A decision tree generation method based on large language model and related equipment

Abstract

The present invention provides a decision tree generation method based on a large language model and related equipment, which are applied to the field of data processing technology. The present application obtains a training sample set, target medical document information and target hospital information; processes the target hospital information to generate attribute information of the target hospital; pre-processes the training sample set to generate a training sample set with target feature data, wherein the target feature data is used to characterize the node information and decision path information of the clinical guideline document information processing; obtains a preset large language model that matches the target feature data, wherein the preset large language model is set based on the target hospital information received within a preset time period; trains the preset large language model based on the training data set with the target feature data to generate a target large language model; processes the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a decision tree generation method based on a large language model and related equipment. Background Art

[0002] Since medical application scenarios require high knowledge accuracy, the defects of general medical knowledge graphs in data accuracy have led to limitations in their application. In recent years, many scholars have conducted research on the construction of medical knowledge graphs for single diseases. At present, the medical fields involved in medical knowledge graphs are already very rich. For example, Weng et al. proposed an automatic construction framework for medical knowledge graphs based on semantic analysis, and based on this framework, constructed a hypertension knowledge graph using the medical records of 886 hypertensive patients. For example, Hong et al. proposed a comprehensive longitudinal assessment based on dietary factors related to acute myocardial infarction and fatal coronary heart disease, and found the impact of diet on cardiovascular and cerebrovascular diseases. Xiangxiang Zeng et al. introduced knowledge graphs into the field of drug discovery to impose explicit structures to integrate heterogeneous biomedical data, and conducted research on drug reuse and adverse drug reaction prediction. Scholars have also studied the construction of rehabilitation medical knowledge graphs.

[0003] Existing medical knowledge bases include encyclopedia knowledge bases and question-answer knowledge bases. This type of knowledge base does not structure knowledge and fails to solve the problems of multi-source knowledge fusion and knowledge point association. Although keyword matching queries can be performed during use, the returned results are poorly correlated with the user's query intent, and users still need to spend a lot of time to find the knowledge content they really need, which fails to solve the problems of low efficiency and poor accuracy of knowledge query.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of this application is to provide a decision tree generation method and related equipment based on a large language model, which at least overcomes the problems existing in the prior art to a certain extent, generates a training sample set with target features by collecting training sample sets, medical document information and hospital information, and generates decision tree information for medical document and hospital attribute information. Through automated feature extraction and node correlation analysis, the efficiency and accuracy of intervention decisions are improved.

[0006] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present invention.

[0007] According to one aspect of the present application, a decision tree generation method based on a large language model is provided, comprising: obtaining a training sample set, target medical document information and target hospital information, wherein the target medical document information includes an acute upper respiratory tract infection disease intervention guideline, and the training sample set includes other clinical guideline document information that matches the target medical document information; processing the target hospital information to generate attribute information of the target hospital, wherein the attribute information of the target hospital is used to characterize historical feedback data of the target hospital for the target medical document information; preprocessing the training sample set to generate a training sample set with target feature data, wherein the target feature data is used to characterize node information and decision path information of clinical guideline document information processing; obtaining a preset large language model that matches the target feature data, wherein the preset large language model is set based on the target hospital information received within a preset time period; training the preset large language model based on the training data set with the target feature data to generate a target large language model; processing the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information.

[0008] In one embodiment of the present application, the target hospital information is processed to generate attribute information of the target hospital, including: processing the target hospital information to generate user hospitalization information, wherein the user hospitalization information includes historical hospitalization information of several patients that matches the target medical document information; performing feature extraction processing on the user hospitalization information to generate user historical physiological characteristic information and user historical behavior information; processing the user historical physiological characteristic information and the user historical behavior information to generate historical feedback data; and generating attribute information of the target hospital based on the historical feedback data.

[0009] In one embodiment of the present application, the training sample set is preprocessed to generate a training sample set with target feature data, including: extracting features from the training sample set to generate an original feature library; classifying the original feature library to generate user feature information, wherein the user feature information is used to characterize the physiological characteristics of the same user for different diseases; processing the user feature information to generate a feature mapping table, wherein the feature mapping table is used to characterize different physiological characteristics of different patients for the same disease; and processing the original feature library based on the feature mapping table to generate target feature data.

[0010] In one embodiment of the present application, feature extraction is performed on the training sample set to generate an original feature library, including: obtaining a historical clinical guideline document data set; processing the historical clinical guideline document data set to obtain abnormal node features and abnormal co-occurrence frequencies matching the abnormal node features; processing the abnormal node features and the abnormal co-occurrence frequencies matching the abnormal node features to generate a number of node information; processing the node information to generate an original feature library; the method also includes a calculation formula for calculating the abnormal co-occurrence frequency, the calculation formula is: ;in, is the probability of A and B co-occurring, are the probabilities of A and B appearing alone, respectively.

[0011] In one embodiment of the present application, the preset large language model is trained based on the training data set with the target feature data to generate the target large language model, including: obtaining any number of data features in the training data set with the target feature data; generating adjacent features based on the distance between the any number of data features and other numbers of data features of the same category, wherein the adjacent features include a preset number of the any number of data features; determining a sampling ratio based on the number of each data feature in the training set; determining a sampling ratio based on the sampling ratio; sampling the adjacent features based on the sampling ratio to generate a preset number of sampled features; Based on any data feature and each sampling feature, multiple data groups are generated, wherein each data group contains a preset number of data samples, and at least one data sample includes identification information; the multiple data groups are processed to generate a training set and a verification set, wherein the training set and the verification set both include a number of data features; based on the training set, the preset large language model is trained to generate a trained large language model; based on the verification set, the trained large language model is processed to generate a verification result; if the data sample containing the identification information in the verification result is node information and decision path information, the trained large language model is used as the target large language model.

[0012] In one embodiment of the present application, the target medical document information and the attribute information of the target hospital are processed based on the target large language model to generate decision tree information, including: processing the target medical document information based on the target large language model to generate initial node information and decision information matching the initial node information, wherein the initial node information is used to characterize user's attention points; processing the attribute information of the target hospital based on the target large language model to generate node correlation information, wherein the node correlation information is used to characterize the correlation between different nodes; processing the initial node information based on the node correlation information to generate a decision adjustment factor; processing the initial node information and the decision information matching the initial node information based on the decision adjustment factor to generate target node information and target decision information matching the target node information; generating decision tree information based on the target node information and the target decision information.

