Large-model-based medical industry-oriented medical record fine adjustment data set generation and evaluation method and system for inquiry dialogue writing

Generate and evaluate the medical record fine-tuning data set for consultation dialogue through data collection, scene construction and model fine-tuning methods, which solves the problem of converting doctor-patient communication into standard medical records, and achieves efficient and accurate medical dialogue generation and protects privacy And improve the model's medical language comprehension ability.

CN119049626APending Publication Date: 2024-11-29SHANDONG LANGCHAO YUNTOU INFORMATION TECH CO LTD
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411118499.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing technology is difficult to convert doctor-patient communication into standard medical records efficiently and accurately. Especially when dealing with the complexity, structural rigor and privacy ethics of medical languages, pre-training large language models requires a large number of doctor-patient consultation and conversations and electronic medical records. The data is fine-tuned, but it is difficult to obtain the dataset.

Method used

Through methods such as data acquisition and preprocessing, scene construction and parameterization, model fine-tuning and optimization, fine-tuning data set generation and data evaluation, the medical record fine-tuning data set used for consultation and dialogue are generated and evaluated to ensure that the data complies with privacy protection and electronic medical records to ensure that the data complies with privacy protection and electronic medical records. Specification, enhance the accuracy and naturalness of the model in the generation of medical conversations.

Benefits of technology

Effectively solve the problem of data source, protect patient privacy, enhance the generalization ability of the model in the medical language environment, and the generated doctor-patient dialogue content is more in line with the complex medical language environment, improving the scientific rigor and personalization of the generated content. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119049626A_ABST
    Figure CN119049626A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a large-model-based medical industry-oriented medical record fine adjustment data set generation and evaluation method and system for inquiry dialogue writing, and the method comprises the following steps: data collection and preprocessing, scene construction and parameterization, model fine adjustment and optimization, fine adjustment data set generation, and data evaluation. The method and the system for generating and evaluating the inquiry dialogue writing medical record fine-tuning data set based on the large model and oriented to the medical industry have the beneficial effects that the problem of data sources is effectively solved by adopting a synthetic data strategy and generating data through simulation in the face of legal and ethical restrictions of real doctor-patient dialogue data acquisition, and the method and the system have high practical value. Meanwhile, the privacy of the patient is protected, and a new path for medical AI data generation is opened up.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a method and system for generating and evaluating a data set for fine-tuning medical record writing in medical consultation dialogues based on a large model for the medical industry. Background Art

[0002] The rapid progress of information technology is leading the digital transformation of the healthcare sector. The promotion of the electronic medical record (EMR) system is a core measure to improve the quality and efficiency of medical services. It not only efficiently stores patient information, but also promotes the flow of information across departments and strengthens the data support system for clinical decision-making and disease prevention and control. However, the preparation of electronic medical records takes up a lot of working time of medical staff, which brings a heavy burden to medical staff.

[0003] Artificial intelligence, especially natural language processing and machine learning technology, provides a new solution to this problem, which is expected to reduce the workload and improve the quality of records by automatically summarizing and generating medical records. Existing technical attempts, such as template filling and keyword extraction, have achieved initial results, but they are difficult to grasp the complexity of medical language, including the diversity of terminology, the rigor of structure and the unstructured characteristics of dialogue. The generated content is often shallow, lacks personalization and clinical value, and is prone to deviations, especially when dealing with special situations.

[0004] Therefore, how to efficiently and accurately convert doctor-patient communication into standard medical records has become an urgent challenge to be solved. The model needs to not only master the art of language, but also have a deep understanding of medical knowledge to ensure that the records are both scientific, rigorous and personalized. Pre-trained large language models provide a powerful platform for automated medical record writing, but they still have to face challenges such as medical language adaptability, structured information understanding and privacy ethics. In order to enhance the capabilities of the pre-trained large prediction model in this scenario, a large amount of doctor-patient consultation dialogue and electronic medical record data are needed to fine-tune the pre-trained large prediction model, but it is very difficult to obtain fine-tuning data sets. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for generating and evaluating a fine-tuning dataset for medical record writing in consultation dialogues based on a large model for the medical industry, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for generating and evaluating a fine-tuning dataset for writing medical records in a medical consultation dialogue based on a large model for the medical industry, the method comprising the following steps:

[0007] Data collection and preprocessing, by collecting medical records and conducting preliminary evaluation, filtering, desensitization and structuring of the records in accordance with electronic medical record specifications to ensure that the data complies with privacy protection and electronic medical record specifications;

[0008] Scenario construction and parameterization: construct multi-department dialogue scenarios based on medical practice, design parameter control systems to adapt to different patient characteristics, and realize dynamic adjustment of dialogue content;

[0009] Model fine-tuning and optimization: Using pre-processed data and scenario parameters, the pre-trained language model is fine-tuned through the prompt engineering of the large model combined with supervised data fine-tuning to improve the accuracy and naturalness of the model in medical dialogue generation;

[0010] Fine-tune dataset generation, design parameterized templates and perform parameter randomization and filling, call the large model interface to generate dialogue samples, and perform quality screening to ensure the purity of the dataset;

[0011] Data evaluation: Establish an evaluation system that includes medical expert review, automated testing, and user feedback to ensure the medical accuracy, logical coherence, and ethical compliance of the generated dialogue content, and use a double-blind scoring system to select high-quality data sets.

[0012] Preferably, data collection and preprocessing also include removing sensitive information, identifying and eliminating records that do not meet standards or are of low quality, and converting unstructured text into a standardized data format.

[0013] Preferably, the scene is constructed and parameterized, and the conversation script is designed according to the needs of different departments, and the personalized generation of the conversation content is achieved through parameterized control.

[0014] Preferably, model fine-tuning and optimization also includes iteratively adjusting the prompt strategy to optimize the performance of the model in specific medical scenarios, and performing supervised fine-tuning in combination with an existing doctor-patient consultation dialogue dataset.

[0015] Preferably, in the step of fine-tuning the data set generation, the parameterized template design includes reserving variable positions to accommodate the filling of different patient information, and using programming techniques to perform parameter randomization and filling to reflect the diversity of the real world;

[0016] Data evaluation includes scoring the consistency, logical rationality, and accuracy of professional terminology between the generated dialogue content and the original medical records, and establishing a veto mechanism for contextual logic to ensure the high quality of the data set.

[0017] A system for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, the system comprising:

[0018] The data collection and preprocessing module collects a large amount of original electronic medical record data from a wide range of channels, including online health platforms and medical institution records. It uses advanced natural language processing technology to deeply clean and organize the data, remove irrelevant information, protect personal privacy, and ensure the consistency and accuracy of medical terminology through standardization.

[0019] The scenario construction and parameterization module simulates the rich and varied consultation scenarios in the real world. By designing different dialogue scripts and parameterized templates, it can deal with different patient conditions and disease conditions. The templates cover basic condition inquiries, symptom descriptions, and also go deep into medical history reviews. The parameterized design allows the system to automatically generate personalized dialogue content based on the input patient characteristics, greatly enhancing the pertinence and effectiveness of the interaction.

[0020] The model fine-tuning and optimization module guides the pre-trained language model to learn in depth in the medical field through carefully annotated training data sets. The data sets not only contain actual consultation records, but also include annotations and structured information reviewed by professional doctors, such as disease codes, drug names and their usage. Through data enhancement strategies such as synonym replacement and context changes, data diversity is further enriched to promote the improvement of model generalization ability.

[0021] Fine-tune the dataset generation module. Based on the customized big model, the doctor-patient consultation dialogue content is generated in batches through parameterized template design, parameter randomization and filling, and calling the big model interface. After generation, a preliminary quality screening is carried out to remove those dialogues that are obviously illogical, repetitive, or contain erroneous medical information to ensure the purity of the dataset.

