, and that the consultation session data is not critically ill consultation data. If both the model identification result and the expert identification result are both the consultation session data and the critical care consultation data, it can be obtained that the consultation session data is the target identification result of the critical care consultation data. If at least one of the model identification result and the expert identification result is that the consultation session data is not the critically ill consultation data, then the target identification result that the consultation session data is not the critically ill consultation data can be obtained.
  • the expert identification result may also include a situation in which it is not identified whether the consultation session data is critical consultation data. At this time, the model recognition result can be used as the target recognition result.
  • the above artificial intelligence-based critical consultation data identification method after obtaining the user's consultation session data, on the one hand, the consultation session data is input into the prediction model to obtain the model identification result, and on the other hand, based on the expert knowledge base, the query session data is obtained. Identify the data of the consultation session to obtain the expert identification result, and then combine the model identification result and the expert identification result to obtain the final identification result.
  • the prediction model is trained based on multi-dimensional feature data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data, the prediction model can learn information of different dimensions during the training process, and combine the contextual semantic environment to better Understand the logic of language, thereby improving the predictive model's ability to recognize critical care data; in this way, the combination of model prediction and expert system for critical care data identification can make up for the lack of relying only on expert systems and improve critical care data. recognition accuracy.
  • the computer device may perform an operation corresponding to the target identification result according to the target identification result.
  • the artificial intelligence-based method for identifying critical-care consultation data further includes: when obtaining a target identification result that the consultation session data is critical-care consultation data, identifying the consultation session to which the consultation session data belongs Access to the doctor's terminal; when it is obtained that the target identification result of the consultation session data is not the critically ill consultation data, continue to advance the consultation session to which the consultation session data belongs.
  • FIG. 3 shows a flowchart of using a predictive model in one embodiment.
  • the consultation session data can be input into the prediction model and the expert system in parallel.
  • the expert system is used to identify whether the consultation session data is critical consultation data, and the expert identification result is obtained.
  • the decision maker combines the model identification results and the expert identification results to obtain the target identification results of whether the consultation session data is critical consultation data.
  • the consultation session to which the consultation session data belongs is connected to the doctor's terminal, and the doctor manually intervenes through the doctor's terminal to review the final identification result of the critically-ill consultation data.
  • the identification when the identification is correct, it can be further processed in a timely manner, such as giving medical advice.
  • the consultation session data is not the target recognition result of the critically ill consultation data
  • the consultation session to which the consultation session data belongs is continued, for example, the artificial intelligence robot continues to interact with the user for consultation.
  • the corresponding next step is immediately performed, so that the user who is in urgent need of help can be effectively answered when the consultation session data is critical Session data is not critical consultation data, and online consultations can be continued in an orderly manner.
  • the consultation session data is stored on the blockchain. It should be emphasized that, in order to further ensure the privacy and security of the above-mentioned consultation session data, the above-mentioned consultation session data may also be stored in a node of a blockchain.
  • the artificial intelligence-based method for identifying data of critical care consultation further includes: collecting historical consultation data and training labels corresponding to the historical consultation data; and the training labels corresponding to the historical consultation data are used to represent the historical consultation data Whether it is critical consultation data; extract entity feature data and entity relationship feature data from historical consultation data to generate multi-dimensional feature data corresponding to historical consultation data; The data is jointly input to the prediction model to be trained, and the prediction recognition result is obtained; the prediction model is trained based on the prediction recognition result of the prediction model and the training label.
  • the multi-dimensional feature data further includes intent feature data
  • the artificial intelligence-based method for identifying critical care data further includes: collecting historical consultation data and training labels corresponding to the historical consultation data;
  • the training label is used to indicate whether the historical consultation data is critical consultation data;
  • entity feature data, entity relationship feature data and intent feature data are extracted from the historical consultation data to generate multi-dimensional feature data corresponding to the historical consultation data;
  • the multi-dimensional feature data corresponding to the consultation data and the historical consultation data are jointly input to the prediction model to be trained to obtain the prediction recognition result; the prediction model is trained based on the prediction recognition result of the prediction model and the training label.
  • the computer device may first collect historical consultation data and training labels corresponding to the historical consultation data.
  • the training label may be the result of manual labeling, indicating whether the historical consultation data is critical consultation data.
  • the computer device can then extract entity feature data and entity relationship feature data from the historical consultation data, and generate multidimensional feature data corresponding to the historical consultation data, the multidimensional feature data includes at least two feature dimensions, and the multidimensional feature data and historical The interview data are collectively used as input data for the predictive model to be trained.
  • the computer equipment can also extract entity feature data, entity relationship feature data and intention feature data from historical consultation data, and generate multidimensional feature data corresponding to the historical consultation data, where the multidimensional feature data includes at least three feature dimensions, and the multidimensional feature data Feature data and historical interview data are used together as input data for the predictive model to be trained.
  • generating multi-dimensional feature data based on multiple feature data may be splicing or fusing the multiple feature data.
  • the computer equipment can obtain the prediction recognition result output by the prediction model to be trained, and then construct a training loss function according to the difference between the prediction recognition result and the training label, and use the back-propagation algorithm to optimize in the direction of minimizing the training loss function.
  • the weights and thresholds of each hidden neuron and input/output neurons can be obtained after training, and then the model parameter file of the trained prediction model can be obtained.
  • the computer device may store the model parameter file in the expert knowledge base.
  • the computer device when designing the input data of the prediction model, fuses various information, for example, the two feature dimensions of the data, the entity dimension and the entity relationship dimension, are fused as the input data of the prediction model. This enables the prediction model to learn at least the effective information of these two feature dimensions during training, and improves the model's ability to recognize critical care data.
  • the intentions of doctors and users to ask and answer can also be used as the basis for identifying critical care data. Therefore, when designing the input data of the prediction model, intention features can be introduced, so that the entity dimension, entity relationship dimension and intention The fusion of these three feature dimensions can enable the prediction model to learn at least the effective information of these three feature dimensions during training, improve the model's ability to recognize critical care data, and expand the direction of critical care data identification. .
  • extracting entity feature data, entity relationship feature data, and intent feature data from historical consultation data, and generating multidimensional feature data corresponding to the historical consultation data includes: using a rule engine to drive the extraction of historical consultation data corresponding to Named entity recognition of historical consultation data using sequence annotation model, and extraction of entity relationship feature data between named entities based on the identified named entities; semantic model to extract intent feature data in historical consultation data ; According to entity feature data, entity relationship feature data and intention feature data, multi-dimensional feature data corresponding to historical consultation data is generated.
  • a rule engine is used to drive the extraction of entity feature data corresponding to historical consultation data, including: performing word segmentation on historical consultation data to obtain word segmentation results; based on expert knowledge base
  • the provided entity label is driven by a rule engine to extract the entity label according to the word segmentation result, and obtain the entity feature data corresponding to the historical consultation data.
  • the computer equipment can perform word segmentation on the historical consultation data to obtain the word segmentation result, and use the rule engine to extract the entity label from the word segmentation result according to the rules in the rule set of the expert knowledge base and the entity label provided from the expert knowledge base, and obtain Entity feature data.
  • entity label entity feature data
  • patient infant
  • symptom vomiting
  • weight loss etc.
  • the rules engine is a component embedded in the application that accepts data input, interprets the rules, and makes decisions based on the rules.
  • the computer equipment can group historical consultation data according to consultation sessions, and generate consultations from historical consultation data of one consultation session according to the response sequence. Then, the sequence labeling model is used to identify the named entities of the medical data sequence, and based on the identified named entities, the entity relationship feature data between the named entities is extracted. For example, named entities (drinking, inducement, abdominal pain), extract the entity relationship: the inducement of abdominal pain is drinking.
  • named entities drinking, inducement, abdominal pain
  • extract the entity relationship the inducement of abdominal pain is drinking.
