Intelligent pre-examination triage method and system

Through deep learning and natural language processing technology, the consultation topics and conversation elements of pediatric pre-examination consultation are integrated to generate personalized reports, solving the problem of insufficient efficiency and accuracy of traditional pre-examination and triage, and improving the diagnosis and treatment efficiency and services of pediatric medical care quality.

CN118983065APending Publication Date: 2024-11-19MATERNAL & CHILD HEALTH HOSPITAL OF TONGZHOU DISTRICT BEIJING
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411063257.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The traditional pre-examination and triage method is insufficient in pediatric medical care, especially in emergency and fever clinics, which leads to poor patient diagnosis and treatment safety and medical resources utilization, and the existing self-service terminal functions are single and humanized and intelligent.

Method used

Using intelligent pre-test and triage methods based on deep learning and natural language processing technology, we can obtain and analyze initial pre-test consultation conversation text through semantic embedding and semantic attention modules, integrate consulting topics and conversation elements, and generate personalized pre-test consultation guidance. Report to optimize the outpatient and emergency medical treatment process.

Benefits of technology

It improves the diagnosis and treatment efficiency and accuracy of pre-examination and triage, provides convenient, efficient, high-quality and safe smart medical services, reduces patient waiting time and improves medical service experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118983065A_ABST
    Figure CN118983065A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides an intelligent pre-examination triage method and system, and the method can intelligently obtain and analyze an initial pre-examination consultation session text, accurately captures a consultation theme and session elements through a dual semantic embedding technology, and provides a solid data basis for subsequent triage work. By means of the method, the hospital seeing process of outpatient and emergency treatment can be optimized, the diagnosis and treatment efficiency of pre-examination triage medical care is improved, and convenient, efficient, high-quality and safe intelligent medical services are provided for patients.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of smart medical technology, and in particular to a smart pre-examination and triage method and system. Background Art

[0002] As the public continues to pay more attention to healthy living, the demand for medical diagnosis and treatment is also growing. Especially in the field of pediatrics, due to the particularity of children's diseases and the characteristics of seasonal and clustered diseases, the contradiction between supply and demand of pediatric medical resources has become increasingly prominent. However, the number of pediatric medical staff has increased relatively slowly, which has seriously affected the efficiency, safety and speed of children's medical treatment.

[0003] In this context, pre-examination and triage, as the primary link in the diagnosis and treatment process, has become increasingly important. Traditional pre-examination and triage methods, such as relying on medical staff to perform manual triage, can no longer meet the efficiency and accuracy requirements of modern medical services. Especially in emergency and fever clinics, the accuracy and efficiency of pre-examination and triage are directly related to the patient's diagnosis and treatment safety and the rational use of medical resources.

[0004] In recent years, with the rapid development of Internet+ and intelligent technology, smart healthcare has become an important part of modern medical technology. Automated and intelligent medical equipment has played an important role in reducing the labor intensity of medical staff and improving the efficiency of diagnosis and treatment. However, the functions of traditional self-service machines on the current market are relatively simple, mainly staying at basic functions such as registration, number collection, and payment, and cannot meet the complex needs of outpatient and emergency pre-examination and triage.

[0005] In addition, the existing self-service terminals are still obviously insufficient in terms of humanization and intelligence. For example, when doctors are prescribing and examining, patients cannot view relevant information in real time, which to some extent affects patients' satisfaction with the medical environment. At the same time, doctors need to perform a series of operations such as patient identification, diagnosis, and prescription at the workstation, which leads to long queues for triage and reduces the efficiency of medical services. Summary of the invention

[0006] In order to improve the above problems, the present application provides an intelligent pre-examination and triage method and system.

[0007] In a first aspect, an embodiment of the present application provides an intelligent pre-diagnosis and triage method, which is applied to an NLP processing system, and the method includes:

[0008] Acquire an initial pre-check consultation conversation text, perform a first semantic embedding on the initial pre-check consultation conversation text to obtain a consultation topic semantic vector of the initial pre-check consultation conversation text, perform a second semantic embedding on the initial pre-check consultation conversation text to obtain a conversation element semantic vector of the initial pre-check consultation conversation text;

[0009] Integrate the consultation topic semantic vector with the conversation element semantic vector to obtain pre-check consultation integrated semantics;

[0010] Obtaining a target response process semantic vector of the target response process log, performing semantic projection on the consultation subject semantic vector to obtain a first session semantic projection vector, combining the target response process semantic vector with the first session semantic projection vector to obtain a first consultation response interaction vector;

[0011] The pre-inspection consultation integrated semantics and the first consultation response interaction vector are entered into a deep structured semantic network, a guidance report is customized using a semantic attention strategy configured according to the pre-inspection consultation integrated semantics and the first consultation response interaction vector, and a pre-inspection consultation guidance report is output.

[0012] Optionally, the step of integrating the consultation topic semantic vector with the session element semantic vector to obtain pre-check consultation integrated semantics includes:

[0013] Performing semantic projection on the consultation topic semantic vector to obtain a second conversation semantic projection vector;

[0014] Reset the prior training vector, combine the second conversation semantic projection vector with the prior training vector, and obtain a second consultation response interaction vector;

[0015] The second consultation response interaction vector is feature-focused using a semantic attention strategy configured according to the conversation element semantic vector to obtain pre-inspection consultation integrated semantics.

[0016] Optionally, the feature focusing of the second consultation response interaction vector using the semantic focus strategy configured according to the conversation element semantic vector to obtain the pre-check consultation integrated semantics includes:

[0017] The second consultation response interaction vector is feature-focused according to the conversation element semantic vector by using a plurality of semantic attention modules with upstream and downstream relationships to obtain pre-inspection consultation integrated semantics;

[0018] Among them, the second consultation response interaction vector serves as input information of the first semantic attention module, and the conversation element semantic vector serves as input information of each semantic attention module.

[0019] Optionally, the feature focusing of the second consultation response interaction vector according to the conversation element semantic vector by multiple semantic attention modules having upstream and downstream relationships to obtain pre-inspection consultation integrated semantics includes:

[0020] Through multiple semantic attention modules with upstream and downstream relationships;

[0021] For each of the semantic attention modules, the input information of the semantic attention module is locally feature focused to obtain a locally focused semantic vector, the locally focused semantic vector is linked feature focused according to the conversation element semantic vector to obtain a first linked focused semantic vector, and the first linked focused semantic vector is semantically projected and then entered into the next semantic attention module;

[0022] The pre-check consultation integrated semantics is obtained according to the output result of the semantic attention module at the end.

[0023] Optionally, the feature focusing of the second consultation response interaction vector according to the conversation element semantic vector by multiple semantic attention modules having upstream and downstream relationships to obtain pre-inspection consultation integrated semantics includes:

[0024] Determining a confidence level of a first semantic vector corresponding to the second consultation-response interaction vector and a confidence level of a second semantic vector corresponding to the conversation element semantic vector;

[0025] According to the first semantic vector confidence, the second consultation-response interaction vector is enhanced to obtain a first consultation-response interaction enhancement vector, and according to the second semantic vector confidence, the conversation element semantic vector is enhanced to obtain a second consultation-response interaction enhancement vector;

[0026] Combining the first consultation-response interaction reinforcement vector with the second consultation-response interaction reinforcement vector to obtain a third consultation-response interaction vector;

[0027] The third consultation response interaction vector is locally feature focused by using a plurality of semantic attention modules having upstream and downstream relationships, and the pre-inspection consultation integrated semantics is obtained according to the output result of the semantic attention module at the end.

[0028] Optionally, the pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the conversation semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the determination of the first semantic vector confidence corresponding to the second consultation response interaction vector and the second semantic vector confidence corresponding to the session element semantic vector includes:

[0029] Entering the second consultation-answer interaction vector and the conversation element semantic vector into the first long short-term memory network for processing respectively, obtaining the first semantic vector confidence corresponding to the second consultation-answer interaction vector and the second semantic vector confidence corresponding to the conversation element semantic vector;

[0030] Among them, the first long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

[0031] Optionally, the deep structured semantic network includes a deep residual subnet, and the guidance report is customized by using a semantic attention strategy configured according to the pre-check consultation integrated semantics and the first consultation response interaction vector, and outputting a pre-check consultation guidance report, including:

[0032] Dynamically construct a perturbation guidance report, input the perturbation guidance report into the deep residual subnet for semantic error correction, and output a pre-check consultation guidance report;

[0033] In the process of semantic error correction, for any residual connection block in the deep residual subnetwork, the output result of the current residual connection block is linked feature focused according to the pre-check consultation integrated semantics to obtain a second linked focused semantic vector, and the output result of the current residual connection block is linked feature focused according to the first consultation response interaction vector to obtain a third linked focused semantic vector, the second linked focused semantic vector and the third linked focused semantic vector are integrated to obtain a fourth linked focused semantic vector, and the fourth linked focused semantic vector is entered into the next residual connection block.