[0013] In one embodiment of the present application, the attribute information of the target hospital is processed based on the target large language model to generate node correlation information, including: processing the attribute information of the target hospital based on the target large language model to generate clinical symptom classification information; processing the clinical symptom classification information to generate symptom attribute information of different users, wherein the symptom attribute information of different users is used to characterize the correlation between different users and different symptoms; processing the symptom attribute information of the different users to generate node correlation information.

[0014] Another aspect of the present application is a decision tree generation device based on a large language model, comprising: an acquisition module, used to acquire a training sample set, target medical document information and target hospital information, wherein the target medical document information includes an acute upper respiratory tract infection disease intervention guideline, and the training sample set includes other clinical guideline document information that matches the target medical document information; acquiring a preset large language model that matches the target feature data, wherein the preset large language model is set based on the target hospital information received within a preset time period; a processing module, used to process the target hospital information to generate attribute information of the target hospital, wherein the attribute information of the target hospital is used to characterize historical feedback data of the target hospital for the target medical document information; preprocessing the training sample set to generate a training sample set with target feature data, wherein the target feature data is used to characterize node information and decision path information for clinical guideline document information processing; training the preset large language model based on the training data set with target feature data to generate a target large language model; processing the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information.

[0015] According to another aspect of the present application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-mentioned decision tree generation method based on a large language model by executing the executable instructions.

[0016] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the above-mentioned decision tree generation method based on a large language model is implemented.

[0017] According to another aspect of the present application, a computer program product is provided, including a computer program, wherein when the computer program is executed by a third processor, the decision tree generation method based on the large language model is implemented.

[0018] The application provides a decision tree generation method based on a large language model and related equipment. The server generates a training sample set with target features by collecting training sample sets, medical document information and hospital information, and generates decision tree information for medical document and hospital attribute information. Through automated feature extraction and node correlation analysis, the efficiency and accuracy of intervention decisions are improved, which helps doctors make more accurate clinical decisions.

[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a decision tree generation method based on a large language model provided by an embodiment of the present application is shown;

[0021] Figure 2 A schematic diagram of the structure of a decision tree generation device based on a large language model provided in an embodiment of the present application is shown;

[0022] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown;

[0023] Figure 4 A schematic diagram of a storage medium provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0025] Combine the following Figure 1To describe the decision tree generation method based on a large language model according to an exemplary embodiment of the present application. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this regard. On the contrary, the embodiments of the present application can be applied to any applicable scenario.

[0026] In one implementation, the present application also proposes a decision tree generation method based on a large language model and related equipment. Figure 1 The following schematically shows a flow chart of a decision tree generation method based on a large language model according to an embodiment of the present application. Figure 1 As shown, the method is applied to a server, comprising:

[0027] S101, obtaining a training sample set, target medical document information, and target hospital information.

[0028] In one embodiment, the target medical document information refers to specific medical information used to generate an intervention decision tree. Taking the acute upper respiratory tract infection disease intervention guidelines as an example, these document information are the basis for constructing an intervention decision tree. They contain comprehensive and structured knowledge about a specific disease (in this case, acute upper respiratory tract infection). The narrative hierarchy of the clinical guidelines refers to the way the information in the guidelines is organized. They are graded in a logical order and importance so that doctors and medical professionals can quickly and accurately find the required information. The following is an explanation of these levels: Epidemiology, which provides statistical information such as the incidence, epidemic trends, geographical distribution, and population distribution of the disease. Etiology, which describes the factors that cause the disease, may include genetic, environmental, lifestyle, etc. Pathology, which involves biological changes in the disease, including changes in tissue structure and function. Pathogenesis, which explains how the disease develops, including the initiating factors of the disease, the influencing process, and the results. Clinical manifestations, which list the symptoms and signs of the disease, which are observed by doctors during clinical intervention. Laboratory tests, including various laboratory tests used to intervene in the disease, such as blood tests, microbial cultures, etc. Intervention and differential intervention, which provides methods and standards for confirming the disease, as well as guidance on how to distinguish similar diseases or symptoms. Intervention, detailing intervention options, including drug intervention, surgical intervention, physical intervention, etc. Prevention, providing methods to prevent disease, which may include vaccination, lifestyle changes, public health measures, etc.

[0029] In the process of building an intervention decision tree, information at these levels is used to define decision nodes and paths. Each node represents a clinical decision point, such as whether to perform a certain laboratory test, or to select an intervention plan based on the test results. The decision tree guides doctors through the entire process from collecting patient information to making the final intervention decision through these nodes and paths. The training sample set is a set of other clinical guideline documents that match the target medical document information. They provide additional data and information for training large language models to identify and extract decision points and paths in clinical guidelines. These sample sets help improve the accuracy and reliability of the model, enabling it to better understand and process complex medical information.

[0030] S102: Process the target hospital information to generate attribute information of the target hospital.

[0031] In one implementation, the target hospital information is processed to generate user hospitalization information, wherein the user hospitalization information includes historical hospitalization information of several patients matching the target medical document information. The historical hospitalization records of patients matching the target medical document information (such as the acute upper respiratory tract infection disease intervention guide) are collected, and the user hospitalization information may include the patient's basic information, hospitalization time, intervention results, intervention process, drug use, surgery status, discharge status, etc.

[0032] Perform feature extraction on the user's hospitalization information to generate the user's historical physiological feature information and user's historical behavior information. Feature extraction is a key step in data analysis, especially when building an intervention decision tree, it helps to identify and select information that has a significant impact on model training and prediction performance from a large amount of raw data. The following is an expanded description of the user's historical physiological feature information and user's historical behavior information:

[0033] User historical physiological characteristics information, including but not limited to the following characteristics:

[0034] Body temperature, record the patient's basal body temperature or temperature changes under specific conditions.

[0035] Blood pressure, including measurements of systolic and diastolic pressure, may reflect the condition of the cardiovascular system.