  • a semantic model to extract the intention feature data in the historical consultation data including: based on the intention label provided by the expert knowledge base, using the semantic model to extract the historical query data.
  • the intent label corresponding to the medical consultation data is obtained, and the intent characteristic data corresponding to the historical medical consultation data is obtained.
  • the computer device can also use the semantic model in the third aspect to extract the intent features of the historical consultation data.
  • the computer equipment can obtain the historical consultation data as samples, manually annotate the intent category labels (intent labels provided by the expert knowledge base), and train the semantic model with supervision. After that, the trained semantic model can be used to extract the historical consultation. Intent characteristics of the data.
  • the artificial intelligence-based method for identifying critical illness data further includes: screening out historical critical consultation data from historical critical consultation data; counting the occurrence frequency of entity keywords in the historical critical consultation data; Entity keywords whose frequency is higher than the preset threshold are used as entity labels to join the expert knowledge base; the consultation intention corresponding to the historical critical consultation data is determined, and the intention label is obtained; the intention label is added to the expert knowledge base.
  • the computer device can also establish an expert knowledge base.
  • the computer equipment can first obtain the critically ill patient data identified in the historical consultation data, and obtain the critically ill sample data set; and then count the occurrence frequency of entity keywords such as incentives, objects, and symptoms included in the critically ill sample data set in the critically ill sample data set, Entity keywords whose occurrence frequency is higher than a preset threshold are filtered and used as entity labels to join the expert knowledge base.
  • the computer equipment can also determine the consultation intention corresponding to the historical critical care consultation data, obtain the intention label, and add the intention label to the expert knowledge base.
  • the labels accurately summarized by the expert system can be used to extract the multi-dimensional features of the training samples, which avoids unnecessary training time for long-tail words and greatly improves the The training time for machine learning model training.
  • FIG. 4 shows a flowchart of training a prediction model in one embodiment.
  • the computer equipment can first perform data preparation, that is, collect historical consultation data and the corresponding training labels of the historical consultation data, and then, on the one hand, based on the expert system, based on the various data provided by the expert knowledge base from Entity feature data is extracted from historical consultation data.
  • entity relationship feature data is extracted from historical consultation data based on semantic understanding
  • intent feature data is extracted from historical consultation data by combining various data and semantic understanding provided by expert knowledge bases.
  • the computer equipment then fuses the feature data extracted from the three aspects, and then combines the historical consultation data as input data to construct a neural network structure and conduct training to obtain a prediction model for identifying critical consultation data.
  • an artificial intelligence-based critical care data identification device including: an acquisition module 501 , a first identification module 502 , a second identification module 503 and a decision module 504 , wherein :
  • Obtaining module 501 is used to obtain the consultation session data corresponding to the target user identifier
  • the first identification module 502 is configured to input the data of the consultation session into a prediction model, and output a model identification result corresponding to the data of the consultation session through the prediction model;
  • the corresponding training labels of the multi-dimensional feature data are obtained by training, the multi-dimensional feature data is extracted and obtained according to historical consultation data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data;
  • the second identification module 503 is configured to determine the expert identification result corresponding to the consultation session data according to the tags hit by the consultation session data in the expert knowledge base;
  • the decision module 504 is configured to combine the model identification result and the expert identification result to obtain a target identification result of whether the consultation session data is critical consultation data.
  • the first identification module 502 is further configured to input the consultation session data into the prediction model, process the consultation session data through a plurality of neurons included in the prediction model, and obtain a model identification result corresponding to the consultation session data ; Output the model recognition result through the prediction model; wherein, the threshold of each neuron and the weight of the connection relationship between each neuron are determined when the prediction model is trained through the multi-dimensional feature data and the corresponding training labels of the multi-dimensional feature data, and the multi-dimensional feature data also Include intent feature data.
  • the multi-dimensional feature data further includes intention feature data
  • the artificial intelligence-based critical consultation data identification device further includes: a training module 505 for collecting historical consultation data and corresponding historical consultation data Training label; the training label corresponding to the historical consultation data is used to indicate whether the historical consultation data is critical consultation data; entity feature data, entity relationship feature data and intent feature data are extracted from the historical consultation data to generate historical consultation data.
  • Corresponding multi-dimensional feature data input the historical consultation data and the multi-dimensional feature data corresponding to the historical consultation data into the prediction model to be trained to obtain the prediction and recognition results; based on the prediction and recognition results of the prediction model and the training labels to train the prediction model .
  • the training module 505 is further configured to use a rule engine to drive the extraction of entity feature data corresponding to the historical consultation data; use a sequence labeling model to perform named entity recognition on the historical consultation data, and extract names based on the identified named entities
  • the entity relationship feature data between entities; the semantic model is used to extract the intent feature data in the historical consultation data; according to the entity feature data, entity relationship feature data and intent feature data, multi-dimensional feature data corresponding to the historical consultation data is generated.
  • the training module 505 is also used to perform word segmentation on the historical consultation data to obtain word segmentation results; based on the entity labels provided by the expert knowledge base, the rule engine is used to drive the extraction of entity labels according to the word segmentation results, and the results of the historical consultation data are obtained. Corresponding entity feature data; based on the intent labels provided by the expert knowledge base, the semantic model is used to extract the intent labels corresponding to the historical consultation data, and the intent feature data corresponding to the historical consultation data is obtained.
  • the training module 505 is further configured to filter out historical critical care consultation data from historical consultation data; count the occurrence frequency of entity keywords in the historical critical consultation data; filter entities whose occurrence frequency is higher than a preset threshold Keywords are used as entity labels to be added to the expert knowledge base; the consultation intentions corresponding to the historical critical consultation data are determined, and the intention labels are obtained; the intention labels are added to the expert knowledge base.
  • the consultation session data is stored in the blockchain; the decision-making module 504 is further configured to, when obtaining the target identification result that the consultation session data is the critically-ill consultation data, store the consultation session data to which the consultation session data belongs.
  • the session is connected to the doctor's terminal; when it is obtained that the target identification result of the consultation session data is not the critical care consultation data, the consultation session to which the consultation session data belongs is continued.
  • the above-mentioned artificial intelligence-based critical consultation data identification device after acquiring the user's consultation session data, on the one hand, input the consultation session data into the prediction model to obtain the model identification result, and on the other hand, based on the expert knowledge base to answer the question. Identify the data of the consultation session to obtain the expert identification result, and then combine the model identification result and the expert identification result to obtain the final identification result.
  • the prediction model is trained based on multi-dimensional feature data, and the multi-dimensional feature data includes entity feature data and entity relationship feature data
  • the prediction model can learn information of different dimensions during the training process, and combine the contextual semantic environment to better Understand the logic of the language, thereby improving the predictive model's ability to recognize the critical care data based on artificial intelligence; in this way, the combination of model prediction and expert system to identify critical data can make up for the deficiency of relying only on the expert system, and improve the performance based on Accuracy of AI-based critical care data identification.
  • Each module in the above-mentioned artificial intelligence-based critical care data identification device may be implemented in whole or in part by software, hardware and combinations thereof.
  • the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
  • a computer device is provided, and the computer device can be a server, and its internal structure diagram can be as shown in FIG. 7 .
  • the computer device includes a processor, memory, a network interface, and a database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the nonvolatile storage medium stores an operating system, a computer program, and a database.
  • the internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
  • the database of the computer equipment is used to store the identification data of the critical care consultation data based on artificial intelligence.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an artificial intelligence-based method for recognizing data of critical care inquiries can be realized.
  • a computer device is provided, and the computer device may be a terminal, and its internal structure diagram may be as shown in FIG. 8 .
  • the computer equipment includes a processor, memory, a network interface, a display screen, and an input device connected by a system bus.
  • the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the nonvolatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the display screen of the computer equipment may be a liquid crystal display screen or an electronic ink display screen
  • the input device of the computer equipment may be a touch layer covered on the display screen, or a button, a trackball or a touchpad set on the shell of the computer equipment , or an external keyboard, trackpad, or mouse.