[0034] Optionally, the pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the session semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the second linkage focused semantic vector and the third linkage focused semantic vector are integrated to obtain a fourth linkage focused semantic vector, including:

[0035] Entering the second linkage focused semantic vector and the third linkage focused semantic vector into a second long short-term memory network for processing respectively, obtaining a third semantic vector confidence corresponding to the second linkage focused semantic vector and a fourth semantic vector confidence corresponding to the third linkage focused semantic vector;

[0036] According to the third semantic vector confidence and the fourth semantic vector confidence, the second linkage focused semantic vector and the third linkage focused semantic vector are enhanced to obtain a fourth linkage focused semantic vector;

[0037] Among them, the second long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

[0038] Optionally, the deep structured semantic network includes a deep residual subnet, and the guidance report is customized by using a semantic attention strategy configured according to the pre-check consultation integrated semantics and the first consultation response interaction vector, and outputting a pre-check consultation guidance report, including:

[0039] Dynamically construct a perturbation guidance report, input the perturbation guidance report into the deep residual subnet for semantic error correction, and output a pre-check consultation guidance report;

[0040] In the process of semantic error correction, for any residual connection block in the deep residual subnetwork, determine the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics, strengthen the first consultation response interaction vector according to the fifth semantic vector confidence to obtain a third consultation response interaction strengthening vector, strengthen the pre-check consultation integrated semantics according to the sixth semantic vector confidence to obtain a fourth consultation response interaction strengthening vector, combine the third consultation response interaction strengthening vector with the fourth consultation response interaction strengthening vector to obtain a fourth consultation response interaction vector, perform linkage feature focusing on the output result of the current residual connection block according to the fourth consultation response interaction vector to obtain a fifth linkage focused semantic vector, and enter the fifth linkage focused semantic vector into the next residual connection block.

[0041] Optionally, the pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the session semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the determination of the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics includes:

[0042] Entering the first consultation-response interaction vector and the pre-check consultation integrated semantics into a third long short-term memory network for processing respectively, and obtaining a fifth semantic vector confidence corresponding to the first consultation-response interaction vector and a sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics;

[0043] Among them, the third long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

[0044] In some independent embodiments, the method further comprises:

[0045] Obtain a pre-examination consultation conversation text example, perform a first semantic embedding on the pre-examination consultation conversation text example, obtain a consultation topic semantic vector example of the pre-examination consultation conversation text example, perform a second semantic embedding on the pre-examination consultation conversation text example, obtain a conversation element semantic vector example of the pre-examination consultation conversation text example;

[0046] Integrate the consultation topic semantic vector example with the conversation element semantic vector example based on the conversation semantic mining core to obtain a pre-inspection consultation integrated semantic example;

[0047] Obtain a response process semantic vector example of a sample response process log, perform semantic projection on the consultation topic semantic vector example based on a first semantic mapping component to obtain a first session semantic projection vector example, combine the response process semantic vector example with the first session semantic projection vector example to obtain a consultation response interaction vector example;

[0048] Entering the pre-check consultation conversation text example, the pre-check consultation integrated semantic example, and the consultation response interaction vector example into a deep structured semantic network, and using a semantic attention strategy configured according to the pre-check consultation integrated semantic example and the consultation response interaction vector example to customize a guidance report for the pre-check consultation conversation text example;

[0049] Lock at least part of the network weights of the deep structured semantic network, and collaboratively debug the session semantic mining core and the first semantic mapping component based on semantic errors identified during the guidance report customization process.

[0050] In some independent embodiments, the deep structured semantic network includes a deep residual subnet, a first feature focusing operator for constructing a semantic attention strategy based on the pre-check consultation integrated semantic example, and a second feature focusing operator for constructing a semantic attention strategy based on the consultation response interaction vector example, the locking of at least part of the network weights of the deep structured semantic network, and the collaborative debugging of the session semantic mining core and the first semantic mapping component based on the semantic errors identified in the guidance report customization process, including:

[0051] Lock the configuration variables of the deep residual subnet and the second feature focusing operator, and collaboratively debug the session semantic mining core, the first semantic mapping component, and the first feature focusing operator based on the semantic errors identified during the guidance report customization process.

[0052] In a second aspect, an embodiment of the present application provides an NLP processing system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the above-mentioned method.

[0053] In a third aspect, an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and the computer program implements the above method when running.

[0054] The embodiment of this application proposes an intelligent pre-examination and triage method based on deep learning and natural language processing technology. This method can intelligently acquire and analyze the initial pre-examination consultation conversation text, accurately capture the consultation topic and conversation elements through dual semantic embedding technology, and provide a solid data foundation for subsequent triage work. Through the method of this application, the outpatient and emergency medical process can be optimized, the diagnosis and treatment efficiency of pre-examination and triage medical care can be improved, and patients can be provided with convenient, efficient, high-quality and safe intelligent medical services. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flow chart of an intelligent pre-examination and triage method provided in an embodiment of the present application.

[0056] Figure 2 A schematic diagram of the structure of an NLP processing system 200 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to better understand the above technical scheme, the technical scheme of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0058] Figure 1 A smart pre-examination and triage method is shown, which is applied to an NLP processing system. The method includes the following steps 110 to 140.

[0059] Step 110, obtain the initial pre-inspection consultation conversation text, perform a first semantic embedding on the initial pre-inspection consultation conversation text to obtain a consultation topic semantic vector of the initial pre-inspection consultation conversation text, perform a second semantic embedding on the initial pre-inspection consultation conversation text to obtain a conversation element semantic vector of the initial pre-inspection consultation conversation text.

[0060] In the embodiment of the present application, the initial pre-examination consultation session text refers to the text description of the patient's symptoms, condition or health problems that is first entered by the patient when using the "smart pre-examination triage machine". This text is the beginning of the patient's interaction with the triage machine, and usually contains the main questions the patient wants to consult and related background information. For example, a patient may enter "I have been coughing, having phlegm, and feeling chest tightness recently. What disease is this?" Such text is the initial pre-examination consultation session text.

[0061] The first semantic embedding (processing) is a natural language processing technique used to convert the text of the initial pre-screening consultation session into a mathematical representation that can be understood by a computer, namely a semantic vector. In this process, the main focus is on the subject content in the text, such as symptoms such as "cough", "sputum" and "chest tightness" in the above example, and possible disease types such as "cold" and "bronchitis", so as to extract the topic of the consultation. This processing helps the machine understand the core issues of the patient's consultation.

[0062] The consultation topic semantic vector is a mathematical representation obtained after the first semantic embedding process, which captures the main consultation topics in the initial pre-screening consultation session text. This vector represents the topic content of the text in a multidimensional space, allowing the machine to judge the topic association between different texts based on the similarity of the vectors. In the above example, the consultation topic semantic vector will reflect that the topic of the patient's consultation is about the symptoms and diagnosis of respiratory diseases.

[0063] The second semantic embedding (processing) is similar to the first semantic embedding, which is also a technique for converting text into semantic vectors. However, the second semantic embedding focuses more on extracting key elements in the conversation, such as the severity and duration of symptoms, the patient's age, gender and other detailed information. This information is crucial for doctors to make accurate diagnoses and treatment recommendations.

[0064] The conversation element semantic vector is a mathematical representation obtained after the second semantic embedding process, which focuses on capturing the key conversation elements in the initial pre-examination consultation conversation text. Different from the consultation topic semantic vector, the conversation element semantic vector focuses more on detailed information, such as the description of symptoms, the patient's personal situation, etc. This information helps doctors understand the patient's condition more comprehensively and make personalized treatment recommendations.

[0065] In step 110, the NLP processing system first receives and obtains the initial pre-examination consultation session text input by the patient. This text is the patient's initial description of his or her own condition or health problem, and contains rich medical information.

[0066] Next, the NLP processing system performs two different semantic embedding processes on this initial pre-examination consultation conversation text. The first is the first semantic embedding, which focuses on analyzing the subject content in the text, such as the main symptoms described by the patient, possible disease types, etc. Through complex algorithms and a large amount of training data, the NLP processing system can accurately extract the topic of the consultation and convert it into a consultation topic semantic vector. This vector represents the subject content of the text in a multidimensional space, providing a basis for subsequent machine understanding and analysis.

[0067] At the same time, the NLP processing system also performs a second semantic embedding process on the text of the initial pre-examination consultation conversation. This step focuses more on extracting key conversation elements in the text, such as the severity of symptoms, the patient's age and gender, the duration of the disease, etc. This information is crucial for doctors to make accurate diagnoses and treatment recommendations. Through the second semantic embedding process, the NLP processing system can convert these key conversation elements into conversation element semantic vectors, providing more comprehensive data support for subsequent medical decisions.

[0068] Step 120: Integrate the consultation topic semantic vector with the session element semantic vector to obtain pre-inspection consultation integrated semantics.

[0069] In an embodiment of the present application, the pre-examination consultation integrated semantics is a comprehensive semantic expression that integrates the information of the consultation topic semantic vector and the conversation element semantic vector. For example, the pre-examination consultation integrated semantics is a comprehensive and in-depth interpretation of the patient's initial pre-examination consultation. It not only covers the main symptoms or health problems consulted by the patient (consultation topics), but also includes details related to these symptoms or problems, such as a specific description of the symptoms, duration, the patient's personal background information, etc. (conversation elements). The formation of this integrated semantics enables a machine or system to understand the patient's consultation content more comprehensively, and then provide more accurate and personalized medical advice and services.

[0070] In step 120, the task performed by the NLP processing system is to integrate the consultation topic semantic vector and the conversation element semantic vector previously obtained through two different semantic embedding processes.

[0071] It is understandable that the above integration process is not a simple merger, but a deep fusion of the information in the two vectors through innovative and complex algorithms. The consultation topic semantic vector provides the core issues of the patient's consultation, that is, the health problems or symptoms they are mainly concerned about. The conversation element semantic vector supplements the specific details about these issues, such as the severity and duration of the symptoms, the patient's medical history, and lifestyle habits.

[0072] During the integration process, the NLP processing system takes into account the correlation and complementarity between the two vectors to ensure that the integrated semantics of the pre-examination consultation can fully and accurately reflect the patient's overall consultation content. This integration not only improves the machine's ability to understand the patient's consultation content, but also provides a richer and more accurate data foundation for subsequent medical decisions.