[0036] Heart rate, which records the number of heartbeats per minute, may be related to heart function.

[0037] Respiratory rate, the number of breaths per minute, may indicate a problem with the respiratory system.

[0038] Oxygen saturation, the concentration of oxygen in the blood, is critical in assessing respiratory function.

[0039] Hematology, including blood routine, blood type, coagulation function, etc.

[0040] Biochemistry, involving indicators such as liver and kidney function, blood sugar, blood lipids, and electrolytes.

[0041] Microbiology, such as bacterial culture and virus detection, is used for intervention in infectious diseases.

[0042] Immunology, including immunoglobulin levels, autoantibodies, etc., may be related to immune status.

[0043] Imaging tests, such as X-rays, CT scans, and MRIs, provide visual information about the body's internal structures.

[0044] Pathological examination, tissue biopsy, cytological examination, etc. are used for disease diagnosis and grading.

[0045] Electrocardiogram, which records the heart's electrical activity, is used to intervene in heart diseases such as arrhythmias.

[0046] Pulmonary function tests, which assess how well the lungs can ventilate and breathe.

[0047] User historical behavior information, including but not limited to the following characteristics:

[0048] Intervention compliance, whether the patient takes medication, undergoes follow-up examinations, and implements the intervention plan as prescribed by the doctor.

[0049] Diet, eating habits, nutritional intake, food preferences, etc.

[0050] Exercise, exercise frequency, type, intensity, etc.

[0051] Smoking and drinking, smoking and drinking habits and amounts.

[0052] Past medical history, historical information on the patient’s past illnesses, surgeries, hospitalization interventions, etc.

[0053] Family medical history, whether there are hereditary diseases or specific diseases in family members.

[0054] Social support, the level of support from family, friends and social networks.

[0055] Assessment of psychological conditions such as psychological stress, anxiety, depression, etc.

[0056] Drug and substance use, including the patient's use of prescription medications, over-the-counter medications, herbal supplements, recreational drugs, etc.

[0057] Environmental exposure, occupational exposure, living environment, chemical substances or radiation exposure, etc.

[0058] Health service utilization, the frequency and type of health care services used by patients, such as outpatient, emergency, and specialty care.

[0059] Through these detailed feature extractions, a comprehensive patient information profile can be constructed, which not only helps to understand the patient's health status and disease process, but also provides support for intervention decisions. This information can be input into the large language model to help the model learn how to make the best intervention recommendations based on the patient's specific situation. In addition, these features can also be used to monitor changes in the patient's health status, evaluate the effectiveness of interventions, and predict the development of the disease and the patient's long-term prognosis.

[0060] Process the user's historical physiological characteristics information and user's historical behavior information to generate historical feedback data, summarize the physiological characteristics information (such as laboratory test results, vital signs, imaging examinations) and behavioral information (such as intervention compliance, living habits) collected from different sources and time points to ensure the consistency and accuracy of the data. It may be necessary to clean the data and remove erroneous or inconsistent records. Analyze the patient's physiological characteristics and behavioral information to identify patterns and trends, such as the patient's response to a specific drug or the speed of recovery after a specific intervention, and use statistical methods and machine learning algorithms to process and interpret the data to discover potential associations and predictive factors. Evaluate the intervention effect by comparing the changes in physiological indicators before and after the intervention, such as blood pressure control, blood sugar level, symptom relief, etc., and analyze the patient's response to different intervention programs to determine which interventions are most effective. Identify possible complications during the intervention, such as surgical sequelae, drug side effects, infection, etc., and analyze the relationship between the occurrence of complications and the patient's physiological characteristics and behavioral information. Track the patient's readmission within a certain period of time, which is an important indicator to measure the effectiveness of the intervention and the patient's health status, and analyze the reasons for readmission, whether it is related to the initial intervention, the patient's lifestyle or compliance.

[0061] Generate the attribute information of the target hospital based on historical feedback data. The attribute information of the target hospital is used to characterize the historical feedback data of the target hospital for the target medical document information. Calculate the ratio of successful cases of specific interventions or surgeries to total cases to evaluate the effectiveness of the hospital in intervening in specific diseases, and analyze the success rates of different intervention methods to determine best practices and room for improvement. Collect patient satisfaction data on hospital services through questionnaires, interviews, or online evaluation systems to evaluate patient satisfaction with doctor communication, quality of care, hospital facilities, waiting times, etc. Track and analyze the use of hospital resources (such as beds, equipment, and drugs) to evaluate the rationality and efficiency of resource allocation, identify resource waste or shortages by comparing resource usage in different time periods, and propose improvement measures.

[0062] Use medical quality indicators (such as infection rate, complication rate, mortality rate) to evaluate the quality of medical services provided by the hospital, and determine the hospital's position in medical quality by comparing with other hospitals or national / international standards. Evaluate the balance between cost and effect of the interventions provided by the hospital to determine the most cost-effective intervention method, analyze the long-term and short-term economic impact of different intervention plans, and provide a basis for hospital financial planning. Evaluate the performance of the hospital in providing continuous medical services, including the coordination of outpatient, inpatient, rehabilitation and follow-up services, and analyze whether the patient's transition between different service links is smooth and whether there is a service gap. Monitor and record medical errors, accidents and adverse events to evaluate the hospital's performance in patient safety, implement preventive measures and improvement strategies to reduce medical errors and improve patient safety. Track patients' clinical outcomes, including intervention effects, recurrence rates, and survival rates, and use these data to evaluate the long-term effects of the hospital in specific disease interventions. Through this attribute information, hospital managers can better understand the hospital's strengths and weaknesses in specific disease interventions, so as to develop improvement strategies and improve the quality of medical services. At the same time, this information can also provide a reference for patients to choose hospitals and doctors, and provide decision support for medical insurance agencies and public health policy makers.

[0063] Throughout the process, the goal is to create an exhaustive dataset that contains not only the patient's clinical information but also hospital-level feedback data. These data are critical for training large language models because they provide real-world data on intervention effects and patient responses in actual clinical settings. Such data can help the model better understand and predict the best intervention path for a specific disease in a specific hospital setting.