  • FIG. 7 or 8 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • a device may include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.
  • a computer device including a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program: acquiring consultation session data corresponding to a target user identifier; The consultation session data is input into the prediction model, and the model identification result corresponding to the consultation session data is output through the prediction model; wherein, the prediction model is trained according to the multi-dimensional feature data and the training labels corresponding to the multi-dimensional feature data, and the multi-dimensional feature data is based on the historical consultation data.
  • the extracted multi-dimensional feature data includes entity feature data and entity relationship feature data; according to the tags hit by the consultation session data in the expert knowledge base, the expert identification result corresponding to the consultation session data is determined; combined with the model identification result and the expert identification As a result, a target identification result of whether the consultation session data is critical consultation data is obtained.
  • inputting the consultation session data into a prediction model, and outputting a model recognition result corresponding to the consultation session data through the prediction model includes: inputting the consultation session data into the prediction model, and using a plurality of neurons included in the prediction model Process the data of the consultation session to obtain the model recognition result corresponding to the data of the consultation session; output the model recognition result through the prediction model; among them, the threshold of each neuron and the weight of the connection relationship between each neuron are obtained through the multi-dimensional feature data. and the training labels corresponding to the multi-dimensional feature data are determined when training the prediction model, and the multi-dimensional feature data also includes intent feature data.
  • the multidimensional feature data also includes intent feature data.
  • the processor executes the computer program, the following steps are also implemented: collecting historical consultation data and training labels corresponding to the historical consultation data; the training labels corresponding to the historical consultation data are used to indicate whether the historical consultation data is critical consultation data; Entity feature data, entity relationship feature data, and intent feature data are extracted from the consultation data to generate multi-dimensional feature data corresponding to historical consultation data; The trained prediction model is used to obtain the prediction recognition result; the prediction model is trained based on the prediction recognition result of the prediction model and the training label.
  • extracting entity feature data, entity relationship feature data, and intent feature data from historical consultation data, and generating multidimensional feature data corresponding to the historical consultation data includes: using a rule engine to drive the extraction of historical consultation data corresponding to Named entity recognition of historical consultation data using sequence annotation model, and extraction of entity relationship feature data between named entities based on the identified named entities; semantic model to extract intent feature data in historical consultation data ; According to entity feature data, entity relationship feature data and intention feature data, multi-dimensional feature data corresponding to historical consultation data is generated.
  • using a rule engine to drive the extraction of entity feature data corresponding to the historical consultation data includes: performing word segmentation on the historical consultation data to obtain a word segmentation result; As a result, entity labels are extracted to obtain entity feature data corresponding to historical consultation data.
  • Using the semantic model to extract the intent feature data in the historical consultation data including: based on the intent labels provided by the expert knowledge base, using the semantic model to extract the intent labels corresponding to the historical consultation data, and obtaining the intent features corresponding to the historical consultation data data.
  • the processor when the processor executes the computer program, the processor further implements the following steps: screening out historical critical illness consultation data from historical consultation data; counting the occurrence frequency of entity keywords in the historical critical illness consultation data; screening the occurrence frequency higher than Entity keywords with preset thresholds are used as entity labels to be added to the expert knowledge base; the consultation intention corresponding to the historical critical consultation data is determined to obtain the intention label; the intention label is added to the expert knowledge base.
  • the consultation session data is stored in the blockchain; when the processor executes the computer program, the processor further implements the following steps: when obtaining the target identification result that the consultation session data is critical care consultation data, then the consultation session The consultation session to which the data belongs is connected to the doctor's terminal; when it is obtained that the consultation session data is not the target identification result of the critical care consultation data, the consultation session to which the consultation session data belongs continues to be promoted.
  • a computer storage medium may be volatile or non-volatile, and a computer program is stored thereon, and when the computer program is executed by a processor, the following Steps: obtaining the consultation session data corresponding to the target user identification; inputting the consultation session data into the prediction model, and outputting the model identification result corresponding to the consultation session data through the prediction model; wherein, the prediction model is based on the multi-dimensional feature data and the multi-dimensional feature data.
  • Corresponding training labels are trained, multi-dimensional feature data is extracted from historical consultation data, and multi-dimensional feature data includes entity feature data and entity relationship feature data;
  • inputting the consultation session data into a prediction model, and outputting a model recognition result corresponding to the consultation session data through the prediction model includes: inputting the consultation session data into the prediction model, and using a plurality of neurons included in the prediction model Process the data of the consultation session to obtain the model recognition result corresponding to the data of the consultation session; output the model recognition result through the prediction model; among them, the threshold of each neuron and the weight of the connection relationship between each neuron are obtained through the multi-dimensional feature data. and the training labels corresponding to the multi-dimensional feature data are determined when training the prediction model, and the multi-dimensional feature data also includes intent feature data.
  • the multi-dimensional feature data further includes intention feature data; when the computer program is executed by the processor, the following steps are further implemented: collecting historical consultation data and training labels corresponding to the historical consultation data; training labels corresponding to the historical consultation data It is used to indicate whether the historical consultation data is critical consultation data; extract entity feature data, entity relationship feature data and intention feature data from the historical consultation data to generate multi-dimensional feature data corresponding to the historical consultation data; The multi-dimensional feature data corresponding to the data and the historical consultation data are jointly input to the prediction model to be trained to obtain the prediction recognition result; the prediction model is trained based on the prediction recognition result of the prediction model and the training label.
  • extracting entity feature data, entity relationship feature data, and intent feature data from historical consultation data, and generating multidimensional feature data corresponding to the historical consultation data includes: using a rule engine to drive the extraction of historical consultation data corresponding to Named entity recognition of historical consultation data using sequence annotation model, and extraction of entity relationship feature data between named entities based on the identified named entities; semantic model to extract intent feature data in historical consultation data ; According to entity feature data, entity relationship feature data and intention feature data, multi-dimensional feature data corresponding to historical consultation data is generated.
  • using a rule engine to drive the extraction of entity feature data corresponding to the historical consultation data includes: performing word segmentation on the historical consultation data to obtain a word segmentation result; As a result, entity labels are extracted to obtain entity feature data corresponding to historical consultation data.
  • Using the semantic model to extract the intent feature data in the historical consultation data including: based on the intent labels provided by the expert knowledge base, using the semantic model to extract the intent labels corresponding to the historical consultation data, and obtaining the intent features corresponding to the historical consultation data data.
  • the following steps are further implemented: screening out historical critical illness consultation data from historical consultation data; counting the occurrence frequency of entity keywords in the historical critical illness consultation data; screening for high occurrence frequency
  • entity keywords at the preset threshold are used as entity labels to join the expert knowledge base; the consultation intention corresponding to the historical critical consultation data is determined, and the intention label is obtained; the intention label is added to the expert knowledge base.
  • the consultation session data is stored in the blockchain; when the computer program is executed by the processor, the following steps are further implemented: when the target identification result that the consultation session data is the critically ill consultation data is obtained, then the consultation session data is obtained.
  • the consultation session to which the session data belongs is connected to the doctor's terminal; when it is obtained that the consultation session data is not the target identification result of the critical care consultation data, the consultation session to which the consultation session data belongs continues to be promoted.
  • the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • Nonvolatile 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 various forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Road (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM double data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain Road (Synchlink) DRAM
  • SLDRAM synchronous chain Road (Synchlink) DRAM
  • Rambus direct RAM
  • DRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pathology (AREA)
  • Molecular Biology (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Neurology (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种基于人工智能的重症问诊数据识别方法、装置、设备及介质，涉及人工智能。方法包括：获取与目标用户标识对应的问诊会话数据（S202）；将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果；其中，预测模型根据多维特征数据以及多维特征数据相应的训练标签训练得到，多维特征数据根据历史问诊数据提取得到、且多维特征数据包括实体特征数据和实体关系特征数据（S204）；根据问诊会话数据在专家知识库中命中的标签，确定与问诊会话数据对应的专家识别结果（S206）；结合模型识别结果和专家识别结果，得到问诊会话数据是否为重症问诊数据的目标识别结果（S208）。采用本方法能够提高重症问诊数据的识别准确率。此外，还涉及区块链技术，用户的问诊会话数据可存储于区块链中。