[0073] The resulting pre-examination consultation integrated semantics is a high-dimensional mathematical representation containing rich medical information, which will serve as an important input for subsequent steps (such as guiding report customization, medical decision support, etc.). In this way, the NLP processing system can respond to patients' consultations more intelligently and provide more personalized and efficient medical services.

[0074] Step 130: obtain the target response process semantic vector of the target response process log, perform semantic projection on the consultation topic semantic vector to obtain a first session semantic projection vector, combine the target response process semantic vector with the first session semantic projection vector to obtain a first consultation response interaction vector.

[0075] In the embodiment of the present application, the target response process log is a log file that records the standard response process of the hospital or clinic for various consultations. It records in detail the standard processing steps, medical advice and service processes for different symptoms or health problems. These logs are formed by medical institutions based on years of practical experience and professional knowledge accumulation, and represent the best response strategies for specific medical problems. By referring to these logs, medical institutions can ensure that every patient can receive unified and high-quality medical services.

[0076] The target response process semantic vector is a mathematical representation that is understandable to computers, namely a semantic vector, which is converted into the target response process log through natural language processing technology. This vector captures the core information of the standard response process of a hospital or clinic for a specific consultation, so that the machine can understand and execute the corresponding medical process based on the vector. This conversion helps to realize the automation and intelligence of the medical process and improve the efficiency and quality of medical services.

[0077] Semantic projection (processing) is a technique that maps the semantic vector of the consultation topic to the semantic vector of the target response process. Through semantic projection, the correlation and similarity between the consultation topic and the target response process can be found, so as to determine the standard response process that best suits the current consultation topic. This processing technology helps to achieve personalized medical services and ensure that each patient can receive targeted medical advice and treatment plans.

[0078] The first conversation semantic projection vector is a mathematical representation obtained by semantically projecting the consultation topic semantic vector. It reflects the correlation and similarity between the consultation topic and the target response process, and provides an important basis for determining the standard response process that best suits the current consultation topic. This vector helps the machine understand the patient's consultation intention more accurately and give corresponding medical advice and treatment plans.

[0079] The first consultation response interaction vector is a mathematical representation obtained by combining the target response process semantic vector with the first conversation semantic projection vector. It integrates the patient's consultation topic and the hospital's standard response process information, providing comprehensive data support for subsequent guidance report customization and medical decision-making. This vector reflects the interaction process between patients and medical institutions, which helps to achieve more efficient and personalized medical services.

[0080] In step 130, the NLP processing system first obtains the target response process log and extracts the target response process semantic vector from it. The vector represents the standard response process of the hospital or clinic for similar consultations and is an important reference for subsequent processing.

[0081] Next, the NLP processing system performs semantic projection processing on the previously obtained consultation topic semantic vector. This process aims to find the correlation and similarity between the consultation topic and the target response process, thereby obtaining the first conversation semantic projection vector. This projection vector reflects the degree of match between the patient's consultation topic and the standard response process of the medical institution.

[0082] Then, the NLP processing system combines the target response process semantic vector with the first conversation semantic projection vector to form the first consultation response interaction vector. This interaction vector integrates the patient's consultation intention and the standard response process information of the medical institution, providing comprehensive data support for subsequent guidance report customization and medical decision-making.

[0083] Through this step, the NLP processing system can more accurately understand the patient's consultation intention and give corresponding medical advice and treatment plans according to the standard response process of the medical institution. This not only improves the efficiency and quality of medical services, but also realizes personalized medical services, ensuring that each patient can receive targeted medical advice and treatment plans.

[0084] Step 140: enter the pre-inspection consultation integrated semantics and the first consultation response interaction vector into a deep structured semantic network, customize the guidance report using a semantic attention strategy configured based on the pre-inspection consultation integrated semantics and the first consultation response interaction vector, and output a pre-inspection consultation guidance report.

[0085] In an embodiment of the present application, a deep structured semantic network is a complex network structure specifically used to process and parse deep semantic information. In the network, different semantic concepts and entities are represented as nodes, and the relationships between these nodes constitute the edges of the network. A deep structured semantic network can not only understand the semantics of a single word or phrase, but also capture the association and mutual influence of these words or phrases in context, so as to more comprehensively understand the semantic content of the entire text. In a medical scenario, such a network can be used to parse and understand the patient's consultation content, as well as the response of a doctor or medical institution.

[0086] The semantic attention strategy configured based on the pre-examination consultation integrated semantics and the first consultation response interaction vector is a customized attention mechanism that determines which parts should be focused on in the deep structured semantic network based on the specific content of the pre-examination consultation integrated semantics and the first consultation response interaction vector. Specifically, the strategy analyzes the key information in the integrated semantics and interaction vectors, such as symptom descriptions, patient concerns, medical institution recommendations, etc., and adjusts the weights and focus points in the network accordingly, so as to more accurately extract and interpret relevant information. In this way, the semantic attention strategy can ensure that the NLP processing system can fully consider the actual needs of patients and the recommendations of medical institutions when generating guidance reports.

[0087] Guided report customization is a process of generating personalized reports based on specific semantic information. In this process, the NLP processing system uses deep structured semantic networks and semantic attention strategies to analyze and interpret the patient's consultation content and the medical institution's response. The system then generates a customized report based on this information, which contains detailed answers to the patient's consultation questions, medical advice, and possible next steps. The report is designed to help patients better understand their health conditions and follow the medical institution's recommendations for appropriate treatment and management.

[0088] The Pre-Examination Consultation Guidance Report is a personalized report based on the patient's pre-examination consultation content and the medical institution's recommendations. It details the patient's symptoms, possible causes, recommended treatment plans, and follow-up precautions. This report is of great guiding significance to patients, helping them better understand their health status and take appropriate actions to improve or manage their condition.

[0089] It can be understood that step 140 is a key link in the entire process, which involves entering the pre-inspection consultation integrated semantics and the first consultation response interaction vector obtained in the previous step into the deep structured semantic network, and using a specific semantic attention strategy to guide the customization of the report.

[0090] First, the NLP processing system takes the pre-inspection consultation integrated semantics and the first consultation response interaction vector as input and enters them into the deep structured semantic network. The network has a strong semantic parsing ability and can deeply understand the connotation and extension of these input information and capture the relationship between the key concepts and entities.

[0091] Next, the system configures the corresponding semantic attention strategy based on the pre-inspection consultation integrated semantics and the content of the first consultation response interaction vector. This strategy will guide the network to focus on the parts related to the consultation topic and medical institution recommendations during the parsing process, ensuring that key information can be accurately extracted and interpreted.

[0092] Then, using the configured semantic attention strategy, the NLP processing system begins to customize the guidance report. Based on the analysis results of the deep structured semantic network, the system combines the patient's actual needs and the medical institution's recommendations to generate a personalized pre-examination consultation guidance report. This report will detail the patient's symptoms, possible causes, treatment options, and subsequent precautions, aiming to help patients better understand their health status and take appropriate actions.

[0093] Finally, the system outputs this customized pre-examination consultation guidance report for reference and use by patients and medical institutions. This report not only provides patients with personalized medical advice and treatment plans, but also provides medical institutions with an efficient and accurate way to respond to patients' inquiries and provide medical services.

[0094] In one possible application scenario, the NLP processing system is used in a "smart pre-examination and triage machine" to optimize the outpatient and emergency medical process and improve the diagnosis and treatment efficiency of pre-examination and triage.

[0095] When a patient comes to the smart pre-examination triage machine for consultation, the NLP processing system first obtains the patient's initial pre-examination consultation conversation text. This text may contain the patient's symptom description, medical history, and their concerns. The NLP processing system performs two different semantic embeddings on this initial pre-examination consultation conversation text. The first semantic embedding focuses on the topic of the consultation, such as whether the patient is consulting about a cold, fever, or other health problems, thereby obtaining the consultation topic semantic vector. The second semantic embedding focuses on extracting key elements in the conversation, such as the severity and duration of the symptoms, thereby obtaining the conversation element semantic vector.

[0096] After obtaining the consultation topic semantic vector and the conversation element semantic vector, the NLP processing system integrates the two vectors to form a more comprehensive pre-examination consultation integrated semantics, which not only reflects the patient's consultation topic, but also includes various key elements in the conversation, providing rich information for the subsequent customization of the guidance report.

[0097] The NLP processing system then obtains the target response process semantic vector of the target response process log, which represents the standard response process of the hospital or clinic for similar consultations. Then, the NLP processing system performs semantic projection on the consultation topic semantic vector to obtain the first conversation semantic projection vector, which reflects the association between the consultation topic and the target response process. Finally, the NLP processing system combines the target response process semantic vector with the first conversation semantic projection vector to obtain the first consultation response interaction vector, which takes into account the patient's consultation topic and conversation elements, and combines the hospital's standard response process, providing an important basis for the subsequent customization of guidance reports.

[0098] After completing the above steps, the NLP processing system enters the pre-examination consultation integrated semantics and the first consultation response interaction vector into the deep structured semantic network, which can perform deep learning and reasoning based on these vectors to generate a guidance report for patient consultation. In the process of generating the report, the NLP processing system configures the corresponding semantic attention strategy based on the pre-examination consultation integrated semantics and the first consultation response interaction vector to ensure that the report can comprehensively and accurately reflect the patient's consultation content and the hospital's response process. Finally, the NLP processing system outputs a pre-examination consultation guidance report, which will serve as an important basis for medical staff to conduct pre-examination and triage of patients, helping medical staff to make diagnosis and treatment decisions quickly and accurately.

