CN-122025082-A - Atropine myopia prevention and control evidence-based dialogue method and system

CN122025082ACN 122025082 ACN122025082 ACN 122025082ACN-122025082-A

Abstract

The application discloses a method and a system for preventing and controlling evidence-based dialogue of atropine myopia, and belongs to the technical field of prevention and control data intellectualization. The method comprises the steps of carrying out semantic analysis on a received user problem, searching evidence-based evidence, carrying out searching enhancement according to a semantic analysis result, obtaining evidence-based evidence and sources and grades of evidence-based evidence from an atropine myopia prevention and control knowledge graph, generating an inference chain, combining the evidence-based evidence with a prompt word, guiding a model to generate inference chain content according to the steps of individual core feature analysis, matching with an atropine myopia prevention and control scheme, safety evaluation and scheme optimization through a thinking chain CoT, and generating answers according to the evidence-based evidence and the inference chain content. The application improves the matching property of each core entity in the output atropine myopia prevention and control scheme and the personalized demand, and improves the accuracy of the atropine myopia prevention and control scheme.

Inventors

LIU YU
Kan hongxing
HU JILI
YU FANGFANG

Assignees

安徽中医药大学

Dates

Publication Date: 20260512
Application Date: 20260114

Claims (13)

1. The atropine myopia prevention and control evidence-based dialogue method is characterized by comprising the following steps of: problem analysis, namely carrying out semantic analysis on received user problems; searching evidence-based evidence, namely searching and enhancing according to the result of semantic analysis, and acquiring evidence-based evidence and the source and grade of the evidence-based evidence from the atropine myopia prevention and control knowledge graph; Generating an inference chain, namely combining evidence-based evidence with the prompt word, and generating inference chain content through a thinking chain CoT guide model according to the steps of individual core feature analysis, matching with an atropine prevention and control scheme, security assessment and scheme optimization; Generating answers, namely generating the myopia prevention and control answers of different types of atropine according to evidence-based evidence and the content of an inference chain.
2. The atropine myopia prevention and control evidence-based dialogue method according to claim 1, wherein constructing the atropine myopia prevention and control knowledge graph comprises: extracting the relation between the entities, namely extracting key entities from each text block in the atropine myopia prevention and control database; the knowledge graph is simplified, namely the entity and the relation obtained by extraction are subjected to de-duplication treatment, the entities with the same name and type are combined, and the relation description of the same source-target entity is unified; the community detection and layering, namely carrying out layering segmentation on the simplified knowledge graph, and identifying core entity communities, wherein each core entity community represents a subdivision subject field; generating community summary, namely generating abstracts of the entity and the relation of each core entity community, and outputting community reports with different granularities; the core entity community comprises a concentration-curative effect community, a concentration-safety community, an individual core characteristic-medication adaptation community and a medication stopping opportunity judging community.
3. The atropine myopia prevention and control evidence-based conversational method of claim 2, wherein the entity and relationship entity types include pharmaceutical entities, disease entities, efficacy metrics, safety metrics, and individual core features; the relationship types in the entity and the relationship comprise concentration-curative effect association, concentration-safety association and individual core characteristic-concentration adaptation association.
4. The atropine myopia prevention and control evidence-based dialogue method according to claim 2, wherein constructing the atropine myopia prevention and control database comprises: deleting sensitive information and error information of the documents related to the prevention and control of the atropine myopia obtained by searching; Dividing a literature obtained by removing sensitive information into functional units of research background, method, result and conclusion based on semantic logic; cutting a document containing a function unit of 'research background-method-result-conclusion' into text blocks of 550-600 token, wherein each text block keeps a citation association relation with an original document; and converting the text blocks obtained by cutting and the corresponding reference association relations into a UTF-8 coding format, and storing the UTF-8 coding format into an atropine myopia prevention and control database.
5. The atropine myopia prevention and control evidence-based dialogue method according to claim 1, wherein the semantic parsing results comprise question types, core entities and user requirements; the problem types comprise clinical diagnosis and treatment decision type problems and myopia prevention and control knowledge and medication consultation type problems; the core entity comprises a drug entity, a disease entity, a curative effect index, a safety index and individual core characteristics; The user demand is the demand of the user on each core entity in the atropine prevention and control scheme.