Description

基于人工智能的重症问诊数据识别方法、装置、设备及介质
本申请要求于2020年9月30日提交中国专利局、申请号为CN202011065413.X、名称为“基于人工智能的重症问诊数据识别方法及装置”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，特别是涉及一种基于人工智能的重症问诊数据识别方法、装置、设备及介质。
背景技术
随着互联网技术和医疗技术的发展，互联网技术在医疗行业中的应用也越来越普遍。比如，用户可以通过在线问诊应用或者在线问诊网站自述症状、咨询病症、了解药物以及寻求就诊指导等。医生人工接诊通过咨询病因、症状等多轮问询，可以对当前问诊数据是否是重症问诊数据做出判断。
然而，随着发起线上问诊的用户量激增，为减轻医生人工鉴别的繁重工作，医疗领域开始采用专家系统进行判别的模式。发明人意识到虽然专家系统可以减轻相关人员的工作量，但是由于线上用户问诊数据的多样性，专家系统对于当前问诊数据是否是重症问诊数据的鉴别准确性存在巨大挑战。
发明内容
一种基于人工智能的重症问诊数据识别方法，，所述方法包括：
获取与目标用户标识对应的问诊会话数据；
将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
一种基于人工智能的重症问诊数据识别装置，所述装置包括：
获取模块，用于获取与目标用户标识对应的问诊会话数据；
第一识别模块，用于将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
第二识别模块，用于根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
决策模块，用于结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现以下步骤：
获取与目标用户标识对应的问诊会话数据；
将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训 练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
一种计算机存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现以下步骤：
获取与目标用户标识对应的问诊会话数据；
将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
附图说明
图1为一个实施例中基于人工智能的重症问诊数据识别方法的应用场景图；
图2为一个实施例中基于人工智能的重症问诊数据识别方法的流程示意图；
图3为一个实施例中使用预测模型的流程框图；
图4为一个实施例中训练预测模型的流程框图；
图5为一个实施例中基于人工智能的重症问诊数据识别装置的结构框图；
图6为另一个实施例中基于人工智能的重症问诊数据识别装置的结构框图；
图7为一个实施例中计算机设备的内部结构图；
图8为另一个实施例中计算机设备的内部结构图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的基于人工智能的重症问诊数据识别方法，可以应用于如图1所示的应用环境中。其中，终端102通过网络与服务器104通过网络进行通信。具体地，终端102可以获取与目标用户标识对应的问诊会话数据，将该问诊会话数据发送至服务器104，服务器104将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果；其中，预测模型根据多维特征数据以及多维特征数据相应的训练标签训练得到，多维特征数据根据历史问诊数据提取得到、且多维特征数据包括实体特征数据和实体关系特征数据；服务器104然后根据问诊会话数据在专家知识库中命中的标签，确定与问诊会话数据对应的专家识别结果；服务器104再结合模型识别结果和专家识别结果，得到问诊会话数据是否为重症问诊数据的目标识别结果。其中，终端102可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备，服务器104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。
在另外的实施例中，终端102或者服务器104也可以分别单独用于执行该基于人工智能的重症问诊数据识别方法。本申请在此不做限定。
在一个实施例中，如图2所示，提供了一种基于人工智能的重症问诊数据识别方法， 以该方法应用于计算机设备为例进行说明，该计算机设备具体可以是图1中的终端或者服务器。该基于人工智能的重症问诊数据识别方法具体包括以下步骤：
步骤202，获取与目标用户标识对应的问诊会话数据。
其中，目标用户标识用于唯一标识一个用户。目标用户标识比如应用账号或者医院就诊卡号等。问诊会话是至少两个用户之间进行问诊交互的过程。至少两个用户包括与患者角色对应的用户和与医生角色对应的用户。问诊会话数据是问诊交互的过程中产生的数据。与医生角色对应的用户可以是医生本人也可以是人工智能机器人。
具体地，终端上可运行有在线问诊应用程序或者在线问诊网站，在线问诊应用程序或者在线问诊网站可提供问诊入口。用户通过终端基于问诊入口输入问诊会话数据以进行线上问诊。
在一个实施例中，问诊会话数据可以是语音数据、文本数据或图像数据等。
在一个实施例中，问诊会话数据可以包括与患者角色对应的问诊会话数据以及与医生角色对应的问诊会话数据。其中，与患者角色对应的问诊会话数据，比如用户基本信息、症状描述信息、症状照片、医学检查报告或者既往史信息等。可以理解，问诊用户可以是患者本人也可以不是患者本人，比如替小孩或者老人问诊的场景。与医生角色对应的问诊会话数据，比如疾病描述信息、症状分析信息、病因分析信息或者针对用户询问数据的答复信息等。
在一个实施例中，问诊会话数据可以是一次问诊过程中一轮或者多轮问答的对话数据。这样可以在用户首次提供的问诊会话数据即可识别问诊会话数据是否为重症问诊数据时尽早识别以进行相应处理；在信息量不足时，则引导用户提供更多的信息，以结合这些信息更加准确的进行重症问诊数据的识别。
其中，重症问诊数据是涉及急危重症的问诊会话数据。“急危重症”为医学术语，通常表示患者所得疾病为某种紧急、濒危的病症，应当尽早进行医学处理，否则可能对患者身体产生重度伤害或导致死亡。比如问诊会话数据中包括急危重症的临床表现症状等。急危重症的临床表现症状比如“昏厥”、“呼吸困难”等。
步骤204，将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果；其中，预测模型根据多维特征数据以及多维特征数据相应的训练标签训练得到，多维特征数据根据历史问诊数据提取得到、且多维特征数据包括实体特征数据和实体关系特征数据。
其中，预测模型是事先训练的、用于识别问诊会话数据是否为重症问诊数据的机器学习模型。机器学习模型可采用神经网络模型、支持向量机或者逻辑回归模型等。神经网络模型比如卷积神经网络、反向传播神经网络、反馈神经网络、径向基神经网络或者自组织神经网络等。
实体特征是反映实体本身特征的数据。比如，问诊会话数据“晚饭期间饮酒，现在腹痛难耐”中包括两个实体，第一实体为“饮酒”，第二实体为“腹痛”。实体关系特征是指反应至少两个实体之间关系的数据。比如问诊会话数据“晚饭期间饮酒，现在腹痛难耐”中“饮酒”与“腹痛”之间的实体关系为“因果关系”，即腹痛的诱因是饮酒。这里，一方面考虑到问诊会话数据中的实体为识别重症问诊数据的重要依据，另一方面还考虑到不同实体之间的关系也会影响识别结果，甚至同一实体在不同语境中也会存在不同的语义，进而也会影响识别结果。比如，问诊会话数据“大姨妈来了，腹痛难耐”中“大姨妈”是指月经，而非亲属称谓。在此，计算机设备则可在设计预测模型的输入数据时，融合多方面的信息，比如在数据的实体维度和实体关系维度这两个特征维度进行融合用作预测模型的输入数据，这样能够使得预测模型在训练中能学习到这两个特征维度的有效信息，提高模型对于重症问诊数据的识别能力。
具体地，计算机设备可将问诊会话数据输入预测模型，通过预测模型包括的多个神经 元对问诊会话数据进行处理，得到问诊会话数据对应的模型识别结果，再通过预测模型输出模型识别结果。其中，神经元是神经网络中最基本的结构，一般情况下，大多数的神经元是处于抑制状态，但是在神经元接收到输入信息，导致它的电位超过一个阈值，那么这个神经元就会被激活，处于“兴奋”状态，进而将输出信息传播至其他的神经元。连接神经元之间的连接线对应一个权重(其值称为权值)，通常不同的连接线对应不同的权重。各神经元的阈值及各神经元之间连接关系的权重，是在通过多维特征数据及多维特征数据相应的训练标签训练预测模型时确定的。神经元包括输入神经元、输出神经元和隐含神经元。
在一个实施例中，该基于人工智能的重症问诊数据识别方法还包括预测模型的训练步骤，该训练步骤具体包括：收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数据和实体关系特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
在一个实施例中，多维特征数据还包括意图特征数据。此时预测模型的训练步骤具体包括：收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
其中，意图特征是反映表达意向的数据。意图特征包括反映医生问询或者应答意图、以及用户问询或者应答意图的特征数据。比如，问诊会话数据“你现在是不是腹痛难耐”所要表达的症状确认意图；问诊会话数据“是的，我现在腹痛难耐”所要表达的是症状确认意图，等等。
这样，考虑到医生以及用户问询和回答的意图也能作为识别重症问诊数据的依据，因此可以在预测模型的输入涉及时，还引入意图特征，从而在实体维度、实体关系维度以及意图维度这三个特征维度进行融合，能够使得预测模型在训练中能学习到这三个特征维度的有效信息，提高模型对于重症问诊数据的识别能力，进而可以扩大重症问诊数据识别的方向。
关于预测模型训练步骤的具体内容可以参考后续实施例中的具体描述。
在一个实施例中，计算机设备可将问诊会话数据转换为预测模型能够处理的数据格式后，再将转换得到的数据输入预测模型。预测模型能够处理的数据格式比如向量格式或者矩阵格式等。
步骤206，根据问诊会话数据在专家知识库中命中的标签，确定与问诊会话数据对应的专家识别结果。
其中，知识库是指专家系统设计所应用的规则集合，包含规则所联系的事实及数据，它们的全体构成知识库。知识库与具体的专家系统有关，在本申请中专家知识库与医疗领域的问诊专家系统相关。专家知识库中的规则是对历史问诊过程中重症问诊数据中高频出现的数据进行提取得到的，规则所联系的事实包括历史问诊过程中的问诊会话数据是否为重症问诊数据，规则所联系的数据包括历史问诊过程中的问诊会话数据中频繁出现的疾病症状等标签。
具体地，计算机设备可采用规则引擎驱动，基于专家知识库的规则集合，确定问诊会话数据在专家知识库中所命中的标签。在该命中的标签为重症问诊数据对应的标签时，得到与问诊会话数据对应的专家识别结果为问诊会话数据是重症问诊数据；在该命中的标签不是重症问诊数据对应的标签时，得到与问诊会话数据对应的专家识别结果为问诊会话数据不是重症问诊数据。
步骤208，结合模型识别结果和专家识别结果，得到问诊会话数据是否为重症问诊数据的目标识别结果。
具体地，模型识别结果包括问诊会话数据是重症问诊数据，以及问诊会话数据不是重症问诊数据这两种识别结果。专家识别结果也包括问诊会话数据是重症问诊数据，以及问诊会话数据不是重症问诊数据这两种识别结果。在模型识别结果和专家识别结果都是问诊会话数据是重症问诊数据，那么可以得到问诊会话数据是重症问诊数据的目标识别结果。在模型识别结果和专家识别结果其中至少一个是问诊会话数据不是重症问诊数据，那么可以得到问诊会话数据不是重症问诊数据的目标识别结果。