[0099] In addition, it is important to note that the embodiments of the present application are mainly concerned with text processing and information extraction, rather than directly diagnosing or treating patients. In detail, it involves obtaining the initial pre-examination consultation conversation text and extracting the consultation topic semantic vector and conversation element semantic vector through two different semantic embedding methods (first semantic embedding and second semantic embedding). These steps are all processing of text data and have nothing to do with directly diagnosing or treating the patient's body.

[0100] Secondly, the further steps in the embodiments of the present application, such as integrating the consultation topic semantic vector with the session element semantic vector, obtaining the semantic vector of the target response process log, performing semantic projection on the consultation topic semantic vector, and combining these vectors, are all processes of data processing and information integration, and also do not involve direct diagnosis or treatment of the patient.

[0101] Finally, the embodiment of the present application uses a deep structured semantic network and a specific semantic attention strategy to customize the guidance report and output a pre-examination consultation guidance report. This process generates a report based on the information extracted and integrated in the previous steps (and the report focuses on the guidance and guidance of the medical process), rather than performing actual medical operations on the patient.

[0102] In summary, the embodiments of the present application are mainly about the processing and information extraction of text data, and the generation of guidance reports based on this information. It does not include direct diagnosis or treatment methods for diseases. Therefore, this technical solution does not belong to disease diagnosis and treatment methods. In addition, please note that although the embodiments of the present application itself does not involve direct diagnosis and treatment of diseases, the generated pre-examination consultation guidance report can be used as a reference for diagnosis and treatment by doctors or other medical professionals later. However, this does not affect the patentability of the embodiments of the present application itself. The invention embodiments focus on the technical solutions themselves at the text level, rather than their subsequent use methods or results.

[0103] By applying the embodiments of the present application, the initial pre-examination consultation conversation text can be intelligently acquired and analyzed, and the consultation topic and conversation elements can be accurately captured through the dual semantic embedding technology, providing a solid data foundation for subsequent triage work. Further, the consultation topic semantic vector and the conversation element semantic vector are efficiently integrated to form pre-examination consultation integrated semantics, which not only enriches the level and details of the information, but also improves the accuracy and pertinence of triage. At the same time, by interacting with the semantic vector of the target response process log, the standard response process suitable for the current consultation situation can be located more quickly, thereby optimizing the efficiency and quality of triage. By entering the pre-examination consultation integrated semantics and the consultation response interaction vector into a deep structured semantic network, and using a specific semantic attention strategy, a personalized pre-examination consultation guidance report can be generated, which not only provides patients with more accurate medical advice, but also provides doctors with a reliable and efficient decision support tool, thereby greatly reducing the waiting time of patients during the triage process and improving the overall medical service experience.

[0104] In summary, the embodiments of the present application realize the intelligence and personalization of medical pre-examination and triage, provide an efficient and accurate communication platform for patients and doctors, and significantly improve the efficiency and quality of medical services.

[0105] It is further noted that, in combination with the above beneficial effects, the embodiment of the present application can intelligently acquire and analyze the initial pre-examination consultation conversation text, and accurately capture the consultation topic and conversation elements through dual semantic embedding technology. This process is mainly the intelligent processing and analysis of text data, and does not involve direct diagnosis or treatment of the disease. It provides a method of information extraction and integration, laying a data foundation for subsequent triage work.

[0106] Secondly, the embodiment of the present application further efficiently integrates the consultation topic semantic vector with the conversation element semantic vector to form pre-examination consultation integrated semantics. This process enriches the level and details of information and improves the accuracy and pertinence of triage. However, this is still the processing and integration of information and does not include actual medical operations or diagnosis of patients.

[0107] Furthermore, by interacting with the semantic vector of the target response process log, the embodiment of the present application can more quickly locate the standard response process suitable for the current consultation situation, thereby optimizing the efficiency and quality of triage. This is also an optimization of information and processes, and does not involve direct medical behavior on patients.

[0108] Finally, the embodiment of the present application generates a personalized pre-examination consultation guidance report through a deep structured semantic network and a specific semantic attention strategy. This report provides medical advice to patients and decision support to doctors. However, these suggestions and support are based on previous data analysis and information processing and do not constitute a direct diagnosis or treatment of patients. In addition, reducing the waiting time of patients in the triage process and improving the medical service experience are all optimizations of the medical service process and do not involve direct diagnosis and treatment of diseases.

[0109] In some optional embodiments, the consultation topic semantic vector is integrated with the session element semantic vector to obtain pre-inspection consultation integrated semantics, including: semantically projecting the consultation topic semantic vector to obtain a second session semantic projection vector; resetting the prior training vector, combining the second session semantic projection vector with the prior training vector to obtain a second consultation response interaction vector; and feature-focusing the second consultation response interaction vector using a semantic attention strategy configured according to the session element semantic vector to obtain pre-inspection consultation integrated semantics.

[0110] In the embodiment of the present application, the system performs a series of sophisticated operations to integrate the consultation topic semantic vector and the conversation element semantic vector, thereby obtaining the pre-inspection consultation integrated semantics. The process first involves semantic projection of the consultation topic semantic vector, which is to map the high-dimensional semantic information of the consultation topic to a new vector space through a specific algorithm model to generate a second conversation semantic projection vector. This step helps to refine and highlight the core meaning of the consultation topic, remove redundant information, and make subsequent processing more accurate and efficient.

[0111] Next, the system resets the prior training vector to eliminate the bias or influence that may be brought by the previous training data, ensuring that each processing is based on the context of the current conversation rather than relying on past experience. Subsequently, the system combines the second conversation semantic projection vector with the reset prior training vector to generate the second consultation response interaction vector. This step integrates the deep meaning and prior knowledge of the conversation topic, laying the foundation for generating richer and more accurate responses.

[0112] In order to further enhance the pertinence and accuracy of semantic integration, the system uses the semantic attention strategy configured according to the semantic vector of the conversation elements to feature-focus the second consultation response interaction vector. This means that the system weights specific features in the interaction vector according to key elements in the conversation, such as symptom descriptions, medical history information, etc., so that these key features occupy a more important position in the subsequent semantic integration. In this way, the system can more accurately capture the core information in the consultation session, thereby generating more appropriate pre-inspection consultation integration semantics.

[0113] Finally, through the above series of refined processing steps, the system successfully integrated the consultation topic semantic vector with the conversation element semantic vector and obtained the pre-inspection consultation integrated semantics. This process not only reflects the system's deep technical accumulation in the field of natural language processing, but also demonstrates its ability to accurately control semantic understanding and response generation in practical applications.

[0114] It can be seen that through innovative semantic projection, vector combination and feature focus technology, a deep understanding and accurate response to consultation conversations have been achieved, which not only greatly improves the accuracy and efficiency of pre-examination and triage, but also provides patients with more personalized and high-quality medical consultation services. At the same time, the application of this technical solution has also effectively reduced the workload of medical staff and promoted the sustainable development of smart medical services.

[0115] In some further optional embodiments, the feature-focusing of the second consultation-response interaction vector using the semantic attention strategy configured according to the session element semantic vector to obtain pre-inspection consultation integrated semantics includes: feature-focusing the second consultation-response interaction vector according to the session element semantic vector through a plurality of semantic attention modules having upstream and downstream relationships to obtain pre-inspection consultation integrated semantics; wherein the second consultation-response interaction vector serves as input information of the first of the semantic attention modules, and the session element semantic vector serves as input information of each of the semantic attention modules.

[0116] In this optional embodiment, the system introduces multiple semantic attention modules with upstream and downstream relationships to achieve feature focus on the second consultation response interaction vector, thereby obtaining pre-inspection consultation integrated semantics. This can more accurately extract and strengthen key information, making the results of pre-inspection consultation more accurate and targeted.

[0117] For example, the system first uses the second consultation response interaction vector as the input information of the first semantic attention module. This vector has integrated the information of the consultation topic and the prior training vector, providing a rich background for subsequent semantic attention. Then, the system uses the conversation element semantic vector as the input information of each semantic attention module. The conversation element semantic vector contains the key elements in the consultation session, such as the patient's symptoms, medical history, etc., which are crucial to determining the results of the pre-examination consultation.

[0118] Under the action of multiple semantic attention modules, the system can gradually feature-focus the second consultation response interaction vector. Each semantic attention module will weight or select specific features in the input vector according to the key information in the conversation element semantic vector, thereby strengthening the expression of these features in the pre-inspection consultation integrated semantics. This gradual focusing process helps the system to more accurately capture the core information in the consultation session and improve the accuracy of the pre-inspection consultation.

[0119] In addition, since these semantic attention modules have upstream and downstream relationships, they can refine the input information in turn. The output of the previous module will be used as the input of the next module, so that each module can further refine and strengthen key information based on the previous module. Through this cascade processing method, the system can gradually deepen its understanding of the consultation conversation and ultimately obtain a more accurate and comprehensive pre-inspection consultation integrated semantics.

[0120] In this way, by introducing multiple semantic attention modules with upstream and downstream relationships, the precise feature focus of the second consultation response interaction vector is achieved, thereby obtaining more accurate and targeted pre-examination consultation integrated semantics. This not only helps to improve the accuracy and efficiency of pre-examination triage, but also provides patients with more personalized and high-quality medical consultation services. At the same time, this technical solution also demonstrates the system's advanced technical application capabilities in the fields of natural language processing and deep learning.