6. The atropine myopia prevention and control evidence-based dialogue method according to claim 1, wherein the searching enhancement is performed according to the result of semantic analysis, and the evidence-based evidence and the source and grade of evidence-based evidence are obtained from the atropine myopia prevention and control knowledge graph, comprising: embedding and calculating each text block in the atropine myopia prevention and control database and a community summary report in the atropine myopia prevention and control knowledge graph to generate 768-dimensional semantic vectors; the local retrieval comprises the steps of locating corresponding entities and associated communities in a knowledge graph for resolving the entities in the obtained user problems, and retrieving text blocks and community summaries in the located communities; Global retrieval, namely, aiming at the cross-subject user problem, retrieving text blocks of all core entity communities related to core entities in the user problem, and screening text blocks with the front sequence through semantic similarity sequencing; And optimizing a search result, namely performing secondary sequencing on text blocks obtained by local search and global search according to the correlation strength between core entities calculated by a Node2Vec algorithm, and preferentially selecting the text blocks with the correlation strength reaching a preset threshold and the evidence grade meeting a preset standard as evidence-based evidence.
7. The atropine myopia prevention and control evidence-based dialogue method according to claim 1, wherein during training of the thought chain CoT guiding model, input features are constructed from three dimensions of individual core features, diagnosis and treatment scene parameters and evidence-based evidence constraints, and the input features comprise individual core features, diagnosis and treatment scene and decision target features and evidence-based evidence constraint features; And when the CoT reasoning engine model is trained, constructing output features from two dimensions of the step-by-step reasoning process and the final decision conclusion, wherein the output features comprise the step-by-step reasoning process features and the final decision conclusion features.
8. The atropine myopia prevention and control evidence-based dialogue method according to claim 1, wherein the step of combining evidence-based evidence with prompt words and generating inference chain content by using a thinking chain CoT guidance model according to the steps of individual core feature analysis-matching atropine prevention and control scheme-security assessment-scheme optimization comprises: Individual core feature analysis: analyzing individual core characteristics; Matching an atropine prevention and control scheme, namely matching the atropine prevention and control scheme with the individual core characteristics in an atropine myopia prevention and control knowledge graph, wherein the matching atropine prevention and control scheme comprises the concentration, the dosage and the dosage frequency of the atropine; Safety assessment of the atropine prevention and control regimen matched from the effects, duration and side effect elimination regimen The scheme is optimized, and the matching atropine prevention and control scheme is adjusted from the aspects of atropine concentration, drug frequency and combined treatment suggestion.
9. The method for a conversation of atropine myopia prevention and control evidence based on any one of claims 1-8, wherein the different types of atropine myopia prevention and control answers include professional version and popular version of atropine myopia prevention and control answers; The content of the professional atropine myopia prevention and control answer comprises quantitative evidence-based parameters, evidence sources and an individualized decision path; the content of the popular atropine myopia prevention and control answer adopts a living language to explain the core conclusion.
10. An atropine myopia prevention and control evidence-based dialogue system, which is characterized by comprising: the question module is used for analyzing the questions, namely carrying out semantic analysis on the received user questions; the retrieval enhancement engine module is used for retrieving evidence-based evidence, carrying out retrieval enhancement according to the result of semantic analysis, and acquiring evidence-based evidence and the source and grade of the evidence-based evidence from the atropine myopia prevention and control knowledge graph; the reasoning engine module is used for generating a reasoning chain, namely combining evidence-based evidence with prompt words, and generating reasoning chain contents according to the steps of individual core feature analysis, matching with an atropine prevention and control scheme, security assessment and scheme optimization through a thinking chain CoT guide model; and the answer generation and evaluation module is used for generating answers by forming different types of atropine myopia prevention and control answers according to evidence-based evidence and reasoning chain content.
11. The atropine myopia prevention and control evidence-based dialogue system according to claim 10, further comprising a knowledge-graph construction module for constructing and storing an atropine myopia prevention and control knowledge graph.
12. The atropine myopia prevention and control evidence-based dialogue system according to claim 11, wherein the knowledge graph construction module is further configured to update the atropine myopia prevention and control knowledge graph once a day.
13. The atropine myopia prevention and control evidence-based dialogue system according to claim 10 is further characterized by comprising a data preprocessing module, wherein the data preprocessing module is used for constructing and storing an atropine myopia prevention and control database.