在一个实施例中，专家识别结果也可以包括未识别出问诊会话数据是否为重症问诊数据这种情况。此时，可以将模型识别结果用作目标识别结果。
上述基于人工智能的重症问诊数据识别方法，在获取到用户的问诊会话数据后，一方面将该问诊会话数据输入预测模型，得到模型识别结果，另一方面基于专家知识库对该问诊会话数据进行识别，得到专家识别结果，再结合这模型识别结果和专家识别结果得到最终的识别结果。由于预测模型是基于多维特征数据训练得到的，且该多维特征数据包括实体特征数据和实体关系特征数据，那么在训练过程中预测模型可以学习到不同维度的信息，并结合上下文语义环境更好地理解语言的逻辑，从而提高预测模型对重症问诊数据的识别能力；这样采用模型预测和专家系统结合的方式进行重症问诊数据识别，可以弥补仅依赖专家系统的不足，提高了重症问诊数据的识别的准确率。
在一个实施例中，计算机设备在得到问诊会话数据是否为重症问诊数据的目标识别结果后，可根据目标识别结果执行与该目标识别结果相应的操作。
在一个具体的实施例中，该基于人工智能的重症问诊数据识别方法还包括：当得到问诊会话数据是重症问诊数据的目标识别结果时，则将问诊会话数据所属的问诊会话接入至医生终端；当得到问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进问诊会话数据所属的问诊会话。
具体地，图3示出了一个实施例中使用预测模型的流程框图。参考该图，计算机设备获取问诊会话数据后可将该问诊会话数据并行输入预测模型和专家系统，一方面通过预测模型识别问诊会话数据是否为重症问诊数据，得到模型识别结果，另一方面则通过专家系统识别问诊会话数据是否为重症问诊数据，得到专家识别结果。然后通过决策器结合模型识别结果和专家识别结果，得到问诊会话数据是否为重症问诊数据的目标识别结果。当得到问诊会话数据是重症问诊数据的目标识别结果时，则将问诊会话数据所属的问诊会话接入至医生终端，医生通过医生终端人工介入以审核重症问诊数据的最终识别结果，在识别正确时可以及时地进一步处理，比如给出就诊建议等。当得到问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进问诊会话数据所属的问诊会话，比如继续通过人工智能机器人与用户交互以进行问诊。
在本实施例中，在得到不同的目标识别结果时，立即进行相应的下一步操作，以在问诊会话数据是重症问诊数据可以使的情况紧急急需帮助的用户得到有效应答，在问诊会话数据不是重症问诊数据，可以有序地继续进行线上问诊。
在一个实施例中，问诊会话数据存储于区块链中。需要强调的是，为进一步保证上述问诊会话数据的私密和安全性，上述问诊会话数据还可以存储于一区块链的节点中。
关于前述实施例中涉及的预测模型的训练步骤的具体内容，可以参考以下实施例中的具体描述。
在一个实施例中，该基于人工智能的重症问诊数据识别方法还包括：收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数据和实体关系特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征 数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
在一个实施例中，多维特征数据还包括意图特征数据；该基于人工智能的重症问诊数据识别方法还包括：收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
具体地，计算机设备可先收集历史问诊数据以及历史问诊数据相应的训练标签。该训练标签可以是人工进行标注的结果，表示历史问诊数据是否为重症问诊数据。计算机设备然后可从历史问诊数据中提取实体特征数据和实体关系特征数据，生成历史问诊数据所对应的多维特征数据，该多维特征数据至少包括两个特征维度，将该多维特征数据和历史问诊数据共同用作待训练的预测模型的输入数据。计算机设备也可从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据，该多维特征数据至少包括三个特征维度，将该多维特征数据和历史问诊数据共同用作待训练的预测模型的输入数据。其中，基于多个特征数据生成多维特征数据可以是将这多个特征数据拼接或者融合。
此后，计算机设备可以得到待训练的预测模型所输出的预测识别结果，然后根据该预测识别结果与训练标签的差异构建训练损失函数，按照最小化该训练损失函数的方向，采用反向传播算法优化预测模型的参数，训练完毕即可将得到的各隐含神经元和输入/输出神经元的权重和阈值，进而得到训练完毕的预测模型的模型参数文件。计算机设备可将该模型参数文件存入专家知识库。
在本实施例中，计算机设备在设计预测模型的输入数据时，融合多方面的信息，比如在数据的实体维度和实体关系维度这两个特征维度进行融合用作预测模型的输入数据，这样能够使得预测模型在训练中能够至少学习到这两个特征维度的有效信息，提高模型对于重症问诊数据的识别能力。
另外，还考虑到医生以及用户问询和回答的意图也能作为识别重症问诊数据的依据，因此可以在设计预测模型的输入数据时，引入意图特征，从而在实体维度、实体关系维度以及意图维度这三个特征维度进行融合，能够使得预测模型在训练中能至少学习到这三个特征维度的有效信息，提高模型对于重症问诊数据的识别能力，进而可以扩大重症问诊数据识别的方向。
可以理解，在实际使用过程中，特征提取的方式很多，主要从以下几个方面进行：
在一个实施例中，从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据，包括：采用规则引擎驱动抽取历史问诊数据对应的实体特征数据；采用序列标注模型对历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取命名实体之间的实体关系特征数据；采用语义模型提取历史问诊数据中的意图特征数据；根据实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据。
(1)在实体特征数据提取方面，在一个具体的实施例中，采用规则引擎驱动抽取历史问诊数据对应的实体特征数据，包括：对历史问诊数据进行分词得到分词结果；基于专家知识库所提供的实体标签，采用规则引擎驱动根据分词结果抽取实体标签，得到历史问诊数据所对应的实体特征数据。
具体地，计算机设备可对历史问诊数据进行分词，得到分词结果，采用规则引擎按照专家知识库的规则集合中的规则，根据从专家知识库提供的实体标签，分词结果中抽取实体标签，得到实体特征数据。比如：实体标签：实体特征数据→患者：婴儿，症状：呕吐、 体重不增等。其中，规则引擎是一种嵌入在应用程序中的组件，接受数据输入，解释规则，并根据规则做出决策。
这样，在提取实体特征数据时，有效地利用了专家知识库中的有效数据，提升了特征提取的效率和有效性，避免了对长尾词耗费不必要的训练时间。
(2)在实体关系特征数据提取方面，在一个具体的实施例中，计算机设备可将历史问诊数据按问诊会话进行分组，将一个问诊会话的历史问诊数据按照应答顺序生成问诊数据序列，然后采用序列标注模型对问诊数据序列进行命名实体识别，并基于识别出的命名实体抽取命名实体之间的实体关系特征数据。比如命名实体(饮酒、诱因、腹痛)，抽取得到实体关系：腹痛的诱因是饮酒。这样，在进行命名实体识别和实体关系抽取时，基于序列的处理方式能有效地结合上下文信息，能够进行更准确的实体间的关系抽取，具有广泛的实用意义。
(3)在意图特征数据提取方面，在一个具体的实施例中，采用语义模型提取历史问诊数据中的意图特征数据，包括：基于专家知识库所提供的意图标签，采用语义模型提取历史问诊数据所对应的意图标签，得到历史问诊数据所对应的意图特征数据。
具体地，计算机设备在训练预测模型的阶段，还可在第三方面采用语义模型来提取历史问诊数据的意图特征。其中，计算机设备可获取历史问诊数据作为样本，人工标注意图类别标签(专家知识库所提供的意图标签)，有监督训练语义模型，此后即可利用训练好的语义模型来提取历史问诊数据的意图特征。
在一个实施例中，该基于人工智能的重症问诊数据识别方法还包括：从历史问诊数据中筛选出历史重症问诊数据；统计历史重症问诊数据中实体关键词的出现频次；筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库；确定历史重症问诊数据所对应的问诊意图，得到意图标签；将意图标签加入专家知识库。
具体地，计算机设备还可建立专家知识库。计算机设备可先获取历史问诊数据中识别出的重症问诊数据，得到重症样本数据集；然后统计重症样本数据集中包括的诱因、对象、症状等实体关键词在重症样本数据集中的出现频次，筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库。此外，计算机设备还可确定历史重症问诊数据所对应的问诊意图，得到意图标签；将意图标签加入专家知识库。
这样，在训练预测模型时提取训练样本的多维特征时，可以利用专家系统准确的归纳总结出的标签，来抽取训练样本的多维特征，避免了对长尾词耗费不必要的训练时间，大大提升了机器学习模型训练的训练时间。
举例说明，图4示出了一个实施例中训练预测模型的流程框图。参考该图，可以看到，计算机设备可先进行数据准备，即收集历史问诊数据以及历史问诊数据相应的训练标签，然后一方面基于专家系统，基于专家知识库所提供的各类数据从历史问诊数据提取实体特征数据，一方面基于语义理解从历史问诊数据提取实体关系特征数据，还结合专家知识库所提供的各类数据和语义理解，从历史问诊数据提取意图特征数据。计算机设备再将三方面提取的特征数据进行融合，再结合历史问诊数据用作输入数据，构建神经网络结构，进行训练，得到用于识别重症问诊数据的预测模型。
应该理解的是，虽然上述实施例的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，上述实施例的流程图中的至少一部分步骤可以包括多个步骤或者多个阶段，这些步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤中的步骤或者阶段的至少一部分轮流或者交替地执行。
在一个实施例中，如图5所示，提供了一种基于人工智能的重症问诊数据识别装置， 包括：获取模块501、第一识别模块502、第二识别模块503和决策模块504，其中：
获取模块501，用于获取与目标用户标识对应的问诊会话数据；
第一识别模块502，用于将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
第二识别模块503，用于根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
决策模块504，用于结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
在一个实施例中，第一识别模块502还用于将问诊会话数据输入预测模型，通过预测模型包括的多个神经元对问诊会话数据进行处理，得到问诊会话数据对应的模型识别结果；通过预测模型输出模型识别结果；其中，各神经元的阈值及各神经元之间连接关系的权重，在通过多维特征数据及多维特征数据相应的训练标签训练预测模型时确定，多维特征数据还包括意图特征数据。
如图6，在一个实施例中，多维特征数据还包括意图特征数据；基于人工智能的重症问诊数据识别装置还包括：训练模块505，用于收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
在一个实施例中，训练模块505还用于采用规则引擎驱动抽取历史问诊数据对应的实体特征数据；采用序列标注模型对历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取命名实体之间的实体关系特征数据；采用语义模型提取历史问诊数据中的意图特征数据；根据实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据。