[0121] In detail, by introducing the semantic attention module for feature focusing, the accuracy and pertinence of pre-examination consultation are significantly improved. This refined processing method enables the system to have a deeper understanding of the core content of the consultation session, and accurately weights and selects the features in the second consultation response interaction vector based on the key information of the conversation element semantic vector. This not only optimizes the pre-examination triage process and improves the efficiency of medical services, but also ensures that patients can obtain more accurate and personalized medical advice. Therefore, the embodiments of the present application have shown significant beneficial effects in improving the quality of medical services, improving patients' medical experience, and promoting the development of smart medicine.

[0122] In the next step, the second consultation response interaction vector is feature-focused according to the conversation element semantic vector through multiple semantic attention modules with upstream and downstream relationships to obtain pre-inspection consultation integrated semantics, including: through multiple semantic attention modules with upstream and downstream relationships; for each of the semantic attention modules, the input information of the semantic attention module is locally feature-focused to obtain a locally focused semantic vector; the locally focused semantic vector is linkage-feature-focused according to the conversation element semantic vector to obtain a first linkage-focused semantic vector; the first linkage-focused semantic vector is semantically projected and then entered into the next semantic attention module; the pre-inspection consultation integrated semantics is obtained according to the output result of the semantic attention module at the end.

[0123] In this embodiment, the system performs feature focusing on the second consultation response interaction vector through a series of refined steps and utilizing multiple semantic attention modules with upstream and downstream relationships, thereby obtaining pre-inspection consultation integrated semantics.

[0124] First, the system processes information through multiple semantic attention modules with upstream and downstream relationships. These modules are designed to gradually deepen the understanding and refine the core meaning of the input information.

[0125] For each semantic attention module, the system first focuses the local features of its input information. This step is to highlight the local key features in the input information for subsequent in-depth processing. Through specific algorithms and technical means, the system can identify and strengthen the local features that are crucial to understanding the consultation conversation, thereby obtaining a local focused semantic vector.

[0126] Next, the system performs linkage feature focusing on the local focused semantic vector based on the conversation element semantic vector. The purpose of linkage feature focusing is to further refine and strengthen key information based on local features and combined with the overall elements and context of the conversation. This step ensures that the system can fully and accurately capture the core meaning of the consultation conversation and generate the first linkage focused semantic vector.

[0127] Subsequently, the system performs semantic projection on the first linkage focused semantic vector. Semantic projection is to map the extracted key information into a vector space that is easier to process and understand. The projected vector not only retains the core meaning of the original information, but is also more suitable as the input of the next semantic attention module.

[0128] The projected first linkage focused semantic vector will be entered into the next semantic attention module for further processing and refinement. In this way, through the cascade processing of multiple modules, the system can gradually deepen its understanding of the consultation conversation until a comprehensive and accurate pre-inspection consultation integrated semantics is obtained.

[0129] Finally, the system obtains the final integrated semantics of pre-examination consultation based on the output of the semantic attention module at the end. This result integrates all aspects of key information in the consultation session and provides accurate and comprehensive data support for subsequent pre-examination triage.

[0130] It can be seen that through multiple semantic attention modules with upstream and downstream relationships, the second consultation response interaction vector is finely focused on features, and finally accurate and comprehensive pre-examination consultation integrated semantics are obtained. This process not only demonstrates the system's advanced technology application capabilities in the fields of natural language processing and deep learning, but also provides strong support for improving the accuracy and efficiency of pre-examination triage.

[0131] In detail, through the embodiments of this application, the system can more deeply understand and refine the key information in the consultation session, thereby obtaining more accurate and comprehensive pre-examination consultation integrated semantics. This not only helps to improve the accuracy of pre-examination triage, ensuring that patients can receive timely and appropriate medical advice, but also optimizes the medical service process and improves the efficiency of medical services.

[0132] In other possible embodiments, the feature focusing of the second consultation-response interaction vector according to the conversation element semantic vector is performed through multiple semantic attention modules with upstream and downstream relationships to obtain pre-inspection consultation integrated semantics, including: determining the first semantic vector confidence corresponding to the second consultation-response interaction vector, and the second semantic vector confidence corresponding to the conversation element semantic vector; reinforcing the second consultation-response interaction vector according to the first semantic vector confidence to obtain a first consultation-response interaction reinforcement vector, reinforcing the conversation element semantic vector according to the second semantic vector confidence to obtain a second consultation-response interaction reinforcement vector; combining the first consultation-response interaction reinforcement vector with the second consultation-response interaction reinforcement vector to obtain a third consultation-response interaction vector; locally feature focusing the third consultation-response interaction vector through multiple semantic attention modules with upstream and downstream relationships, and obtaining pre-inspection consultation integrated semantics according to the output result of the semantic attention module at the end.

[0133] For example, through a series of complex processing steps, the system achieves feature focusing on the second consultation response interaction vector, and then obtains the pre-inspection consultation integrated semantics.

[0134] First, the system determines the confidence of the first semantic vector corresponding to the second consultation response interaction vector, and the confidence of the second semantic vector corresponding to the conversation element semantic vector. Here, confidence can be understood as the system's trust in these two vectors or the reliability of the information they contain. This step is to be able to more accurately strengthen key information later.

[0135] Next, the system strengthens the second consultation-response interaction vector according to the confidence of the first semantic vector to obtain the first consultation-response interaction strengthening vector. The strengthening process may involve the adjustment of weights or the highlighting of specific features, with the aim of highlighting more important information in the vector. Similarly, the system will also strengthen the conversation element semantic vector according to the confidence of the second semantic vector to obtain the second consultation-response interaction strengthening vector.

[0136] Then, the system combines the first consultation-response interaction reinforcement vector with the second consultation-response interaction reinforcement vector to form a third consultation-response interaction vector. This combination process may be linear or nonlinear, depending on the system design and the characteristics of the data. The combined vector incorporates the reinforced consultation-response interaction information and conversation element information, providing a richer and more accurate data foundation for subsequent feature focusing.

[0137] Finally, the system focuses the local features of the third consultation response interaction vector through multiple semantic attention modules with upstream and downstream relationships. These modules process the vector in turn, and each module may further refine and strengthen the vector according to its specific focus. After this series of processing, the system obtains the final pre-inspection consultation integrated semantics based on the output results of the semantic attention module at the end.

[0138] In this way, it not only reflects the level of refinement and intelligence of the system in processing natural language, but also shows how to improve the accuracy and effectiveness of pre-inspection consultation integrated semantics through the introduction of confidence and the processing of multi-level semantic attention modules. In detail, by introducing confidence and multi-level semantic attention modules, the embodiment of the present application can more accurately strengthen and refine the key information in the consultation response interaction vector, thereby obtaining more accurate and comprehensive pre-inspection consultation integrated semantics, which not only helps to improve the system's ability to understand the consultation content, but also provides more reliable data support for subsequent medical services.

[0139] Under some exemplary design ideas, the pre-inspection consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the conversation semantic mining core, and the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component. The determination of the first semantic vector confidence corresponding to the second consultation response interaction vector and the second semantic vector confidence corresponding to the session element semantic vector includes: respectively entering the second consultation response interaction vector and the session element semantic vector into the first long short-term memory network for processing, to obtain the first semantic vector confidence corresponding to the second consultation response interaction vector and the second semantic vector confidence corresponding to the session element semantic vector; wherein, the first long short-term memory network is obtained by collaborative debugging with the conversation semantic mining core and the first semantic mapping component.

[0140] Under some exemplary design ideas, the system determines the first semantic vector confidence of the second consultation response interaction vector and the second semantic vector confidence of the conversation element semantic vector in an innovative way, and further uses these confidences to obtain the pre-inspection consultation integrated semantics.

[0141] First, the system uses a component called "Conversational Semantic Mining Core", which can integrate the semantic vector of the consultation topic with the semantic vector of the conversation elements to obtain the pre-inspection consultation integrated semantics. This step is the core of the overall process because it integrates the key information of the consultation topic and the conversation elements, providing a comprehensive semantic basis for subsequent processing.

[0142] Next, to determine the confidence of the two key vectors, the system uses a deep learning model called the first long short-term memory network (LSTM). LSTM is a special type of recurrent neural network (RNN) that can learn and remember long-term dependencies and is well suited for processing sequential data such as natural language text. In this scenario, LSTM is trained to evaluate the importance or confidence of consultation response interactions and conversation elements.

[0143] The system enters the second consultation response interaction vector and the conversation element semantic vector into the first long short-term memory network for processing. The LSTM model analyzes the intrinsic structure and semantic information of these vectors and outputs the confidence of each vector. Specifically, for the second consultation response interaction vector, LSTM will output a first semantic vector confidence, indicating the system's trust in the information contained in the vector; for the conversation element semantic vector, LSTM will also output a second semantic vector confidence.

[0144] It is worth noting that this first long short-term memory network does not exist in isolation. It is obtained in the process of co-debugging with the session semantic mining core and the first semantic mapping component. This means that the training and optimization of LSTM is closely integrated with other key components of the entire system to ensure its performance and accuracy in the entire processing flow.

[0145] Through this collaborative debugging and integrated processing approach, the system can more accurately evaluate the importance of consultation response interactions and conversation elements, and thus generate more accurate pre-inspection consultation integrated semantics. This not only improves the system's ability to understand complex consultation scenarios, but also provides more reliable data support for subsequent medical services. In detail, by introducing the conversation semantics mining core and the collaborative debugging LSTM model, the embodiment of the present application can more accurately evaluate the confidence of the consultation response interaction vector and the conversation element semantic vector, thereby generating more accurate and comprehensive pre-inspection consultation integrated semantics. This method of integration and collaborative processing not only improves the system's ability to understand natural language, but also provides more intelligent and reliable support for medical services.