[0022] The evaluation module formulates data quality assessment standards, builds a multi-level evaluation system that includes medical expert review, automated testing and user feedback, conducts a comprehensive review of the model's output, including the medical accuracy, logical coherence and compliance with ethical standards of the model-generated dialogue, establishes data quality assessment standards, double-blindly scores the data according to the standards, and continuously iterates and optimizes to ensure high-quality output of doctor-patient dialogue content, meeting or even exceeding the high standards of fine-tuning data sets.

[0023] Preferably, the data collection and preprocessing module also includes removing sensitive information, identifying and eliminating records that do not meet the specifications or are of low quality, and converting unstructured text into a standardized data format.

[0024] Preferably, the scenario construction and parameterization module designs dialogue scripts according to the needs of different departments, and realizes personalized generation of dialogue content through parameterized control.

[0025] Preferably, the model fine-tuning and optimization module also includes iteratively adjusting the prompt strategy to optimize the performance of the model in specific medical scenarios, and performing supervised fine-tuning in combination with an existing doctor-patient consultation dialogue dataset.

[0026] Preferably, the data set generation step module is fine-tuned, and the parameterized template design includes reserving variable positions to accommodate the filling of different patient information, and using programming techniques to perform parameter randomization and filling to reflect the diversity of the real world;

[0027] The evaluation module includes scoring the consistency, logical rationality and accuracy of professional terminology between the generated dialogue content and the original medical records, and establishing a veto mechanism for contextual logic to ensure the high quality of the data set.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention proposes a method and system for generating and evaluating a fine-tuning dataset for medical record writing in consultation dialogues based on a large model for the medical industry. Faced with legal and ethical restrictions on the acquisition of real doctor-patient dialogue data, the present invention adopts a synthetic data strategy to generate data through simulation, which effectively solves the problem of data source, while protecting patient privacy and opening up a new path for medical AI data generation; the fine-tuning dataset generation module enriches data diversity through data enhancement strategies such as synonym replacement and context change, promotes the generalization ability of the model when dealing with different medical scenarios, and makes the doctor-patient dialogue content generated by the model more suitable for the complex medical language environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Embodiment 1, the present invention provides a technical solution: a method for generating and evaluating a fine-tuning dataset for writing medical records for consultation dialogues based on a large model for the medical industry, the method comprising the following steps:

[0033] The first step is data collection and preprocessing. This step collects medical records, conducts a preliminary assessment of the collected electronic medical record records according to the electronic medical record specifications, filters out records that do not meet the specifications, desensitizes the records, removes irrelevant information, and analyzes the electronic medical record records in a structured manner. The collected raw data must undergo strict screening and preprocessing to ensure that it complies with the electronic medical record specifications of the relevant departments. The key tasks in the preprocessing stage include removing sensitive information, such as patient name and address, to protect patient privacy; at the same time, identifying and eliminating records that do not meet the specifications or are of low quality to ensure the quality of data for subsequent processing. Through structured parsing, unstructured text information is converted into a standardized, machine-friendly data format. For example, "headache for 2 days" is parsed into "symptoms: headache, duration: 2 days".

[0034] The second step is scenario construction and parameterization. This step is to construct a dialogue scenario closely related to medical practice after the data has been structured. This link requires in-depth analysis of the unique needs of different departments (such as internal medicine, surgery, and pediatrics) to design dialogue scripts that are close to real situations. Each scenario corresponds to a specific department and disease background, such as hypertension diagnosis in the cardiovascular department or pregnancy management in the obstetrics and gynecology department. By abstracting the key parameters in the medical records (such as patient gender, age, specific disease information, etc.), a flexible parameter control system is designed so that the generated dialogue can be dynamically adjusted according to the actual patient situation, thereby improving the authenticity and practicality of the dialogue. According to different scenarios, the mapping relationship between the scene and the medical record module is set, and variable parameters such as patient gender, age, disease information, and department information are extracted for parameterized control of the generation of doctor-patient dialogues.