在一个实施例中，训练模块505还用于对历史问诊数据进行分词得到分词结果；基于专家知识库所提供的实体标签，采用规则引擎驱动根据分词结果抽取实体标签，得到历史问诊数据所对应的实体特征数据；基于专家知识库所提供的意图标签，采用语义模型提取历史问诊数据所对应的意图标签，得到历史问诊数据所对应的意图特征数据。
在一个实施例中，训练模块505还用于从历史问诊数据中筛选出历史重症问诊数据；统计历史重症问诊数据中实体关键词的出现频次；筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库；确定历史重症问诊数据所对应的问诊意图，得到意图标签；将意图标签加入专家知识库。
在一个实施例中，问诊会话数据存储于区块链中；决策模块504还用于当得到问诊会话数据是重症问诊数据的目标识别结果时，则将问诊会话数据所属的问诊会话接入至医生终端；当得到问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进问诊会话数据所属的问诊会话。
上述基于人工智能的重症问诊数据识别装置，在获取到用户的问诊会话数据后，一方面将该问诊会话数据输入预测模型，得到模型识别结果，另一方面基于专家知识库对该问诊会话数据进行识别，得到专家识别结果，再结合这模型识别结果和专家识别结果得到最终的识别结果。由于预测模型是基于多维特征数据训练得到的，且该多维特征数据包括实体特征数据和实体关系特征数据，那么在训练过程中预测模型可以学习到不同维度的信息，并结合上下文语义环境更好地理解语言的逻辑，从而提高预测模型对基于人工智能的重症 问诊数据识别能力；这样采用模型预测和专家系统结合的方式进行重症问诊数据识别，可以弥补仅依赖专家系统的不足，提高了基于人工智能的重症问诊数据识别的准确率。
关于基于人工智能的重症问诊数据识别装置的具体限定可以参见上文中对于基于人工智能的重症问诊数据识别方法的限定，在此不再赘述。上述基于人工智能的重症问诊数据识别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非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的数据库用于存储基于人工智能的重症问诊数据识别数据。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。该计算机程序被处理器执行时以实现一种基于人工智能的重症问诊数据识别方法。
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终端，其内部结构图可以如图8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、显示屏和输入装置。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非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。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。该计算机程序被处理器执行时以实现一种基于人工智能的重症问诊数据识别方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图7或8中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种计算机设备，包括存储器和处理器，该存储器存储有计算机程序，该处理器执行计算机程序时实现以下步骤：获取与目标用户标识对应的问诊会话数据；将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果；其中，预测模型根据多维特征数据以及多维特征数据相应的训练标签训练得到，多维特征数据根据历史问诊数据提取得到、且多维特征数据包括实体特征数据和实体关系特征数据；根据问诊会话数据在专家知识库中命中的标签，确定与问诊会话数据对应的专家识别结果；结合模型识别结果和专家识别结果，得到问诊会话数据是否为重症问诊数据的目标识别结果。
在一个实施例中，将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果，包括：将问诊会话数据输入预测模型，通过预测模型包括的多个神经元对问诊会话数据进行处理，得到问诊会话数据对应的模型识别结果；通过预测模型输出模型识别结果；其中，各神经元的阈值及各神经元之间连接关系的权重，在通过多维特征数据及多维特征数据相应的训练标签训练预测模型时确定，多维特征数据还包括意图特征数据。
在一个实施例中，多维特征数据还包括意图特征数据。处理器执行计算机程序时还实现以下步骤：收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数 据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
在一个实施例中，从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据，包括：采用规则引擎驱动抽取历史问诊数据对应的实体特征数据；采用序列标注模型对历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取命名实体之间的实体关系特征数据；采用语义模型提取历史问诊数据中的意图特征数据；根据实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据。
在一个实施例中，采用规则引擎驱动抽取历史问诊数据对应的实体特征数据，包括：对历史问诊数据进行分词得到分词结果；基于专家知识库所提供的实体标签，采用规则引擎驱动根据分词结果抽取实体标签，得到历史问诊数据所对应的实体特征数据。采用语义模型提取历史问诊数据中的意图特征数据，包括：基于专家知识库所提供的意图标签，采用语义模型提取历史问诊数据所对应的意图标签，得到历史问诊数据所对应的意图特征数据。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：从历史问诊数据中筛选出历史重症问诊数据；统计历史重症问诊数据中实体关键词的出现频次；筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库；确定历史重症问诊数据所对应的问诊意图，得到意图标签；将意图标签加入专家知识库。
在一个实施例中，问诊会话数据存储于区块链中；处理器执行计算机程序时还实现以下步骤：当得到问诊会话数据是重症问诊数据的目标识别结果时，则将问诊会话数据所属的问诊会话接入至医生终端；当得到问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进问诊会话数据所属的问诊会话。
在一个实施例中，提供了一种计算机存储介质，所述计算机存储介质可以是易失性的，也可以是非易失性的，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：获取与目标用户标识对应的问诊会话数据；将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果；其中，预测模型根据多维特征数据以及多维特征数据相应的训练标签训练得到，多维特征数据根据历史问诊数据提取得到、且多维特征数据包括实体特征数据和实体关系特征数据；根据问诊会话数据在专家知识库中命中的标签，确定与问诊会话数据对应的专家识别结果；结合模型识别结果和专家识别结果，得到问诊会话数据是否为重症问诊数据的目标识别结果。
在一个实施例中，将问诊会话数据输入预测模型，通过预测模型输出与问诊会话数据对应的模型识别结果，包括：将问诊会话数据输入预测模型，通过预测模型包括的多个神经元对问诊会话数据进行处理，得到问诊会话数据对应的模型识别结果；通过预测模型输出模型识别结果；其中，各神经元的阈值及各神经元之间连接关系的权重，在通过多维特征数据及多维特征数据相应的训练标签训练预测模型时确定，多维特征数据还包括意图特征数据。
在一个实施例中，多维特征数据还包括意图特征数据；计算机程序被处理器执行时还实现以下步骤：收集历史问诊数据以及历史问诊数据相应的训练标签；历史问诊数据相应的训练标签用于表示历史问诊数据是否为重症问诊数据；从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据；将历史问诊数据和历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；基于预测模型的预测识别结果与训练标签训练预测模型。
在一个实施例中，从历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据，包括：采用规则引擎驱动抽取历史问 诊数据对应的实体特征数据；采用序列标注模型对历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取命名实体之间的实体关系特征数据；采用语义模型提取历史问诊数据中的意图特征数据；根据实体特征数据、实体关系特征数据和意图特征数据，生成历史问诊数据所对应的多维特征数据。
在一个实施例中，采用规则引擎驱动抽取历史问诊数据对应的实体特征数据，包括：对历史问诊数据进行分词得到分词结果；基于专家知识库所提供的实体标签，采用规则引擎驱动根据分词结果抽取实体标签，得到历史问诊数据所对应的实体特征数据。采用语义模型提取历史问诊数据中的意图特征数据，包括：基于专家知识库所提供的意图标签，采用语义模型提取历史问诊数据所对应的意图标签，得到历史问诊数据所对应的意图特征数据。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：从历史问诊数据中筛选出历史重症问诊数据；统计历史重症问诊数据中实体关键词的出现频次；筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库；确定历史重症问诊数据所对应的问诊意图，得到意图标签；将意图标签加入专家知识库。
在一个实施例中，问诊会话数据存储于区块链中；计算机程序被处理器执行时还实现以下步骤：当得到问诊会话数据是重症问诊数据的目标识别结果时，则将问诊会话数据所属的问诊会话接入至医生终端；当得到问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进问诊会话数据所属的问诊会话。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
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，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，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。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种基于人工智能的重症问诊数据识别方法，其中，所述方法包括：
    获取与目标用户标识对应的问诊会话数据；
    将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
    根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
    结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
  2. 根据权利要求1所述的方法，其中，所述将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果，包括：