[0146] In some preferred embodiments, the deep structured semantic network includes a deep residual subnet, and the use of the semantic attention strategy configured according to the pre-check consultation integrated semantics and the first consultation response interaction vector to customize the guidance report and output the pre-check consultation guidance report includes: dynamically constructing a perturbation guidance report, entering the perturbation guidance report into the deep residual subnet for semantic error correction, and outputting the pre-check consultation guidance report; in the process of semantic error correction, for any residual connection block in the deep residual subnet, the output result of the current residual connection block is linked feature focused according to the pre-check consultation integrated semantics to obtain a second linked focused semantic vector, the output result of the current residual connection block is linked feature focused according to the first consultation response interaction vector to obtain a third linked focused semantic vector, the second linked focused semantic vector and the third linked focused semantic vector are integrated to obtain a fourth linked focused semantic vector, and the fourth linked focused semantic vector is entered into the next residual connection block.

[0147] For example, the system's deep structured semantic network innovatively introduces a deep residual subnet. The design of this network structure helps improve the training efficiency and performance of the model, especially when dealing with complex semantic tasks. Next, the system will use the semantic attention strategy configured by the pre-inspection consultation integrated semantics and the first consultation response interaction vector to guide the customization of the report, and finally output the pre-inspection consultation guidance report.

[0148] First, the system dynamically constructs a perturbation guidance report. The purpose of this step is to provide a benchmark in the subsequent semantic error correction process, by introducing small perturbations to test and improve the model's error correction capabilities. The perturbation guidance report can be regarded as an initial version of the guidance report, which may contain some intentionally introduced errors or inconsistencies.

[0149] Next, this perturbation guidance report is fed into the deep residual subnet for semantic error correction. In the deep residual subnet, the information is processed by multiple residual connection blocks. Each residual connection block performs a specific transformation on its input and combines the transformed result with the original input through a skip connection, thereby effectively preserving the original information and alleviating the problem of gradient vanishing.

[0150] In the process of semantic error correction, the system performs linkage feature focusing on the output results of each residual connection block according to the pre-check consultation integrated semantics and the first consultation response interaction vector. Specifically, the system focuses the output results of the current residual connection block according to the pre-check consultation integrated semantics to obtain the second linkage focused semantic vector. This vector highlights the key features related to the pre-check consultation integrated semantics.

[0151] Similarly, the system will also focus the output result of the current residual connection block according to the first consultation-response interaction vector to obtain the third linkage-focused semantic vector. This vector emphasizes the key information related to the consultation-response interaction.

[0152] Subsequently, the system integrates the second linkage focused semantic vector and the third linkage focused semantic vector to obtain the fourth linkage focused semantic vector. This integrated vector incorporates the key information of the pre-inspection consultation integrated semantics and the consultation response interaction, providing richer input for the next residual connection block.

[0153] Finally, the system enters the fourth linkage focused semantic vector into the next residual connection block to continue the semantic error correction process. Through the cascade processing of multiple residual connection blocks, the system can gradually correct the errors in the perturbation guidance report and finally output a pre-inspection consultation guidance report that has been finely corrected.

[0154] By introducing a deep residual subnet and a linkage feature focusing mechanism, the embodiments of the present application can more effectively customize semantic error correction and guidance reports. This method not only improves the accuracy and readability of guidance reports, but also enhances the system's ability to handle complex semantic relationships. At the same time, by dynamically constructing perturbation guidance reports and performing step-by-step error correction in residual connection blocks, the system can better adapt to various semantic scenarios and challenges in practical applications, providing more reliable and intelligent support for pre-inspection consultation.

[0155] In the next possible embodiment, the pre-inspection consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the session semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the second linkage focused semantic vector and the third linkage focused semantic vector are integrated to obtain a fourth linkage focused semantic vector, including: respectively entering the second linkage focused semantic vector and the third linkage focused semantic vector into the second long short-term memory network for processing to obtain the third semantic vector confidence corresponding to the second linkage focused semantic vector, and the fourth semantic vector confidence corresponding to the third linkage focused semantic vector; based on the third semantic vector confidence and the fourth semantic vector confidence, the second linkage focused semantic vector and the third linkage focused semantic vector are strengthened to obtain a fourth linkage focused semantic vector; wherein, the second long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

[0156] In the following possible embodiment, the system further refines the process of how to integrate the second linkage focused semantic vector and the third linkage focused semantic vector to obtain the fourth linkage focused semantic vector.

[0157] First, the pre-inspection consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the conversation element semantic vector through the conversation semantic mining core. This integration process combines the key information in the consultation topic and the conversation, providing rich context for subsequent semantic processing. At the same time, the first conversation semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, which helps to map the consultation topic to a more processable semantic space.

[0158] Next, the system enters the second linkage focused semantic vector and the third linkage focused semantic vector into the second long short-term memory network (LSTM) for processing. LSTM is a deep learning model suitable for sequence data, which can capture long-term dependencies in the sequence. Here, LSTM is used to evaluate the confidence of the two linkage focused semantic vectors, that is, the reliability and importance of the information they each contain.

[0159] Through LSTM processing, the system obtains the confidence of the third semantic vector corresponding to the second linkage focused semantic vector, and the confidence of the fourth semantic vector corresponding to the third linkage focused semantic vector. These confidences reflect the credibility of the information in the vector and the degree of contribution to the final integrated vector.

[0160] Then, the system strengthens the second linkage focused semantic vector and the third linkage focused semantic vector according to the third semantic vector confidence and the fourth semantic vector confidence. The strengthening process may involve the adjustment of weights or the enhancement of specific features to ensure that more reliable information can occupy a greater weight during the integration process.

[0161] Finally, the enhanced second linkage focused semantic vector and the third linkage focused semantic vector are integrated together to form the fourth linkage focused semantic vector. This vector incorporates the key information evaluated and enhanced by LSTM, providing a more accurate and comprehensive data foundation for subsequent processing.

[0162] It is worth noting that the second long short-term memory network in the embodiment of the present application does not exist in isolation. It is debugged in collaboration with the session semantic mining core, the first semantic mapping component, etc. This means that in the design and optimization process of the entire system, the training and adjustment of LSTM is closely integrated with other key components to ensure its performance and accuracy in the entire processing flow.

[0163] In this way, by introducing the second long short-term memory network to evaluate and strengthen the linkage focused semantic vector, the embodiment of the present application can more accurately integrate the second linkage focused semantic vector and the third linkage focused semantic vector, thereby obtaining a more accurate and comprehensive fourth linkage focused semantic vector. This not only improves the system's ability to understand complex consulting scenarios, but also provides more reliable data support for subsequent medical services. At the same time, the performance and accuracy of the entire processing flow are optimized through collaborative debugging, further improving the intelligence level of the system.

[0164] In another preferred embodiment, the deep structured semantic network includes a deep residual subnet, and the guidance report is customized based on the semantic attention strategy configured according to the pre-check consultation integrated semantics and the first consultation response interaction vector, and the pre-check consultation guidance report is output, including: dynamically constructing a perturbation guidance report, entering the perturbation guidance report into the deep residual subnet for semantic error correction, and outputting the pre-check consultation guidance report; in the process of semantic error correction, for any residual connection block in the deep residual subnet, determining the fifth semantic vector confidence corresponding to the first consultation response interaction vector, and the sixth semantic vector corresponding to the pre-check consultation integrated semantics. The confidence of the first consultation-response interaction vector is enhanced according to the fifth semantic vector confidence to obtain a third consultation-response interaction enhancement vector; the pre-inspection consultation integrated semantics is enhanced according to the sixth semantic vector confidence to obtain a fourth consultation-response interaction enhancement vector; the third consultation-response interaction enhancement vector is combined with the fourth consultation-response interaction enhancement vector to obtain a fourth consultation-response interaction vector; the output result of the current residual connection block is linked feature focused according to the fourth consultation-response interaction vector to obtain a fifth linked focused semantic vector; the fifth linked focused semantic vector is entered into the next residual connection block.

[0165] In another preferred embodiment, the system further refines the operational flow of the deep residual subnet in the deep structured semantic network, particularly when configuring the semantic attention strategy using the pre-check consultation integrated semantics and the first consultation response interaction vector.

[0166] First, the system dynamically constructs a perturbation guidance report. This perturbation guidance report is generated by making slight modifications or perturbations to the original guidance report in order to provide a benchmark for verification and optimization in the subsequent semantic error correction process.

[0167] Next, the system enters the perturbation guidance report into the deep residual subnet for semantic error correction. The deep residual subnet consists of multiple residual connection blocks, each of which performs specific processing on the input data and combines the processing results with the original input through jump connections, thereby effectively solving the gradient vanishing problem in deep neural networks and improving the network's training efficiency and performance.

[0168] In the process of semantic error correction, for any residual connection block in the deep residual subnet, the system first determines the confidence of the fifth semantic vector corresponding to the first consultation response interaction vector, and the confidence of the sixth semantic vector corresponding to the pre-check consultation integrated semantics. These confidence values ​​reflect the reliability and importance of the corresponding semantic vectors.

[0169] Subsequently, the system strengthens the first consultation-response interaction vector according to the fifth semantic vector confidence, and obtains the third consultation-response interaction reinforcement vector. Similarly, the system also strengthens the pre-check consultation integrated semantics according to the sixth semantic vector confidence, and obtains the fourth consultation-response interaction reinforcement vector. The reinforcement process may involve weighting or adjusting certain dimensions of the vector to highlight more important features.