[0035] The third step is model fine-tuning and optimization. Based on the pre-processed electronic medical record data and carefully designed scenario parameters, the fine-tuning dataset generation module uses advanced natural language processing technology, especially the prompt engineering skills of large models, to guide the model to generate high-quality doctor-patient consultation dialogues. This process not only needs to consider how to make the generated dialogue cover all the necessary information in the medical record, but also to ensure the natural fluency and logical coherence of the dialogue. During the fine-tuning process, the prompt strategy is continuously iterated to optimize the performance of the model in specific scenarios, ensuring that the generated content is consistent with medical knowledge and consultation scenarios, and combining the existing doctor-patient consultation dialogues and electronic medical record data sets to conduct multiple rounds of supervised fine-tuning of the model to improve the model dialogue generation effect.

[0036] The fourth step is to generate fine-tuning dataset. This step includes the following sub-steps:

[0037] 1. Parameterized template design. Design a set of flexible dialogue templates that include the basic framework of the dialogue, such as greetings, asking about the chief complaint, medical history review, symptom details exploration, etc., and reserve variable positions in the template, such as {patient_name}, {gender}, {chief_complaint}, etc., so that they can be filled in according to the electronic medical record data and scenario parameters.

[0038] 2. Parameter randomization and filling. Based on the structured information obtained in the data preprocessing stage, the patient's name, gender, chief complaint and other variable values ​​are randomly selected or set according to certain rules (such as gender ratio and age distribution) through programming. The selection of these parameters, synonym replacement, and scenario change strategies are intended to reflect the diversity of the real world, while also ensuring that the generated dialogue covers a wide range of medical records.

[0039] 3. Calling the big model interface. Using the script, the filled parameterized template is used as a prompt to call the big model's question-and-answer interface. The role of the big model is to generate coherent and reasonable doctor-patient dialogue content based on given scenarios and parameters.

[0040] 4. Batch generation and initial quality screening. By repeatedly executing steps 1 to 3 above, a large number of doctor-patient conversation samples can be generated in batches. After generation, a preliminary quality screening is carried out to remove those conversations that are obviously illogical, repetitive, or contain incorrect medical information to ensure the purity of the data set.

[0041] The fifth step is data evaluation. Develop a comprehensive evaluation system that includes multiple dimensions, such as consistency between the conversation content and the original medical record, logical rationality, and accuracy of professional terminology. For example, the more medical record items such as "chief complaint" and "history of current illness" mentioned in the conversation, the higher the score; conversely, if there is content unrelated to the medical record or incorrect medical terminology, points will be deducted accordingly. Establish a veto mechanism for contextual logic, and any conversation that seriously violates medical common sense or is logically confusing will be directly judged as unqualified. Double-blind scoring is used to ensure objectivity and fairness of the scoring, and only data with a score of 60 or above and a score difference of less than 10 points will be included in the final data set to ensure the quality of the data set and the reliability of the model.

[0042] Specific scoring rules: According to the structured content of the electronic medical record, the quality of the generated dialogue is evaluated according to the medical record items such as chief complaint, current medical history, past medical history, personal history, marital history, menstrual and reproductive history, family history, physical examination, and auxiliary examinations. When the above items in the medical record appear or are mentioned in the dialogue, N points are added for each item that appears; for items that are not included in the medical record but are missing in the dialogue content, the relevance of the content to the electronic medical record is evaluated. If there is obvious irrelevance to the context logic or fabricated disease descriptions, points are deducted, and M points are deducted for each item that appears; the accuracy of the medical terms in the dialogue is evaluated. If there is inaccuracy, L points are deducted for each item that appears; the entire dialogue context logic is rejected, and if the logic is unreasonable and differs greatly from the actual scenario, it is judged to be 0 points; 60 points are used as the baseline score, and the quality of the generated doctor-patient dialogue is scored as a passing standard of 60 points. The scoring is conducted in a double-blind manner according to the scoring criteria, and data with a score difference of more than 10 points are not included in the data set.

