CN-121983213-A - Medical record dialogue data generation method, system, terminal and storage medium

CN121983213ACN 121983213 ACN121983213 ACN 121983213ACN-121983213-A

Abstract

The invention discloses a medical record dialogue data generation method, a system, a terminal and a storage medium, wherein the method comprises the steps of obtaining medical record data and historical question-answer sequence information of a target user, obtaining a target medical record state according to the medical record data and the historical question-answer sequence information, and carrying out information evaluation to obtain an information evaluation result; the method comprises the steps of determining an instruction according to an information evaluation result to obtain a target instruction, generating a medical record question-answer pair, performing quality inspection to obtain a quality inspection result, obtaining medical record dialogue data according to the quality inspection result, updating historical question-answer sequence information to obtain an updating result, performing iterative evaluation on the medical record data to obtain an iterative evaluation result, obtaining a medical record dialogue data set corresponding to the medical record data if a preset end condition is met, and storing the question-answer sequence information of the medical record dialogue data set. The invention improves the information integrity, thereby reducing the risk of information loss and improving the information density and the reasoning depth of the generated dialogue data.

Inventors

HUANG JINCHENG
QIN XINGDE

Assignees

深圳心语心言智能技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260210

Claims (10)

1. A medical record session data generation method, characterized in that the medical record session data generation method comprises: acquiring medical record data and historical question-answer sequence information of a target user, obtaining a target medical record state according to the medical record data and the historical question-answer sequence information, and carrying out information evaluation on the target medical record state to obtain an information evaluation result; Determining a target instruction according to the information evaluation result, generating a corresponding medical record question-answer pair according to the target instruction, performing quality inspection processing on the medical record question-answer pair to obtain a quality inspection result, and obtaining medical record dialogue data according to the quality inspection result; updating the historical question-answer sequence information according to the medical record dialogue data to obtain an updating result, and carrying out iterative evaluation on the medical record data according to the updating result to obtain an iterative evaluation result; And if the iteration evaluation result meets a preset ending condition, obtaining a medical record dialogue data set corresponding to the medical record data according to the iteration evaluation result, and storing question-answer sequence information of the medical record dialogue data set.
2. The medical record dialogue data generating method according to claim 1, wherein the obtaining the medical record data and the historical question-answer sequence information of the target user obtains a target medical record state according to the medical record data and the historical question-answer sequence information, and performs information evaluation on the target medical record state to obtain an information evaluation result, and the method specifically comprises: Acquiring medical record data of a target user, carrying out question-answer sequence matching on the target user to obtain a question-answer matching result, and extracting historical question-answer sequence information in the question-answer matching result; Obtaining a target medical record state according to the medical record data and the historical question-answer sequence information, and performing information saturation calculation on medical record information in the target medical record state to obtain a saturation value; And obtaining a candidate action set of the target medical record state, carrying out utility value calculation on each candidate action in the candidate action set to obtain an action utility value calculation result, and obtaining an information evaluation result according to the saturation value and the action utility value calculation result.
3. The medical record dialogue data generating method according to claim 2, wherein the calculating of the information saturation of the medical record information in the target medical record state specifically includes: ; Wherein, the For the medical record information saturation evaluation function, For the number of medical entities identified in the medical record information, For the number of entities identified in the medical record information, For the number of medical events identified in the medical record information, For the number of events identified in the medical record information, The noise information is scored as such, Is the first The medical record information input by the wheel, 、 And Are weight coefficients.
4. The medical record dialog data generation method according to claim 2, wherein the performing utility value calculation on each candidate action in the candidate action set specifically includes: ; Wherein, the A function is calculated for the action utility value, As a candidate action, Is the first Historical question-answer sequence information before the round of processing, Is the first The medical record information input by the wheel, For the correlation of actions with current medical record information and historical question-answer sequence information, New information utility values that may be introduced for an action, In order for the action to be at risk of causing an error, 、 And Are super parameters.
5. The medical record dialogue data generating method according to claim 2, wherein the determining a target instruction according to the information evaluation result, and generating a corresponding medical record question-answer pair according to the target instruction, specifically comprises: if the information evaluation result is that the saturation value is larger than a first preset threshold value and the action utility value calculation result has high utility value action, determining that the target instruction is a problem action instruction; performing information analysis on the medical record information according to the generated problem action instruction to obtain a corresponding medical entity, and performing problem generation according to the medical entity and the context constraint of the historical question-answering sequence information to obtain a medical record problem; And carrying out information positioning on the medical record information according to the medical record questions to obtain target answer information, obtaining medical record answers according to the target answer information, and obtaining medical record question-answer pairs according to the medical record questions and the medical record answers.
6. The medical record dialogue data generating method according to claim 1, wherein the quality inspection processing is performed on the medical record question-answer pair to obtain a quality inspection result, and medical record dialogue data is obtained according to the quality inspection result, specifically comprising: Performing confidence calculation on the medical record question-answer pair to obtain question-answer confidence, performing logic analysis on the medical record question-answer pair to obtain a logic analysis result, and obtaining a quality inspection result according to the question-answer confidence and the logic analysis result; If the quality inspection result is that the question and answer confidence coefficient is larger than a second preset threshold value and the logic analysis result is that logic is normal, judging that the quality inspection of the medical record question and answer pair passes, and obtaining medical record dialogue data according to the medical record question and answer pair.
7. The medical record dialogue data generation method according to claim 6, wherein the step of performing confidence calculation on the medical record question-answer pair to obtain question-answer confidence, performing logic analysis on the medical record question-answer pair to obtain a logic analysis result, and obtaining a quality inspection result according to the question-answer confidence and the logic analysis result, and further comprising: if the quality inspection result is that the question and answer confidence is smaller than or equal to the second preset threshold value or the logic analysis result is logic abnormality, judging that the quality inspection of the medical record question and answer pair is not passed; and carrying out feedback processing on the medical record question-answer pair to obtain a feedback result, and carrying out instruction processing on the feedback result, wherein the instruction processing comprises a question modifying action instruction, an answer modifying action instruction and an ending action instruction.
8. A medical record session data generation system, the medical record session data generation system comprising: the medical record information evaluation module is used for acquiring medical record data and historical question-answer sequence information of a target user, obtaining a target medical record state according to the medical record data and the historical question-answer sequence information, and carrying out information evaluation on the target medical record state to obtain an information evaluation result; the medical record dialogue quality inspection module is used for determining a target instruction according to the information evaluation result, generating a corresponding medical record question-answer pair according to the target instruction, performing quality inspection processing on the medical record question-answer pair to obtain a quality inspection result, and obtaining medical record dialogue data according to the quality inspection result; The iteration evaluation module is used for updating the historical question-answer sequence information according to the medical record dialogue data to obtain an updating result, and carrying out iteration evaluation on the medical record data according to the updating result to obtain an iteration evaluation result; And the dialogue data collection module is used for obtaining a medical record dialogue data set corresponding to the medical record data according to the iteration evaluation result if the iteration evaluation result meets a preset ending condition, and storing the question-answer sequence information of the medical record dialogue data set.
9. A terminal comprising a memory, a processor and a program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the medical record session data generation method according to any one of claims 1-7.
10. A computer readable storage medium, having stored thereon a computer program, the computer readable storage medium having stored thereon a medical record session data generating program which, when executed by a processor, implements the steps of the medical record session data generating method according to any of claims 1-7.