[0170] Then, the system combines the third consultation-response interaction reinforcement vector with the fourth consultation-response interaction reinforcement vector to obtain a fourth consultation-response interaction vector. This combination process may involve operations such as vector concatenation and weighted summation, aiming to fuse the two reinforced vectors into a more comprehensive representation.

[0171] Next, the system performs linkage feature focusing on the output of the current residual connection block based on the fourth consultation response interaction vector. This process aims to make targeted adjustments and optimizations to the output of the residual connection block based on the key information of the consultation response interaction and the pre-inspection consultation integrated semantics, thereby obtaining the fifth linkage focused semantic vector.

[0172] Finally, the system enters the fifth linkage focused semantic vector into the next residual connection block to continue the subsequent semantic error correction process. Through the cascade processing of multiple residual connection blocks, the system can gradually correct the errors in the perturbation guidance report and output a pre-inspection consultation guidance report that has been finely corrected.

[0173] By introducing a deep residual subnet and a linkage feature focusing mechanism, and combining the confidence of the semantic vector for enhanced processing, the embodiment of the present application can more effectively customize semantic error correction and guidance reports. This method not only improves the accuracy and readability of the guidance report, but also enhances the system's ability to handle complex semantic relationships. At the same time, by dynamically constructing perturbation guidance reports and performing step-by-step error correction and enhancement processing in the residual connection block, the system can better adapt to various semantic scenarios and challenges in practical applications, providing more reliable and intelligent support for pre-inspection consultation.

[0174] Under some possible design ideas, the pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the session semantic mining core, and the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component. The determination of the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics includes: respectively entering the first consultation response interaction vector and the pre-check consultation integrated semantics into the third long short-term memory network for processing, to obtain the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics; wherein, the third long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

[0175] For another example, the system enters the first consultation response interaction vector and the pre-inspection consultation integrated semantics into the third long short-term memory network (LSTM) for processing. LSTM is a special recurrent neural network (RNN) that can learn long-term dependencies and effectively solve the gradient vanishing or gradient exploding problems that may occur in traditional RNNs when processing long sequences. Here, LSTM is used to evaluate the confidence of semantic vectors, that is, the reliability and importance of the information they contain.

[0176] When the first consultation response interaction vector and the pre-inspection consultation integrated semantics are input into the LSTM, the network generates a confidence value for each vector by analyzing the internal structure and associated information of these vectors. For the first consultation response interaction vector, the LSTM will output a fifth semantic vector confidence, which reflects the importance and accuracy of the vector in the consultation response interaction. Similarly, for the pre-inspection consultation integrated semantics, the LSTM will output a sixth semantic vector confidence, which reflects the credibility and relevance of this integrated semantics in the overall consultation context.

[0177] It is worth noting that the third long short-term memory network in the embodiment of the present application does not exist in isolation. It is debugged in collaboration with the session semantic mining core, the first semantic mapping component, etc. This means that the training and optimization of LSTM is closely integrated with other key components in the system to ensure its performance and accuracy in the entire semantic processing flow.

[0178] In this way, the system can more accurately evaluate the confidence of the first consultation response interaction vector and the pre-inspection consultation integrated semantics, providing a more reliable data basis for subsequent semantic error correction and guidance report customization. In detail, by introducing the third long short-term memory network to evaluate the confidence of the semantic vector, and working in conjunction with the conversation semantic mining core, the first semantic mapping component, etc., this design idea can more accurately measure the importance and reliability of the consultation response interaction vector and the pre-inspection consultation integrated semantics in the overall consultation context. This not only improves the system's ability to understand complex consultation scenarios, but also provides more solid data support for subsequent semantic error correction and guidance report customization. At the same time, the performance and accuracy of the entire processing flow are optimized through collaborative debugging, further enhancing the intelligence and practicality of the system.

[0179] In some independent embodiments, the method further includes: obtaining a pre-examination consultation conversation text example, performing a first semantic embedding on the pre-examination consultation conversation text example, obtaining a consultation topic semantic vector example of the pre-examination consultation conversation text example, performing a second semantic embedding on the pre-examination consultation conversation text example, obtaining a session element semantic vector example of the pre-examination consultation conversation text example; integrating the consultation topic semantic vector example with the session element semantic vector example based on a session semantic mining core to obtain a pre-examination consultation integrated semantic example; obtaining a response process semantic vector example of a sample response process log, performing semantic projection on the consultation topic semantic vector example based on a first semantic mapping component to obtain a first session A semantic projection vector example is provided, and the response process semantic vector example is combined with the first conversation semantic projection vector example to obtain a consultation-response interaction vector example; the pre-check consultation conversation text example, the pre-check consultation integrated semantic example, and the consultation-response interaction vector example are entered into a deep structured semantic network, and a guidance report is customized for the pre-check consultation conversation text example using a semantic attention strategy configured according to the pre-check consultation integrated semantic example and the consultation-response interaction vector example; at least part of the network weights of the deep structured semantic network are locked, and based on the semantic errors identified during the guidance report customization process, the conversation semantic mining core and the first semantic mapping component are collaboratively debugged.

[0180] In this embodiment, the system's approach further covers multiple key steps, aiming to optimize the pre-screening consultation process through deep learning and natural language processing techniques.

[0181] First, the system obtains pre-screening consultation conversation text examples, which are collected from real consultation conversations for training and optimization. Then, the system performs two different types of semantic embedding operations on these conversation text examples.

[0182] The first is the first semantic embedding, through which the system is able to extract examples of consultation topic semantic vectors. These vectors capture the core topics and key information in the consultation conversation and are the basis for subsequent processing.

[0183] The second is the second semantic embedding, which helps the system get examples of conversation element semantic vectors. These vectors focus on specific details and elements in the conversation, such as symptom description, medical history, etc., which are crucial for accurately understanding the patient's condition.

[0184] Next, the system uses the conversation semantic mining core to integrate the consultation topic semantic vector examples with the conversation element semantic vector examples to generate pre-inspection consultation integrated semantic examples. This step is to fuse the topic and detail information together to form a more comprehensive and richer semantic representation.

[0185] Then, the system obtains the response process semantic vector examples of the sample response process log, which reflect the response process and handling methods of the doctor or medical system to the patient consultation. At the same time, based on the first semantic mapping component, the system performs semantic projection on the consultation topic semantic vector example to obtain the first session semantic projection vector example. This projection vector is based on retaining the topic information and converting it into a form that is more suitable for comparison and combination with the response process semantic vector.

[0186] Subsequently, the system combines the response process semantic vector example with the first conversation semantic projection vector example to generate a consultation response interaction vector example. This vector integrates the information of the consultation topic and the doctor's response process, providing an important basis for the subsequent customization of the guidance report.

[0187] The next key step is to input the pre-inspection consultation conversation text examples, pre-inspection consultation integrated semantic examples, and consultation response interaction vector examples into a deep structured semantic network. This network has powerful feature extraction and representation learning capabilities, and can deeply understand and process these complex semantic information.

[0188] In the network, the system uses the semantic attention strategy configured based on the pre-check consultation integrated semantic example and the consultation response interaction vector example to guide the report customization of the pre-check consultation conversation text example. This strategy ensures that the report can accurately focus on the key information and potential problems in the consultation.

[0189] Finally, in order to further improve the accuracy and efficiency of the system, the system locks at least part of the network weights of the deep structured semantic network, and based on the semantic errors identified during the guidance report customization process, the session semantic mining core and the first semantic mapping component are collaboratively debugged. This debugging method can optimize the performance of key components in a targeted manner, making the entire system more intelligent and efficient.

[0190] In this way, through this series of refined processing steps, the system can more accurately understand and process the pre-examination consultation conversation text and generate high-quality guidance reports. At the same time, using the deep structured semantic network and collaborative debugging mechanism, the system can continuously learn and optimize its own performance, so as to better serve the medical pre-examination consultation scenario, which not only improves the efficiency and quality of medical services, but also brings patients a more convenient and accurate medical consultation experience.

[0191] In other independent embodiments, the deep structured semantic network includes a deep residual subnet, a first feature focusing operator for constructing a semantic attention strategy based on the pre-inspection consultation integrated semantic example, and a second feature focusing operator for constructing a semantic attention strategy based on the consultation response interaction vector example, and the locking of at least part of the network weights of the deep structured semantic network, and collaboratively debugging the session semantic mining core and the first semantic mapping component based on semantic errors identified in the guidance report customization process, includes: locking the configuration variables of the deep residual subnet and the second feature focusing operator, and collaboratively debugging the session semantic mining core, the first semantic mapping component and the first feature focusing operator based on semantic errors identified in the guidance report customization process.

[0192] For example, the system's deep structured semantic network is designed to be more complex and sophisticated to adapt to more diverse pre-inspection consultation scenarios. Specifically, this network includes multiple components such as the deep residual subnet, the first feature focusing operator, and the second feature focusing operator.

[0193] The deep residual subnet is an important component of the deep structured semantic network. It can effectively solve the gradient vanishing and representation bottleneck problems of deep neural networks during training, thereby improving the performance of the network. Here, the deep residual subnet is used to extract and process the deep semantic features of the pre-screening consultation conversation text.

[0194] The first feature focusing operator and the second feature focusing operator are used to construct semantic attention strategies based on the pre-inspection consultation integrated semantic example and the consultation response interaction vector example respectively. These operators can make the network pay more attention to the information that has an important impact on the consultation results by weighting the features or selectively strengthening certain features.

[0195] During the collaborative debugging process, the system first locks the configuration variables of the deep residual subnet and the second feature focusing operator, which means that during the debugging process, the internal parameters and structures of these components will remain unchanged. The benefit of this is that it can reduce the complexity of debugging and make the debugging process more focused on the conversation semantic mining core, the first semantic mapping component, and the first feature focusing operator, which are directly related to the understanding and processing of the consultation conversation text.