    将所述问诊会话数据输入预测模型，通过所述预测模型包括的多个神经元对所述问诊会话数据进行处理，得到所述问诊会话数据对应的模型识别结果；
    通过所述预测模型输出所述模型识别结果；
    其中，各所述神经元的阈值及各所述神经元之间连接关系的权重，在通过多维特征数据及所述多维特征数据相应的训练标签训练所述预测模型时确定，所述多维特征数据还包括意图特征数据。
  3. 根据权利要求1所述的方法，其中，所述多维特征数据还包括意图特征数据；所述方法还包括：
    收集历史问诊数据以及所述历史问诊数据相应的训练标签；所述历史问诊数据相应的训练标签用于表示所述历史问诊数据是否为重症问诊数据；
    从所述历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据；
    将所述历史问诊数据和所述历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；
    基于所述预测模型的预测识别结果与所述训练标签训练所述预测模型。
  4. 根据权利要求3所述的方法，其中，所述从所述历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据，包括：
    采用规则引擎驱动抽取所述历史问诊数据对应的实体特征数据；
    采用序列标注模型对所述历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取所述命名实体之间的实体关系特征数据；
    采用语义模型提取所述历史问诊数据中的意图特征数据；
    根据所述实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据。
  5. 根据权利要求4所述的方法，其中，所述采用规则引擎驱动抽取所述历史问诊数据对应的实体特征数据，包括：
    对所述历史问诊数据进行分词得到分词结果；
    基于所述专家知识库所提供的实体标签，采用规则引擎驱动根据所述分词结果抽取实体标签，得到所述历史问诊数据所对应的实体特征数据；
    所述采用语义模型提取所述历史问诊数据中的意图特征数据，包括：
    基于所述专家知识库所提供的意图标签，采用语义模型提取所述历史问诊数据所对应 的意图标签，得到所述历史问诊数据所对应的意图特征数据。
  6. 根据权利要求5所述的方法，其中，所述方法还包括：
    从所述历史问诊数据中筛选出历史重症问诊数据；
    统计所述历史重症问诊数据中实体关键词的出现频次；
    筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库；
    确定所述历史重症问诊数据所对应的问诊意图，得到意图标签；
    将所述意图标签加入所述专家知识库。
  7. 根据权利要求1-6中任一项所述的方法，其中，所述问诊会话数据存储于区块链中；
    所述方法还包括：
    当得到所述问诊会话数据是重症问诊数据的目标识别结果时，则将所述问诊会话数据所属的问诊会话接入至医生终端；
    当得到所述问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进所述问诊会话数据所属的问诊会话。
  8. 一种基于人工智能的重症问诊数据识别装置，其中，所述装置包括：
    获取模块，用于获取与目标用户标识对应的问诊会话数据；
    第一识别模块，用于将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
    第二识别模块，用于根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
    决策模块，用于结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
  9. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其中，所述处理器执行所述计算机程序时实现如下步骤：
    获取与目标用户标识对应的问诊会话数据；
    将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
    根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
    结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
  10. 根据权利要求9所述的计算机设备，其中，所述将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果，包括：
    将所述问诊会话数据输入预测模型，通过所述预测模型包括的多个神经元对所述问诊会话数据进行处理，得到所述问诊会话数据对应的模型识别结果；
    通过所述预测模型输出所述模型识别结果；
    其中，各所述神经元的阈值及各所述神经元之间连接关系的权重，在通过多维特征数据及所述多维特征数据相应的训练标签训练所述预测模型时确定，所述多维特征数据还包括意图特征数据。
  11. 根据权利要求9所述的计算机设备，其中，所述多维特征数据还包括意图特征数据；所述处理器执行所述计算机程序时还实现如下步骤：
    收集历史问诊数据以及所述历史问诊数据相应的训练标签；所述历史问诊数据相应的 训练标签用于表示所述历史问诊数据是否为重症问诊数据；
    从所述历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据；
    将所述历史问诊数据和所述历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；
    基于所述预测模型的预测识别结果与所述训练标签训练所述预测模型。
  12. 根据权利要求11所述的计算机设备，其中，所述从所述历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据，包括：
    采用规则引擎驱动抽取所述历史问诊数据对应的实体特征数据；
    采用序列标注模型对所述历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取所述命名实体之间的实体关系特征数据；
    采用语义模型提取所述历史问诊数据中的意图特征数据；
    根据所述实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据。
  13. 根据权利要求12所述的计算机设备，其中，所述采用规则引擎驱动抽取所述历史问诊数据对应的实体特征数据，包括：
    对所述历史问诊数据进行分词得到分词结果；
    基于所述专家知识库所提供的实体标签，采用规则引擎驱动根据所述分词结果抽取实体标签，得到所述历史问诊数据所对应的实体特征数据；
    所述采用语义模型提取所述历史问诊数据中的意图特征数据，包括：
    基于所述专家知识库所提供的意图标签，采用语义模型提取所述历史问诊数据所对应的意图标签，得到所述历史问诊数据所对应的意图特征数据。
  14. 根据权利要求13所述的计算机设备，其中，所述处理器执行所述计算机程序时还实现如下步骤：
    从所述历史问诊数据中筛选出历史重症问诊数据；
    统计所述历史重症问诊数据中实体关键词的出现频次；
    筛选出现频次高于预设阈值的实体关键词用作实体标签加入专家知识库；
    确定所述历史重症问诊数据所对应的问诊意图，得到意图标签；
    将所述意图标签加入所述专家知识库。
  15. 根据权利要求9-14中任一项所述的计算机设备，其中，所述问诊会话数据存储于区块链中；所述处理器执行所述计算机程序时还实现如下步骤：
    当得到所述问诊会话数据是重症问诊数据的目标识别结果时，则将所述问诊会话数据所属的问诊会话接入至医生终端；
    当得到所述问诊会话数据不是重症问诊数据的目标识别结果时，则继续推进所述问诊会话数据所属的问诊会话。
  16. 一种计算机存储介质，其上存储有计算机程序，其中，所述计算机程序被处理器执行时实现如下步骤：
    获取与目标用户标识对应的问诊会话数据；
    将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果；其中，所述预测模型根据多维特征数据以及所述多维特征数据相应的训练标签训练得到，所述多维特征数据根据历史问诊数据提取得到、且所述多维特征数据包括实体特征数据和实体关系特征数据；
    根据所述问诊会话数据在专家知识库中命中的标签，确定与所述问诊会话数据对应的专家识别结果；
    结合所述模型识别结果和所述专家识别结果，得到所述问诊会话数据是否为重症问诊数据的目标识别结果。
  17. 根据权利要求16所述的计算机存储介质，其中，所述将所述问诊会话数据输入预测模型，通过所述预测模型输出与所述问诊会话数据对应的模型识别结果，包括：
    将所述问诊会话数据输入预测模型，通过所述预测模型包括的多个神经元对所述问诊会话数据进行处理，得到所述问诊会话数据对应的模型识别结果；
    通过所述预测模型输出所述模型识别结果；
    其中，各所述神经元的阈值及各所述神经元之间连接关系的权重，在通过多维特征数据及所述多维特征数据相应的训练标签训练所述预测模型时确定，所述多维特征数据还包括意图特征数据。
  18. 根据权利要求16所述的计算机存储介质，其中，所述多维特征数据还包括意图特征数据；所述计算机程序被处理器执行时还实现如下步骤：
    收集历史问诊数据以及所述历史问诊数据相应的训练标签；所述历史问诊数据相应的训练标签用于表示所述历史问诊数据是否为重症问诊数据；
    从所述历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据；
    将所述历史问诊数据和所述历史问诊数据所对应的多维特征数据，共同输入待训练的预测模型，得到预测识别结果；
    基于所述预测模型的预测识别结果与所述训练标签训练所述预测模型。
  19. 根据权利要求18所述的计算机存储介质，其中，所述从所述历史问诊数据中提取实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据，包括：
    采用规则引擎驱动抽取所述历史问诊数据对应的实体特征数据；
    采用序列标注模型对所述历史问诊数据进行命名实体识别，并基于识别出的命名实体抽取所述命名实体之间的实体关系特征数据；
    采用语义模型提取所述历史问诊数据中的意图特征数据；
    根据所述实体特征数据、实体关系特征数据和意图特征数据，生成所述历史问诊数据所对应的多维特征数据。
  20. 根据权利要求19所述的计算机存储介质，其中，所述采用规则引擎驱动抽取所述历史问诊数据对应的实体特征数据，包括：
    对所述历史问诊数据进行分词得到分词结果；
    基于所述专家知识库所提供的实体标签，采用规则引擎驱动根据所述分词结果抽取实体标签，得到所述历史问诊数据所对应的实体特征数据；
    所述采用语义模型提取所述历史问诊数据中的意图特征数据，包括：
    基于所述专家知识库所提供的意图标签，采用语义模型提取所述历史问诊数据所对应的意图标签，得到所述历史问诊数据所对应的意图特征数据。
PCT/CN2021/084349 2020-09-30 2021-03-31 基于人工智能的重症问诊数据识别方法、装置、设备及介质 WO2022068160A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011065413.XA CN112201359B (zh) 2020-09-30 2020-09-30 基于人工智能的重症问诊数据识别方法及装置
CN202011065413.X 2020-09-30