Description

Atropine myopia prevention and control evidence-based dialogue method and system Technical Field The application belongs to the technical field of prevention and control data intellectualization, and particularly relates to an atropine myopia prevention and control evidence-based dialogue method and system. Background In the myopia prevention and control means, the low-concentration atropine eye drops are effective medicaments which are verified by a multi-center Random Control Test (RCT) at present, and the clinical optimization application of the low-concentration atropine eye drops has great strategic significance for inhibiting myopia high-incidence situations. However, the concentration riddle of low concentration atropine is always pending, the concentration range studied by laboratory research is 0.01% -0.5%, single 0.01% concentration is commonly adopted in clinical practice, the concentration has slight side effect but insufficient long-term prevention and control effect, the application value of other concentrations such as 0.02%, 0.025%, 0.05% and the like is still controversial, and the concentration can be optimally balanced among curative effect (myopia control rate), safety (photophobia, near-vision blurring, drug withdrawal rebound) and long-term tolerance, no unified standard is formed, and the maximum release of the prevention and control potential of atropine is directly restricted. In addition, the existing technical scheme of myopia prevention and control of low-concentration atropine depends on the experience of doctors or single guideline recommendation, and precise prevention and control are difficult to realize. The known traditional evidence-based medical method has obvious limitation in the field of prevention and control of low-concentration atropine myopia: ① The evidence integration and update lag, namely the atropine clinical test period is long, the evidence generation related to concentration optimization is slow, the latest research result is difficult to be quickly converted into clinical practice, and the guideline update lags behind the research progress by about 9 years; ② The individual decision transformation is difficult, the traditional evidence-based medicine emphasizes the optimal evidence of the group average effect, the complex situation of an individual patient cannot be directly adapted, a doctor needs to expend a great deal of effort to integrate massive research data and judge by combining with the specific situation of the patient, and the operation difficulty is extremely high; ③ The doctor-patient information is asymmetric, the basic doctor faces heavy knowledge integration burden, and the patients and families lack authoritative and understandable personalized medicine guiding channels, so that the treatment compliance is affected. With the rapid development of artificial intelligence technology, a Large Language Model (LLMs) based on a transducer architecture particularly shows strong capability in the aspects of natural language understanding, generation, reasoning and knowledge integration, and provides a revolutionary tool for breaking through the bottleneck of traditional evidence-based medicine. At present, the medical field has emerged a plurality of specialized large language models, such as Med-PaLM series which are good at clinical reasoning and text generation, med-Flamingo which is combined with information search to optimize evidence-based response, doctorGLM which is adaptive to Chinese context, and the like, and meanwhile, the search enhancement generation (RAG) technology effectively makes up the defect of the large language models in knowledge accuracy by introducing an external authoritative knowledge base, and lays a technical foundation for the research and development of intelligent auxiliary systems in the medical field. Disclosure of Invention In order to solve the problems, the application provides a method and a system for preventing and controlling evidence-based dialogue of atropine myopia. The first object of the present application is to provide an atropine myopia prevention and control evidence-based dialogue method, comprising: problem analysis, namely carrying out semantic analysis on received user problems; searching evidence-based evidence, namely searching and enhancing according to the result of semantic analysis, and acquiring evidence-based evidence and the source and grade of the evidence-based evidence from the atropine myopia prevention and control knowledge graph; Generating an inference chain, namely combining evidence-based evidence with the prompt word, and generating inference chain content through a thinking chain CoT guide model according to the steps of individual core feature analysis, matching with an atropine prevention and control scheme, security assessment and scheme optimization; Generating answers, namely generating the myopia prevention and control answers of different types of atropine according