[0196] Next, the system collaboratively debugs the conversation semantic mining core, the first semantic mapping component, and the first feature focusing operator based on the semantic errors identified during the guidance report customization process. "Semantic errors" here may include misunderstandings, omissions, or misinterpretations of key information in the consultation conversation text. By continuously adjusting and optimizing the parameters and structures of these components, the system can gradually correct these errors and improve its ability to understand and process consultation conversation texts.

[0197] By introducing a deep residual subnet and a feature focus operator, combined with a collaborative debugging method that locks some network weights, this embodiment enables the system to more accurately understand and process key information in the pre-examination consultation conversation text. This not only improves the accuracy and pertinence of the pre-examination consultation guidance report, but also provides more reliable data support for subsequent medical services. At the same time, by optimizing the performance of key components through collaborative debugging, the robustness and generalization ability of the system are further enhanced. Furthermore, Figure 2 Schematic diagram of the structure of an NLP processing system 200 provided in an embodiment of the present application. Figure 2 The NLP processing system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0198] Alternatively, if Figure 2 As shown, the NLP processing system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0199] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0200] Alternatively, if Figure 2 As shown, the NLP processing system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices.

[0201] Optionally, the NLP processing system 200 may implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application, which will not be described in detail here for the sake of brevity.

[0202] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and performed. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, and other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0203] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0204] It should be understood that the above-mentioned memory is exemplary but not restrictive. For example, the memory in the embodiments of the present application may also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is to say, the memory in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.

[0205] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program implements the above method when running.

[0206] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations.

Claims

1. A smart pre-examination and triage method, characterized in that: Applied to an NLP processing system, the method comprises: Acquire an initial pre-check consultation conversation text, perform a first semantic embedding on the initial pre-check consultation conversation text to obtain a consultation topic semantic vector of the initial pre-check consultation conversation text, perform a second semantic embedding on the initial pre-check consultation conversation text to obtain a conversation element semantic vector of the initial pre-check consultation conversation text; Integrate the consultation topic semantic vector with the conversation element semantic vector to obtain pre-check consultation integrated semantics; Obtaining a target response process semantic vector of the target response process log, performing semantic projection on the consultation subject semantic vector to obtain a first session semantic projection vector, combining the target response process semantic vector with the first session semantic projection vector to obtain a first consultation response interaction vector; The pre-inspection consultation integrated semantics and the first consultation response interaction vector are entered into a deep structured semantic network, a guidance report is customized using a semantic attention strategy configured according to the pre-inspection consultation integrated semantics and the first consultation response interaction vector, and a pre-inspection consultation guidance report is output.

2. The intelligent pre-examination and triage method according to claim 1, characterized in that: The step of integrating the consultation topic semantic vector with the session element semantic vector to obtain pre-check consultation integrated semantics includes: Performing semantic projection on the consultation topic semantic vector to obtain a second conversation semantic projection vector; Reset the prior training vector, combine the second conversation semantic projection vector with the prior training vector, and obtain a second consultation response interaction vector; The second consultation response interaction vector is feature-focused using a semantic attention strategy configured according to the conversation element semantic vector to obtain pre-inspection consultation integrated semantics.

3. The intelligent pre-examination and triage method according to claim 2, characterized in that: The feature focusing of the second consultation response interaction vector by using the semantic attention strategy configured according to the conversation element semantic vector to obtain the pre-check consultation integrated semantics includes: The second consultation response interaction vector is feature-focused according to the conversation element semantic vector by using a plurality of semantic attention modules with upstream and downstream relationships to obtain pre-inspection consultation integrated semantics; Among them, the second consultation response interaction vector serves as input information of the first semantic attention module, and the conversation element semantic vector serves as input information of each semantic attention module.

4. The intelligent pre-examination and triage method according to claim 3, characterized in that: The feature focusing of the second consultation response interaction vector according to the conversation element semantic vector by using a plurality of semantic attention modules having upstream and downstream relationships to obtain pre-inspection consultation integrated semantics includes: Through multiple semantic attention modules with upstream and downstream relationships; For each of the semantic attention modules, the input information of the semantic attention module is locally feature focused to obtain a locally focused semantic vector, the locally focused semantic vector is linked feature focused according to the conversation element semantic vector to obtain a first linked focused semantic vector, and the first linked focused semantic vector is semantically projected and then entered into the next semantic attention module; The pre-check consultation integrated semantics is obtained according to the output result of the semantic attention module at the end.

5. The intelligent pre-examination and triage method according to claim 3, characterized in that: The feature focusing of the second consultation response interaction vector according to the conversation element semantic vector by using a plurality of semantic attention modules having upstream and downstream relationships to obtain pre-inspection consultation integrated semantics includes: Determining a confidence level of a first semantic vector corresponding to the second consultation-response interaction vector and a confidence level of a second semantic vector corresponding to the conversation element semantic vector; According to the first semantic vector confidence, the second consultation-response interaction vector is enhanced to obtain a first consultation-response interaction enhancement vector, and according to the second semantic vector confidence, the conversation element semantic vector is enhanced to obtain a second consultation-response interaction enhancement vector; Combining the first consultation-response interaction reinforcement vector with the second consultation-response interaction reinforcement vector to obtain a third consultation-response interaction vector; The third consultation response interaction vector is locally feature focused by using a plurality of semantic attention modules having upstream and downstream relationships, and the pre-inspection consultation integrated semantics is obtained according to the output result of the semantic attention module at the end.

6. The intelligent pre-diagnosis and triage method according to claim 5, characterized in that: The pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the conversation semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the determination of the first semantic vector confidence corresponding to the second consultation response interaction vector and the second semantic vector confidence corresponding to the session element semantic vector includes: Entering the second consultation-answer interaction vector and the conversation element semantic vector into the first long short-term memory network for processing respectively, obtaining the first semantic vector confidence corresponding to the second consultation-answer interaction vector and the second semantic vector confidence corresponding to the conversation element semantic vector; Among them, the first long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

7. The intelligent pre-examination and triage method according to claim 1, characterized in that: The deep structured semantic network includes a deep residual subnet, and the guidance report is customized by using a semantic attention strategy configured according to the pre-check consultation integrated semantics and the first consultation response interaction vector, and outputting a pre-check consultation guidance report, including: Dynamically construct a perturbation guidance report, input the perturbation guidance report into the deep residual subnet for semantic error correction, and output a pre-check consultation guidance report; In the process of semantic error correction, for any residual connection block in the deep residual subnetwork, the output result of the current residual connection block is linked feature focused according to the pre-check consultation integrated semantics to obtain a second linked focused semantic vector, and the output result of the current residual connection block is linked feature focused according to the first consultation response interaction vector to obtain a third linked focused semantic vector, the second linked focused semantic vector and the third linked focused semantic vector are integrated to obtain a fourth linked focused semantic vector, and the fourth linked focused semantic vector is entered into the next residual connection block.

8. The intelligent pre-examination and triage method according to claim 7, characterized in that: The pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the session semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the second linkage focused semantic vector and the third linkage focused semantic vector are integrated to obtain a fourth linkage focused semantic vector, including: Entering the second linkage focused semantic vector and the third linkage focused semantic vector into a second long short-term memory network for processing respectively, obtaining a third semantic vector confidence corresponding to the second linkage focused semantic vector and a fourth semantic vector confidence corresponding to the third linkage focused semantic vector; According to the third semantic vector confidence and the fourth semantic vector confidence, the second linkage focused semantic vector and the third linkage focused semantic vector are enhanced to obtain a fourth linkage focused semantic vector; Among them, the second long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

9. The intelligent pre-examination and triage method according to claim 1, characterized in that: The deep structured semantic network includes a deep residual subnet, and the guidance report is customized by using a semantic attention strategy configured according to the pre-check consultation integrated semantics and the first consultation response interaction vector, and outputting a pre-check consultation guidance report, including: Dynamically construct a perturbation guidance report, input the perturbation guidance report into the deep residual subnet for semantic error correction, and output a pre-check consultation guidance report; In the process of semantic error correction, for any residual connection block in the deep residual subnetwork, determine the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics, strengthen the first consultation response interaction vector according to the fifth semantic vector confidence to obtain a third consultation response interaction reinforcement vector, strengthen the pre-check consultation integrated semantics according to the sixth semantic vector confidence to obtain a fourth consultation response interaction reinforcement vector, combine the third consultation response interaction reinforcement vector with the fourth consultation response interaction reinforcement vector to obtain a fourth consultation response interaction vector, perform linkage feature focusing on the output result of the current residual connection block according to the fourth consultation response interaction vector to obtain a fifth linkage focused semantic vector, and enter the fifth linkage focused semantic vector into the next residual connection block; Among them, the pre-check consultation integrated semantics is obtained by integrating the consultation topic semantic vector with the session element semantic vector based on the session semantic mining core, the first session semantic projection vector is obtained by semantically projecting the consultation topic semantic vector based on the first semantic mapping component, and the determination of the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics includes: respectively entering the first consultation response interaction vector and the pre-check consultation integrated semantics into the third long short-term memory network for processing, to obtain the fifth semantic vector confidence corresponding to the first consultation response interaction vector and the sixth semantic vector confidence corresponding to the pre-check consultation integrated semantics; wherein, the third long short-term memory network is obtained by collaborative debugging with the session semantic mining core and the first semantic mapping component.

10. A NLP processing system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 9.