[0043] For example 2, please refer to Figure 1 Based on the first embodiment, a system for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model and oriented to the medical industry is proposed. The system includes:

[0044] Data collection and preprocessing module, which collects a large amount of original electronic medical record data from a wide range of channels, including online health platforms and medical institution records, and uses advanced natural language processing technology to deeply clean and organize these data, remove irrelevant information, protect personal privacy, and ensure the consistency and accuracy of medical terminology through standardized processing. This process is not only related to data availability, but also a key step in protecting patient privacy and data compliance.

[0045] The scenario construction and parameterization module is designed to simulate the rich and varied consultation scenarios in the real world. By designing different dialogue scripts and parameterized templates, the system can flexibly respond to different patient conditions and disease conditions. These templates not only cover basic condition inquiries and symptom descriptions, but also go deep into aspects such as medical history review. The parameterized design allows the system to automatically generate personalized dialogue content based on the input patient characteristics, greatly enhancing the pertinence and effectiveness of the interaction.

[0046] Model fine-tuning and optimization module: This module guides the pre-trained language model to learn in depth in the medical field through carefully annotated training data sets. The data set not only contains actual consultation records, but also includes annotations and structured information reviewed by professional doctors, such as disease codes, drug names and their usage. Through data enhancement strategies, such as synonym replacement and context changes, data diversity is further enriched to promote the improvement of model generalization ability. This step is the core to ensure that the model can accurately generate professional medical advice and guidance.

[0047] Fine-tune the dataset generation module. This module is based on the customized big model mentioned above. It generates doctor-patient consultation dialogue content in batches through parameterized template design, parameter randomization and filling, and calling the big model interface. After generation, implement preliminary quality screening to remove those dialogues that are obviously illogical, repetitive, or contain erroneous medical information to ensure the purity of the dataset.

[0048] The evaluation module comprehensively reviews the output of the model by developing data quality evaluation standards and building a multi-level evaluation system including medical expert review, automated testing, and user feedback. This includes testing the medical accuracy, logical coherence, and compliance with ethical standards of the dialogue generated by the model. Data quality assessment standards are established, and data is double-blindly scored according to the standards. Continuous iteration and optimization are carried out to ensure high-quality output of doctor-patient dialogue content, meeting or even exceeding the high standards of fine-tuning data sets.

[0049] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating and evaluating a fine-tuning dataset for medical record writing in a medical consultation dialogue based on a large model, which is characterized by: The method comprises the following steps: Data collection and preprocessing, by collecting medical records and conducting preliminary evaluation, filtering, desensitization and structuring of the records in accordance with electronic medical record specifications to ensure that the data complies with privacy protection and electronic medical record specifications; Scenario construction and parameterization: construct multi-department dialogue scenarios based on medical practice, design parameter control systems to adapt to different patient characteristics, and realize dynamic adjustment of dialogue content; Model fine-tuning and optimization: Using pre-processed data and scenario parameters, the pre-trained language model is fine-tuned through the prompt engineering of the large model combined with supervised data fine-tuning to improve the accuracy and naturalness of the model in medical dialogue generation; Fine-tune dataset generation, design parameterized templates and perform parameter randomization and filling, call the large model interface to generate dialogue samples, and perform quality screening to ensure the purity of the dataset; Data evaluation: Establish an evaluation system that includes medical expert review, automated testing, and user feedback to ensure the medical accuracy, logical coherence, and ethical compliance of the generated dialogue content, and use a double-blind scoring system to select high-quality data sets.

2. According to claim 1, a method for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: Data collection and preprocessing also include removing sensitive information, identifying and eliminating records that do not meet specifications or are of low quality, and converting unstructured text into standardized data formats.

3. According to claim 1, a method for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: Scenario construction and parameterization: design conversation scripts according to the needs of different departments, and realize personalized generation of conversation content through parameterized control.

4. According to claim 1, a method for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: Model fine-tuning and optimization also include iteratively adjusting the prompt strategy to optimize the model's performance in specific medical scenarios, and performing supervised fine-tuning based on existing doctor-patient consultation dialogue datasets.