[0064] S103, preprocessing the training sample set to generate a training sample set with target feature data.

[0065] In one implementation, feature extraction is performed on the training sample set to generate an original feature library, and the original feature library is classified and processed to generate user feature information, wherein the user feature information is used to characterize the physiological characteristics of the same user for different diseases. The user feature information is processed to generate a feature mapping table, wherein the feature mapping table is used to characterize different physiological characteristics of different patients for the same disease, and the original feature library is processed based on the feature mapping table to generate target feature data, and the target feature data is used to characterize the node information and decision path information of the clinical guideline document information processing.

[0066] Specifically, all data that may be relevant to intervention decisions are extracted from the training sample set, including the patient's physiological measurements, laboratory test results, clinical observation records, etc. The original feature library is a data set containing a wide range of information, which provides a basis for subsequent feature selection and data processing. The data in the original feature library is classified according to different diseases to identify and distinguish the features related to different diseases. User feature information refers to a set of features related to a specific user, which can characterize the user's physiological characteristics when facing different diseases.

[0067] A feature map is a data structure that links the characteristics of different patients with specific diseases to show the different physiological characteristics of different patients with the same disease. This map helps to understand the different responses of different patients to the same disease and how to adjust the intervention plan based on these differences. For example, some people get much better after taking 1-2 doses of medicine, while others may suffer from recurrence of other diseases due to physical factors. Use feature maps to filter and reorganize the data in the original feature library to generate more accurate and targeted target feature data. This process may involve data conversion, normalization, dimensionality reduction and other processing to ensure the quality and availability of feature data. Target feature data is a processed and optimized feature set that is specifically used to characterize the node information and decision path information of clinical guideline document information processing. These data can be directly used to train large language models so that they can understand and generate intervention decision trees that are consistent with the actual clinical decision-making process.

[0068] Throughout the process, the goal is to transform raw, potentially very complex medical data into a set of clear, organized features that accurately reflect the patient's clinical condition and intervention needs. Such feature data are critical for training large language models because they directly affect the model's predictive accuracy and reliability in clinical applications. In this way, it can be ensured that the model can take into account the individual differences of patients and the complexity of the disease when dealing with specific intervention problems.

[0069] In another implementation, feature extraction is performed on the training sample set to generate an original feature library, which specifically includes: obtaining a historical clinical guideline document data set, collecting historical clinical guideline documents, which may include disease definitions, classifications, intervention standards, intervention recommendations, etc. The data set may come from medical journals, professional databases, public health agencies or medical institutions.

[0070] The historical clinical guideline document dataset is processed to obtain abnormal node features and abnormal co-occurrence frequencies that match the abnormal node features. The collected documents are digitized. If the documents are in paper form, they may need to be scanned and converted into electronic text using optical character recognition (OCR) technology. The data is cleaned and irrelevant information such as headers, footers, and advertisements are removed to ensure the accuracy and availability of the data. The document content is analyzed to identify abnormal nodes related to disease interventions and interventions. These nodes represent key features of the disease or key steps of the intervention. Abnormal node features include symptoms, signs, laboratory test results, intervention responses, etc. Determine the frequency with which abnormal node features co-occur in the document. This helps identify which features often appear together and may indicate specific disease patterns or intervention combinations. The co-occurrence frequency is detailed below and will not be repeated here.

[0071] The abnormal node features and the abnormal co-occurrence frequencies that match the abnormal node features are processed to generate several node information, and the node information is processed to generate the original feature library. In the process of constructing the intervention decision tree, the node information is the basis for understanding the relationship between clinical guidelines and patient data. The node information generated by combining the abnormal node features and their co-occurrence frequencies is a key component of this process. Each abnormal node feature should have a clear definition or description of its specific meaning in the clinical guidelines, such as a symptom, test result, or intervention recommendation. The description may include the measurement method of the feature, the normal value range, and its role in disease intervention or intervention. The correlation in the node information refers to the statistical association between different features. For example, the co-occurrence of certain symptoms may indicate a specific disease. The correlation can be determined by various statistical tests, such as the Pearson correlation coefficient, the Spearman rank correlation coefficient, or the chi-square test. Importance scoring is a method to evaluate the influence of node features in intervention decisions. Some features may receive higher scores because of their significant impact on the intervention or intervention results. The score can be determined by machine learning algorithms (such as feature importance evaluation in random forests) or the logic of expert systems. Node information should also include the context in which the feature appears, such as the frequency of occurrence of a specific symptom in a specific age group or a specific population. Contextual information helps to understand the clinical significance of the feature in different situations. For some diseases, the appearance or change of symptoms may change over time. The node information should be able to reflect this time dependency. Time series data can be used to analyze the progression of the disease or the effect of intervention. Node information can include the level of evidence supporting the description of each feature, such as randomized controlled trials and cohort studies. There may be interactions between certain features, that is, the effect of one feature may depend on the presence or level of another feature. The node information should be able to capture and describe these interactions so that they can be correctly represented in the decision tree.

[0072] Specifically, examples of abnormal node features are as follows: Fever: body temperature exceeding the normal range (e.g., oral temperature > 37.5°C). Cough: dry cough or cough with sputum, duration, frequency, and nature. White blood cell count: the number of white blood cells in a blood test, which may be elevated in infection.

[0073] Examples of the relevance of node information are as follows: Fever and cough: In patients with URTI, fever and cough co-occur frequently, indicating that they may have a strong correlation in intervention. Fever: abnormally high body temperature, which may be a sign of infection, cough: a mechanism for the respiratory tract to clear secretions, which may indicate an upper respiratory tract infection. Fever and white blood cell count: a higher proportion of patients with fever had elevated white blood cell counts, indicating that there may be a positive correlation between them, white blood cell count: an indicator of immune system activity, which is usually elevated during infection.

[0074] In addition, regarding the importance score: fever and cough may be key symptoms in URTI intervention and therefore may receive a higher importance score, while white blood cell count, while helpful for intervention, may not be as immediate as symptoms and therefore may receive a slightly lower score. In addition, in children, fever may be an early and prominent symptom of URTI, while in adults, cough may be a more common presenting symptom. Fever may occur early in the infection, while cough may last longer later in the infection. If a patient has both fever and cough but a normal white blood cell count, this may indicate a viral infection rather than a bacterial infection.