Publications (1)

Publication Number Publication Date
WO2022068160A1 true WO2022068160A1 (zh) 2022-04-07

Family

ID=74013788

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/084349 WO2022068160A1 (zh) 2020-09-30 2021-03-31 基于人工智能的重症问诊数据识别方法、装置、设备及介质

Country Status (2)

Country Link
CN (1) CN112201359B (zh)
WO (1) WO2022068160A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115759113A (zh) * 2022-11-08 2023-03-07 贝壳找房(北京)科技有限公司 识别对话信息中的句子语义的方法及装置
CN117725961A (zh) * 2024-02-18 2024-03-19 智慧眼科技股份有限公司 医疗意图识别模型训练方法、医疗意图识别方法及设备
CN118349600A (zh) * 2024-06-17 2024-07-16 成都中科合迅科技有限公司 基于物化视图的多维度关系展现控制方法和系统

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112201359B (zh) * 2020-09-30 2024-05-03 平安科技（深圳）有限公司 基于人工智能的重症问诊数据识别方法及装置
CN113284631B (zh) * 2021-04-09 2024-08-27 阿里健康科技(中国)有限公司 线上问诊交互方法
CN114780538A (zh) * 2022-03-25 2022-07-22 北京京东拓先科技有限公司 一种信息存储方法及装置、存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108986908A (zh) * 2018-05-31 2018-12-11 平安医疗科技有限公司 问诊数据处理方法、装置、计算机设备和存储介质
CN109036506A (zh) * 2018-07-25 2018-12-18 平安科技（深圳）有限公司 互联网医疗问诊的监管方法、电子装置及可读存储介质
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
US20190311814A1 (en) * 2018-04-06 2019-10-10 Curai, Inc. Systems and methods for responding to healthcare inquiries
CN110675944A (zh) * 2019-09-20 2020-01-10 京东方科技集团股份有限公司 分诊方法及装置、计算机设备及介质
CN112201359A (zh) * 2020-09-30 2021-01-08 平安科技（深圳）有限公司 基于人工智能的重症问诊数据识别方法及装置

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040122704A1 (en) * 2002-12-18 2004-06-24 Sabol John M. Integrated medical knowledge base interface system and method
CA2715825C (en) * 2008-02-20 2017-10-03 Mcmaster University Expert system for determining patient treatment response
CN104217132A (zh) * 2014-09-30 2014-12-17 中国科学院苏州生物医学工程技术研究所 一种用于病患康复的专家系统和诊疗系统
EP3255573A1 (en) * 2016-06-10 2017-12-13 Electronics and Telecommunications Research Institute Clinical decision supporting ensemble system and clinical decison supporting method using the same
CN107247868B (zh) * 2017-05-18 2020-05-12 深思考人工智能机器人科技（北京）有限公司 一种人工智能辅助问诊系统
CN107669250A (zh) * 2017-11-02 2018-02-09 佛山鑫进科技有限公司 一种基于大数据分析的人体健康度监测方法
CN109119130A (zh) * 2018-07-11 2019-01-01 上海夏先机电科技发展有限公司 一种基于云计算的大数据健康管理系统及方法
CN109935336B (zh) * 2019-01-15 2023-10-31 北京思普科软件股份有限公司 一种儿童呼吸科疾病的智能辅助诊断系统
CN110838368B (zh) * 2019-11-19 2022-11-15 广州西思数字科技有限公司 一种基于中医临床知识图谱的主动问诊机器人
CN111145903B (zh) * 2019-12-18 2024-08-06 东北大学 获取眩晕症问诊文本的方法、装置、电子设备及问诊系统
CN111430029B (zh) * 2020-03-24 2021-04-16 浙江达美生物技术有限公司 基于人工智能的多维度脑卒中预防筛查方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190311814A1 (en) * 2018-04-06 2019-10-10 Curai, Inc. Systems and methods for responding to healthcare inquiries
CN108986908A (zh) * 2018-05-31 2018-12-11 平安医疗科技有限公司 问诊数据处理方法、装置、计算机设备和存储介质
CN109036506A (zh) * 2018-07-25 2018-12-18 平安科技（深圳）有限公司 互联网医疗问诊的监管方法、电子装置及可读存储介质
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
CN110675944A (zh) * 2019-09-20 2020-01-10 京东方科技集团股份有限公司 分诊方法及装置、计算机设备及介质
CN112201359A (zh) * 2020-09-30 2021-01-08 平安科技（深圳）有限公司 基于人工智能的重症问诊数据识别方法及装置

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115759113A (zh) * 2022-11-08 2023-03-07 贝壳找房(北京)科技有限公司 识别对话信息中的句子语义的方法及装置
CN115759113B (zh) * 2022-11-08 2023-11-03 贝壳找房(北京)科技有限公司 识别对话信息中的句子语义的方法及装置
CN117725961A (zh) * 2024-02-18 2024-03-19 智慧眼科技股份有限公司 医疗意图识别模型训练方法、医疗意图识别方法及设备
CN118349600A (zh) * 2024-06-17 2024-07-16 成都中科合迅科技有限公司 基于物化视图的多维度关系展现控制方法和系统

Also Published As

Publication number Publication date
CN112201359A (zh) 2021-01-08
CN112201359B (zh) 2024-05-03

Similar Documents

Publication Publication Date Title
WO2022068160A1 (zh) 基于人工智能的重症问诊数据识别方法、装置、设备及介质
Zeberga et al. [Retracted] A Novel Text Mining Approach for Mental Health Prediction Using Bi‐LSTM and BERT Model
US11810671B2 (en) System and method for providing health information
US11232365B2 (en) Digital assistant platform
US20180218127A1 (en) Generating a Knowledge Graph for Determining Patient Symptoms and Medical Recommendations Based on Medical Information
US20180218126A1 (en) Determining Patient Symptoms and Medical Recommendations Based on Medical Information
EP3776247A1 (en) Systems and methods for responding to healthcare inquiries
Li et al. Reliable medical diagnosis from crowdsourcing: Discover trustworthy answers from non-experts
US10755197B2 (en) Rule-based feature engineering, model creation and hosting
Wanyan et al. Deep learning with heterogeneous graph embeddings for mortality prediction from electronic health records
Gupta et al. A novel deep similarity learning approach to electronic health records data
Arumugham et al. An explainable deep learning model for prediction of early‐stage chronic kidney disease
Hasan et al. Improving Medical Image Decision‐Making by Leveraging Metacognitive Processes and Representational Similarity
Mirtaheri et al. A self-attention TCN-based model for suicidal ideation detection from social media posts
JP2023527686A (ja) テキストデータの状態識別及び分類のためのシステム及び方法
Zaghir et al. Real-world patient trajectory prediction from clinical notes using artificial neural networks and UMLS-based extraction of concepts
CN117238527B (zh) 基于大数据的医疗信息管理方法及系统
Cui et al. Intelligent recommendation for departments based on medical knowledge graph
US11170172B1 (en) System and method for actionizing comments
Galitsky LLM-Based Personalized Recommendations in Health
Theodorou et al. Synthesize extremely high-dimensional longitudinal electronic health records via hierarchical autoregressive language model
CN115358817A (zh) 基于社交数据的智能产品推荐方法、装置、设备及介质
US20200303033A1 (en) System and method for data curation
Echajei et al. Integrating Causal Inference and Machine Learning for Early Diagnosis and Management of Diabetes.
Damen et al. Pastel: a semantic platform for assisted clinical trial patient recruitment

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: 21873827

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: 21873827

Country of ref document: EP

Kind code of ref document: A1