Description

Medical record dialogue data generation method, system, terminal and storage medium Technical Field The present invention relates to the field of data processing technologies, and in particular, to a medical record dialogue data generating method, a system, a terminal, and a computer readable storage medium. Background With the increasing medical intelligence application of LLM (Large Language Models, large-scale pre-training language model), constructing high-quality conversational medical data sets for training and evaluation becomes a key bottleneck, although the prior art has proposed using large language models or multi-agent frameworks for structured processing or conversational generation of medical record text data, mainly adopting two methods, (1) direct conversion based on single LLM (usually aided with prompt engineering, instruction fine-tuning and advanced abstracting to reduce context length), and (2) multi-agent-based clinical simulation systems. However, the existing method has the defects that the processing capability of the long-time-sequence medical record is limited, the information integrity is difficult to guarantee, a single model has higher error rate in terms of fact consistency and robustness in the face of noise data, the information density and reasoning depth of the generated question-answer or dialogue data are insufficient, and the generated content cannot be continuously participated in subsequent processing as a system state due to the lack of explicit state management of a history output result in the process of generating a dialogue, so that the cross-round consistency is difficult to guarantee. Accordingly, the prior art is still in need of improvement and development. Disclosure of Invention The invention mainly aims to provide a medical record dialogue data generation method, a system, a terminal and a storage medium, and aims to solve the problems that the existing medical record dialogue data generation method cannot guarantee information integrity and cross-round consistency, and is low in reasoning depth, so that accuracy is low. In order to achieve the above object, the present invention provides a medical record dialogue data generating method, which includes the following steps: acquiring medical record data and historical question-answer sequence information of a target user, obtaining a target medical record state according to the medical record data and the historical question-answer sequence information, and carrying out information evaluation on the target medical record state to obtain an information evaluation result; Determining a target instruction according to the information evaluation result, generating a corresponding medical record question-answer pair according to the target instruction, performing quality inspection processing on the medical record question-answer pair to obtain a quality inspection result, and obtaining medical record dialogue data according to the quality inspection result; updating the historical question-answer sequence information according to the medical record dialogue data to obtain an updating result, and carrying out iterative evaluation on the medical record data according to the updating result to obtain an iterative evaluation result; And if the iteration evaluation result meets a preset ending condition, obtaining a medical record dialogue data set corresponding to the medical record data according to the iteration evaluation result, and storing question-answer sequence information of the medical record dialogue data set. Optionally, in the medical record dialogue data generating method, the obtaining medical record data and historical question-answer sequence information of the target user obtains a target medical record state according to the medical record data and the historical question-answer sequence information, and performs information evaluation on the target medical record state to obtain an information evaluation result, which specifically includes: Acquiring medical record data of a target user, carrying out question-answer sequence matching on the target user to obtain a question-answer matching result, and extracting historical question-answer sequence information in the question-answer matching result; Obtaining a target medical record state according to the medical record data and the historical question-answer sequence information, and performing information saturation calculation on medical record information in the target medical record state to obtain a saturation value; And obtaining a candidate action set of the target medical record state, carrying out utility value calculation on each candidate action in the candidate action set to obtain an action utility value calculation result, and obtaining an information evaluation result according to the saturation value and the action utility value calculation result. Optionally, in the medical record dialogue data generating method, the calculating