[0075] In addition, in clinical guidelines, intervention rules are usually described in natural language text. For some complex intervention scenarios that need to consider multiple situations, this description method is usually complicated and it is not easy to sort out the decision logic. Taking the blood lipid intervention method as an example, it is necessary to formulate different intervention plans according to the patient's risk stratification, such as "for patients with triglycerides <5.7mmol / L, or no triglyceride value, and LDL-C not meeting the standard", it is necessary to decide which drug to use and the intensity of drug use based on whether the patient has "applied lipid-lowering drugs", such as "moderate-intensity statins", "statins + fibrates + ezetimibe", etc. This type of key medical information on patient characteristics, medication conditions, and medication methods is the key node of intervention decision-making; the ordered connection from one node to another is the decision path, such as "for patients with triglycerides <5.7mmol / L, or no triglyceride value, and LDL-C not meeting the standard", if "no lipid-lowering drugs are used", "moderate-intensity statins" should be used for intervention.

[0076] Through such analysis, medical professionals can better understand how different symptoms and test results jointly influence intervention decisions. In the decision tree, this node information will be used to guide each step from initial intervention to intervention selection. For example, if a patient presents with a high fever and persistent cough, but the white blood cell count is normal, the decision tree may point to the branch of antiviral intervention. On the contrary, if the white blood cell count is elevated, it may point to the branch of antibiotic intervention. This evidence-based decision-making process helps to improve the targeting and effectiveness of interventions.

[0077] Integrate all node information into a database or data structure to form an original feature library, which is the basis for subsequent feature engineering and model training. The original feature library may include a large number of features and requires further screening and optimization to improve the performance of the model.

[0078] Throughout the process, the goal is to transform raw, potentially very complex medical data into a set of clear, organized features that accurately reflect the patient's clinical condition and intervention needs. Such feature data are critical for training large language models because they directly affect the model's predictive accuracy and reliability in clinical applications. In this way, it can be ensured that the model can take into account the individual differences of patients and the complexity of the disease when dealing with specific intervention problems.

[0079] In another implementation, the present application also includes a calculation formula for calculating the abnormal co-occurrence frequency, and the calculation formula is: ;in, is the probability of A and B co-occurring, are the probabilities of A and B appearing alone, respectively.

[0080] The purpose of this formula is to identify and quantify whether there is a significant co-occurrence relationship between different features (such as symptoms, test results, etc.) in clinical guideline documents. This co-occurrence relationship may indicate the importance and relevance of these features in specific disease interventions. By calculating PMI, it can help extract meaningful information from large amounts of medical data, improve the accuracy of interventions and interventions, analyze clinical text data, and discover potential connections between diseases, symptoms, and interventions. This information can be used to build an intervention decision tree and optimize the clinical decision-making process.

[0081] S104: Acquire a preset large language model that matches the target feature data.

[0082] In one embodiment, the preset large language model refers to a model pre-trained in a specific field (such as healthcare) that can understand and generate natural language text. Standard hospital information refers to data collected from a specific hospital, which may include patient medical records, intervention records, laboratory test results, etc. The preset large language model can be customized or fine-tuned based on this information to better adapt to the data characteristics and intervention practices of a specific hospital. The preset time period refers to data collected within a specific time period, which can be several months, a year, or longer. The purpose of selecting this time period is to capture the latest medical practices and trends and ensure the timeliness and relevance of the model.

[0083] During the implementation process, the target feature data needs to be matched with the preset large language model to ensure that the model can understand and process the data. The matching process may involve data preprocessing, such as text cleaning, word segmentation, and removal of stop words, to ensure that the data format is suitable for the input requirements of the model. When the model matches the data, it can be used to analyze and interpret medical texts and extract key information, such as disease intervention criteria, intervention plans, prognostic factors, etc. In this way, the preset large language model can become a powerful tool for medical data analysis and decision support, helping medical professionals improve the quality and efficiency of interventions.

[0084] S105: Training the preset large language model based on the training data set with the target feature data to generate a target large language model.

[0085] In one embodiment, any number of data features in a training data set with target feature data is obtained, and adjacent features are generated based on the distance between any number of data features and other numbers of data features of the same category, wherein the adjacent features include a preset number of any number of data features. Based on the number of each data feature in the training set, a sampling ratio is determined, and based on the sampling ratio, a sampling ratio is determined. Adjacent features are sampled based on the sampling ratio to generate a preset number of sampling features. Based on any data feature and each sampling feature, multiple data groups are generated, wherein each data group contains a preset number of data samples, and at least one data sample includes identification information. The multiple data groups are processed to generate a training set and a verification set, wherein the training set and the verification set both include a number of data features, a preset large language model is trained based on the training set, and a trained large language model is generated, and the trained large language model is processed based on the verification set to generate a verification result, and if the data sample containing the identification information in the verification result is node information and decision path information, the trained large language model is used as the target large language model.

[0086] Specifically, data features are extracted from the training data set. These features may include the patient's physiological indicators, intervention responses, historical hospitalization information, etc. Data features are the original input information used to train the model, and they represent different aspects of the patient's health status. Based on the similarity or distance between data features (such as Euclidean distance or cosine similarity), adjacent features are identified and generated. Adjacent features refer to features that are close to each other in the feature space, and they may jointly affect intervention decisions. According to the distribution and number of data features, the sampling ratio is determined to ensure that the data in the training set and validation set are representative. The sampling ratio refers to the ratio of samples selected from adjacent features, which is used to balance the data set and prevent overfitting. According to the sampling ratio, a certain number of samples are randomly selected from adjacent features to generate sampling features. The sampling features are used to construct the training set and validation set, which are the basis for model training and evaluation.