5. According to claim 1, a method for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: Fine-tuning the dataset generation step, parameterized template design includes reserving variable positions to accommodate the filling of different patient information, and using programming techniques to randomize and fill parameters to reflect the diversity of the real world; Data evaluation includes scoring the consistency, logical rationality, and accuracy of professional terminology between the generated dialogue content and the original medical records, and establishing a veto mechanism for contextual logic to ensure the high quality of the data set.

6. A method for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry according to any one of claims 1 to 5, characterized in that: The system comprises: The data collection and preprocessing module collects a large amount of original electronic medical record data from a wide range of channels, including online health platforms and medical institution records. It uses advanced natural language processing technology to deeply clean and organize the data, remove irrelevant information, protect personal privacy, and ensure the consistency and accuracy of medical terminology through standardization. The scenario construction and parameterization module simulates the rich and varied consultation scenarios in the real world. By designing different dialogue scripts and parameterized templates, it can deal with different patient conditions and disease conditions. The templates cover basic condition inquiries, symptom descriptions, and also go deep into medical history reviews. The parameterized design allows the system to automatically generate personalized dialogue content based on the input patient characteristics, greatly enhancing the pertinence and effectiveness of the interaction. The model fine-tuning and optimization module guides the pre-trained language model to learn in depth in the medical field through carefully annotated training data sets. The data sets not only contain actual consultation records, but also include annotations and structured information reviewed by professional doctors, such as disease codes, drug names and their usage. Through data enhancement strategies such as synonym replacement and context changes, data diversity is further enriched to promote the improvement of model generalization ability. Fine-tune the dataset generation module. Based on the customized big model, the doctor-patient consultation dialogue content is generated in batches through parameterized template design, parameter randomization and filling, and calling the big model interface. After generation, a preliminary quality screening is carried out to remove those dialogues that are obviously illogical, repetitive, or contain erroneous medical information to ensure the purity of the dataset. The evaluation module formulates data quality assessment standards, builds a multi-level evaluation system that includes medical expert review, automated testing and user feedback, conducts a comprehensive review of the model's output, including the medical accuracy, logical coherence and compliance with ethical standards of the model-generated dialogue, establishes data quality assessment standards, double-blindly scores the data according to the standards, and continuously iterates and optimizes to ensure high-quality output of doctor-patient dialogue content, meeting or even exceeding the high standards of fine-tuning data sets.

7. According to claim 6, a system for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: The data collection and preprocessing module also includes removing sensitive information, identifying and eliminating records that do not meet specifications or are of low quality, and converting unstructured text into standardized data formats.

8. According to claim 6, a system for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: The scenario construction and parameterization module designs conversation scripts according to the needs of different departments and realizes personalized generation of conversation content through parameterized control.

9. According to claim 6, a large model-based medical industry-oriented data set generation and evaluation system for fine-tuning medical record writing in consultation dialogues, characterized in that: The model fine-tuning and optimization module also includes iterative adjustment of the prompt strategy to optimize the performance of the model in specific medical scenarios, and supervised fine-tuning based on the existing doctor-patient consultation dialogue dataset.

10. According to claim 6, a system for generating and evaluating a medical record fine-tuning dataset for medical consultation dialogue writing based on a large model for the medical industry, characterized in that: Fine-tune the dataset generation step module. The parameterized template design includes reserving variable positions to accommodate the filling of different patient information, and using programming techniques to randomize and fill parameters to reflect the diversity of the real world. The evaluation module includes scoring the consistency, logical rationality and accuracy of professional terminology between the generated dialogue content and the original medical records, and establishing a veto mechanism for contextual logic to ensure the high quality of the data set.

Citation Information

Cited By

  • Medical digital human simulation test management system and method based on situational dialogue

    CN119782154A

  • Medical text structured processing method based on Prompt technology

    CN120015344A

  • Industrial instruction data construction method and system based on large model

    CN120409690A

  • Model optimization method and apparatus, device, and storage medium

    WO2026129157A1