[0087] Combine each data feature with the sampling feature to generate multiple data sets, each of which contains a certain number of data samples. Each data set may contain identification information, such as patient ID or case number, for tracking and verification. Separate training sets and validation sets from multiple data sets for model training and evaluation. The training set is used to train the large language model, while the validation set is used to evaluate the performance and generalization ability of the model. Use the training set to train the preset large language model, adjust the model parameters to minimize the prediction error, and the training process may include multiple cycles (epochs). In each cycle, the model learns how to better extract information from the data features. Use the validation set to evaluate the trained large language model and generate validation results. The validation results are used to determine whether the model can accurately predict or classify new data samples. If the validation results meet specific criteria, such as accuracy, recall, or F1 score, and can correctly identify node information and decision path information, the model is selected as the target large language model, which is the model ultimately used for actual intervention decision support.

[0088] Through this process, it can be ensured that the large language model has a high degree of accuracy and reliability in specific medical fields. The model can be used to assist doctors in making key decisions such as diagnosis, intervention selection, and prediction of disease development.

[0089] S106: Process the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information.

[0090] In one embodiment, the target medical document information is processed based on the target large language model to generate initial node information and decision information matching the initial node information, wherein the initial node information is used to characterize the user's attention. The target large language model is used to analyze the target medical document information, such as the intervention guide for acute upper respiratory tract infection diseases, to extract key information from the document and generate initial node information, which may include the diagnostic criteria, intervention principles, preventive measures, etc. of the disease. Decision information matching the initial node information is created to guide clinical decisions, such as the selection of intervention plans, the determination of drug dosages, etc. The initial node information and decision information together constitute the basic structure of the decision tree.

[0091] The target hospital's attribute information is processed based on the target large language model, and the node correlation information is used to characterize the correlation between different nodes. The target hospital's attribute information, such as the hospital's historical feedback data, intervention success rate, patient satisfaction, etc., is analyzed using the same large language model, and the node correlation information is extracted, which indicates the connection between different medical decision points, such as the relationship between specific symptoms and intervention plans.

[0092] The initial node information is processed based on the node correlation information to generate a decision adjustment factor. In the process of constructing an intervention decision tree, node correlation information is a key link, which helps determine the relationship and interaction between different decision nodes. The following is a detailed description of how to use node correlation information to generate decision adjustment factors: Risk assessment adjustment factors are based on the patient's clinical data and historical data to assess the patient's likelihood of developing a certain disease or developing a certain complication. For example, if a patient has high-risk factors such as hypertension or diabetes, the decision tree may recommend more frequent monitoring or more aggressive intervention plans.

[0093] Consider the patient's personal preferences for intervention options, such as resistance to surgery, preference for medication, etc. This information can be reflected as different branches in the decision tree, allowing doctors to adjust the intervention plan according to the patient's choice. Consider the resource limitations of the hospital or clinic, such as drug inventory, equipment availability, professional staffing, etc. If certain resources are limited, the decision tree may recommend the use of alternative intervention plans or referral to other medical institutions. Collect data related to node associations, including patients' medical records, hospital resource usage, patient feedback, etc., and analyze the data to determine the association between different nodes, such as the association between certain symptoms and specific intervention plans. Based on the analysis results, generate decision adjustment factors that can quantify the impact of risk, preference, and resource limitations.

[0094] Based on the decision adjustment factor, the initial node information and the decision information matching the initial node information are processed to generate the target node information and the target decision information matching the target node information, and the decision tree information is generated based on the target node information and the target decision information. The initial node information usually contains the basic characteristics of the disease and possible intervention options. The decision adjustment factor is applied to these initial nodes to consider the influence of risk assessment, patient preference, resource availability and other factors. The target node information is adjusted node information that more accurately reflects the decision-making needs in a specific clinical scenario. For example, if a patient has a specific risk of complications, the target node information may include additional monitoring or preventive measures.

[0095] The decision information that matches the initial node information may include recommended intervention options, examination sequences, or patient education materials. After the decision adjustment factors are applied, these decision information will be updated to ensure that they are consistent with the target node information and adapted to the patient's specific situation. The target decision information is the final, personalized decision guidance, which takes into account all relevant adjustment factors. This information provides doctors with a clear guide to action to help them make the most appropriate clinical decision. Based on the target node information and target decision information, a complete decision tree is constructed. The decision tree graphically shows the entire process from initial symptoms to the final intervention decision. Each node represents a step in the decision-making process, and each branch of the decision tree represents a different clinical pathway. Based on the patient's specific situation and the doctor's judgment, the leaf nodes of the tree usually represent the final decision, such as intervention selection, further examination, or observation and waiting. Before actual application, the decision tree needs to be validated to ensure that it accurately reflects clinical guidelines and best practices. Validation may include comparison with expert clinical judgment or testing in an actual clinical setting.

[0096] As new clinical data and medical evidence emerge, decision trees need to be updated regularly, and the updating process may involve re-evaluating decision adjustment factors, as well as adjusting target nodes and target decision information. Through this process, decision trees become a dynamic and adaptable tool that can help medical professionals make the best decisions in an ever-changing clinical environment. This data-driven approach improves the efficiency and quality of interventions, helps achieve personalized medicine and improves patient outcomes.

[0097] Decision adjustment factors are used to fine-tune the path of the decision tree to ensure that each branch reflects the latest clinical guidelines, patient preferences, and resource conditions. Fine-tuning may involve adding new decision nodes, modifying existing decision paths, or adjusting the priority of decision nodes. Before actual application, the adjusted decision tree is verified and tested to ensure that it can provide accurate decision support. In this way, the decision tree is not only based on medical knowledge and clinical data, but also takes into account the personalized needs of patients and the actual conditions of the medical environment. This helps to improve the practicality of the decision tree and the quality of clinical decisions, ultimately improving the intervention effect and satisfaction of patients, improving the efficiency and quality of decision-making, and ensuring that patients receive personalized interventions based on the latest medical evidence and hospital practices.

[0098] In another embodiment, the attribute information of the target hospital is processed based on the target large language model to generate node correlation information, including: processing the attribute information of the target hospital based on the target large language model to generate clinical symptom classification information, processing the clinical symptom classification information to generate symptom attribute information of different users, wherein the symptom attribute information of different users is used to characterize the correlation of different users with different symptoms, and processing the symptom attribute information of different users to generate node correlation information. Specifically, the target large language model is used to analyze the attribute information of the target hospital. This information may include the hospital's historical intervention data, patient feedback, medical resource usage, etc. Through model analysis, clinical symptoms are classified to identify the characteristics of different diseases or conditions. For example, the model may identify a typical set of symptoms associated with a specific disease, such as the association between fever, cough and respiratory infection.

[0099] For each classified symptom, the model generates attribute information related to different patients. This information may include the patient's description of a specific symptom, the severity of the symptom, the time of symptom onset, etc. The symptom attribute information of different users is used to characterize the correlation between different users for different symptoms. For example, the model may find that certain symptoms often occur together, or that the appearance of certain symptoms indicates the development of other symptoms. The symptom attribute information of different users is further processed to generate node correlation information, which describes the relationship between different nodes in the decision tree, such as a symptom node may lead to a specific diagnosis or intervention node. Node correlation information is used to construct or optimize the intervention decision tree to ensure that the structure of the tree can reflect the real-world clinical decision-making process. In the decision tree, each node may be connected to other nodes based on correlation information to form a decision path. Using node correlation information, each branch of the decision tree can be refined to ensure that they can adapt to the specific circumstances of different patients. For example, if a symptom is found to have a strong correlation with the effect of a specific intervention plan, this information can be used to adjust the intervention recommendations in the decision tree.

[0100] As new data is continuously input, the target large language model can continuously learn and optimize to improve the accuracy of node association information. The iterative training of the model helps to improve the predictive ability and clinical applicability of the intervention decision tree. Through this process, the target large language model can not only process and analyze a large amount of medical data, but also understand and predict the complex relationship between clinical symptoms, thereby providing doctors with more precise and personalized intervention recommendations.

[0101] In this application, the server collects training sample sets, target medical document information (such as acute upper respiratory tract infection intervention guidelines) and target hospital information, which include historical hospitalization records, patient physiological and behavioral characteristics, and historical feedback data from the hospital. By preprocessing these data, a training sample set with target features is generated, which are used to characterize the node information and decision paths in the clinical guidelines. A suitable preset large language model is selected, which is set based on the information received by the target hospital within a preset time period. The model is trained with a training data set to generate a target large language model that can understand and process complex medical text data, extract key information, and generate initial node information and decision information.

[0102] In addition, the present application also includes processing the attribute information of the target hospital and generating node relevance information. This includes the classification of clinical symptoms, symptom attribute information of different users, and node relevance information generated based on this information. This information helps to characterize the relevance between different nodes and provides an important basis for the construction of a decision tree. Finally, the target large language model is used to process the target medical document information and hospital attribute information to generate decision tree information. The decision tree information includes target node information and target decision information, which together constitute the structure of the decision tree to guide doctors in clinical decision-making. Through automated feature extraction, node relevance analysis and decision tree generation, the efficiency and accuracy of intervention decisions are improved.

[0103] In one embodiment, if Figure 2 As shown, the present application also provides a decision tree generation device based on a large language model, comprising:

[0104] The acquisition module 201 is used to acquire a training sample set, target medical document information and target hospital information, wherein the target medical document information includes an acute upper respiratory tract infection disease intervention guide, and the training sample set includes other clinical guideline document information that matches the target medical document information; acquire a preset large language model that matches the target feature data, wherein the preset large language model is set based on the target hospital information received within a preset time period;

[0105] The processing module 202 is used to process the target hospital information and generate attribute information of the target hospital, wherein the attribute information of the target hospital is used to characterize the historical feedback data of the target hospital for the target medical document information; pre-process the training sample set to generate a training sample set with target feature data, wherein the target feature data is used to characterize the node information and decision path information of the clinical guideline document information processing; train the preset large language model based on the training data set with the target feature data to generate a target large language model; process the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information.

[0106] The present application embodiment provides an electronic device, such as Figure 3 As shown, the electronic device 3 includes a first processor 300, a memory 301, a bus 302 and a communication interface 303, wherein the first processor 300, the communication interface 303 and the memory 301 are connected via the bus 302; the memory 301 stores a computer program that can be run on the first processor 300, and when the first processor 300 runs the computer program, the decision tree generation method based on the large language model provided in any of the aforementioned embodiments of the present application is executed.

[0107] The memory 301 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 303 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0108] The bus 302 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 301 is used to store a program, and the first processor 300 executes the program after receiving an execution instruction. The decision tree generation method based on a large language model disclosed in any implementation of the embodiment of the present application may be applied to the first processor 300, or implemented by the first processor 300.

[0109] The first processor 300 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the first processor 300. The above-mentioned first processor 300 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a readily available 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-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be embodied as a hardware decoding processor to execute, or a combination of hardware and software modules in the decoding processor to execute. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 301, and the first processor 300 reads the information in the memory 301 and completes the steps of the above method in combination with its hardware.

[0110] The electronic device provided by the above-mentioned embodiments of the present application and the decision tree generation method based on a large language model provided by the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0111] The present application embodiment provides a computer-readable storage medium, such as Figure 4 As shown, the computer-readable storage medium 401 stores a computer program, and when the computer program is read and executed by the second processor 402, the decision tree generation method based on the large language model as described above is implemented.

[0112] The technical solution of the embodiment of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling an electronic device (which may be an air conditioner, a refrigeration device, a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in the embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0113] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the decision tree generation method based on a large language model provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0114] An embodiment of the present application provides a computer program product, including a computer program, wherein the computer program is executed by a third processor to implement the method described above.

[0115] The computer program product provided by the above-mentioned embodiments of the present application and the decision tree generation method based on a large language model provided by the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0116] It should be noted that, in this application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or still includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0117] Each embodiment in the present application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for evaluating the decision tree generation method based on a large language model, electronic device, electronic device, and readable storage medium embodiment, since it is basically similar to the above-mentioned decision tree generation method based on a large language model, the description is relatively simple, and the relevant parts can be referred to the partial description of the above-mentioned decision tree generation method based on a large language model.

[0118] Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application shall be subject to the scope defined by the claims.

Claims

1. A decision tree generation method based on a large language model, characterized in that: include: Acquire a training sample set, target medical document information, and target hospital information, wherein the target medical document information includes an acute upper respiratory tract infection disease intervention guideline, and the training sample set includes other clinical guideline document information that matches the target medical document information; Processing the target hospital information to generate attribute information of the target hospital, wherein the attribute information of the target hospital is used to represent historical feedback data of the target hospital for the target medical document information; Preprocessing the training sample set to generate a training sample set with target feature data, wherein the target feature data is used to characterize node information and decision path information of clinical guideline document information processing; Acquire a preset large language model that matches the target feature data, wherein the preset large language model is set based on the target hospital information received within a preset time period; Training the preset large language model based on the training data set with target feature data to generate a target large language model; Processing the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information; The target medical document information and the attribute information of the target hospital are processed based on the target large language model to generate decision tree information, including: processing the target medical document information based on the target large language model to generate initial node information and decision information matching the initial node information, wherein the initial node information is used to characterize the user's attention items; processing the attribute information of the target hospital based on the target large language model to generate node correlation information, wherein the node correlation information is used to characterize the correlation between different nodes; processing the initial node information based on the node correlation information to generate a decision adjustment factor; processing the initial node information and the decision information matching the initial node information based on the decision adjustment factor to generate target node information and target decision information matching the target node information; generating decision tree information based on the target node information and the target decision information; The attribute information of the target hospital is processed based on the target large language model to generate node correlation information, including: processing the attribute information of the target hospital based on the target large language model to generate clinical symptom classification information; processing the clinical symptom classification information to generate symptom attribute information of different users, wherein the symptom attribute information of different users is used to characterize the correlation between different users and different symptoms; processing the symptom attribute information of different users to generate node correlation information.

2. The method according to claim 1, characterized in that The target hospital information is processed to generate attribute information of the target hospital, including: Processing the target hospital information to generate user hospitalization information, wherein the user hospitalization information includes historical hospitalization information of several patients matching the target medical document information; Performing feature extraction processing on the user's hospitalization information to generate user historical physiological feature information and user historical behavior information; Processing the user's historical physiological characteristic information and the user's historical behavior information to generate historical feedback data; The attribute information of the target hospital is generated based on the historical feedback data.

3. The method according to claim 1, characterized in that Preprocessing the training sample set to generate a training sample set with target feature data includes: Extracting features from the training sample set to generate an original feature library; Classifying the original feature library to generate user feature information, wherein the user feature information is used to characterize the physiological features of the same user for different diseases; Processing the user characteristic information to generate a characteristic mapping table, wherein the characteristic mapping table is used to characterize different physiological characteristics of different patients for the same disease; The original feature library is processed based on the feature mapping table to generate target feature data.

4. The method according to claim 3, characterized in that Extracting features from the training sample set to generate an original feature library includes: Obtain historical clinical guideline document datasets; Processing the historical clinical guideline document dataset to obtain abnormal node features and abnormal co-occurrence frequencies matching the abnormal node features; Processing the abnormal node features and the abnormal co-occurrence frequencies matching the abnormal node features to generate a number of node information; Process the node information to generate the original feature library; The method also includes a calculation formula for calculating the abnormal co-occurrence frequency, and the calculation formula is: ； in, is the probability of A and B co-occurring, are the probabilities of A and B appearing alone, respectively.

5. The method according to claim 1, characterized in that The preset large language model is trained based on the training data set with target feature data to generate a target large language model, including: Get any number of data features in the training data set with target feature data; Generate adjacent features based on the distance between the any number of data features and other numbers of data features of the same category, wherein the adjacent features include a preset number of the any number of data features; Determining a sampling ratio based on the number of each data feature in the training data set; Based on the sampling ratio, determining a sampling ratio; Sampling the adjacent features based on the sampling ratio to generate a preset number of sampling features; Based on any data feature and each sampling feature, generating multiple data groups, wherein each data group includes a preset number of data samples, and at least one data sample includes identification information; Processing multiple data sets to generate training sets and validation sets, wherein both the training sets and the validation sets include a number of data features; Training the preset large language model based on the training set to generate a trained large language model; Processing the trained large language model based on the verification set to generate a verification result; If the data sample containing identification information in the verification result is node information and decision path information, the trained large language model is used as the target large language model.

6. A decision tree generation device based on a large language model, characterized in that: The device comprises: An acquisition module is used to acquire a training sample set, target medical document information and target hospital information, wherein the target medical document information includes an acute upper respiratory tract infection disease intervention guide, and the training sample set includes other clinical guideline document information that matches the target medical document information; acquire a preset large language model that matches the target feature data, wherein the preset large language model is set based on the target hospital information received within a preset time period; A processing module is used to process the target hospital information to generate attribute information of the target hospital, wherein the attribute information of the target hospital is used to characterize historical feedback data of the target hospital for the target medical document information; preprocess the training sample set to generate a training sample set with target feature data, wherein the target feature data is used to characterize node information and decision path information of clinical guideline document information processing; train the preset large language model based on the training data set with target feature data to generate a target large language model; process the target medical document information and the attribute information of the target hospital based on the target large language model to generate decision tree information; The target medical document information and the attribute information of the target hospital are processed based on the target large language model to generate decision tree information, including: processing the target medical document information based on the target large language model to generate initial node information and decision information matching the initial node information, wherein the initial node information is used to characterize the user's attention items; processing the attribute information of the target hospital based on the target large language model to generate node correlation information, wherein the node correlation information is used to characterize the correlation between different nodes; processing the initial node information based on the node correlation information to generate a decision adjustment factor; processing the initial node information and the decision information matching the initial node information based on the decision adjustment factor to generate target node information and target decision information matching the target node information; generating decision tree information based on the target node information and the target decision information; The attribute information of the target hospital is processed based on the target large language model to generate node correlation information, including: processing the attribute information of the target hospital based on the target large language model to generate clinical symptom classification information; processing the clinical symptom classification information to generate symptom attribute information of different users, wherein the symptom attribute information of different users is used to characterize the correlation between different users and different symptoms; processing the symptom attribute information of different users to generate node correlation information.

7. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; The first processor is configured to execute the decision tree generation method based on a large language model as described in any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the decision tree generation method based on the large language model described in any one of claims 1 to 5 is implemented.

Images