Nursing knowledge management system based on RAGflow

Abstract

The invention provides a nursing knowledge management system based on RAGflow, and belongs to the technical field of nursing information systems. The system comprises a data access and analysis module which is used for carrying out structured processing on a nursing document and outputting an analysis result in a uniform format; the knowledge slicing and governance module is used for carrying out semantic slicing on the analysis result, generating metadata and eliminating repeated contents; the indexing and retrieval module is used for constructing and maintaining a hybrid index comprising a dense vector index and a sparse inverted index; the query understanding and task decomposition module is used for identifying a query problem, extracting entity data and decomposing the query problem into execution subtasks; the RAG generation and evidence alignment module is used for generating a question and answer result with an evidence snapshot; the evidence snapshot comprises a source document and a document content position; and the agent arrangement module is used for calling a preset task template and tool according to the nursing role and executing a preset target task. According to the method, the interpretability and the traceability of the question and answer result can be realized.

Description

Technical Field

[0001] This application relates to the field of nursing information systems, and more particularly to the RAGflow-based nursing knowledge management system. Background Technology

[0002] In related technologies, the traditional architecture of nursing information systems can work when the amount of information is small, but when faced with large-scale, heterogeneous, and unstructured nursing knowledge (such as tables, flowcharts, photocopies, path diagrams, etc.), the generated results often produce illusions (incorrect answers).

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this application is to propose a nursing knowledge management system based on RAGflow, which aims to achieve accurate retrieval and evidence backtracking of multi-source heterogeneous knowledge, thereby improving the level of intelligence in clinical and teaching scenarios.

[0005] To achieve the above objectives, one aspect of this application proposes a nursing knowledge management system based on RAGflow, the system including a data access and parsing module, a knowledge slicing and governance module, an indexing and retrieval module, a query understanding and task decomposition module, a RAG generation and evidence alignment module, and an agent orchestration module; The data access and parsing module is used to perform structured processing on nursing documents and output parsing results in a unified format. The knowledge slicing and governance module is connected to the data access and parsing module. The knowledge slicing and governance module is used to perform semantic slicing on the parsing results, generate metadata, and eliminate duplicate content. The indexing and retrieval module is connected to the knowledge slicing and governance module. The indexing and retrieval module is used to construct and maintain a hybrid index containing dense vector indexes and sparse inverted indexes. The query understanding and task decomposition module is used to identify the query question, extract entity data, and decompose the query question into execution sub-tasks. The RAG generation and evidence alignment module is connected to the indexing and retrieval module. The RAG generation and evidence alignment module is used to generate question-and-answer results with evidence snapshots. The evidence snapshots include the source document and the document content location. The intelligent agent orchestration module is connected to the query understanding and task decomposition module. The intelligent agent orchestration module is used to call preset task templates and tools according to the nursing role and execute preset target tasks.

[0006] In some embodiments, the data access and parsing module performs structured processing on nursing documents using OCR and layout parsing algorithms, identifies text blocks, table boundaries, and image elements, and outputs parsing results in a unified format; the layout parsing algorithm is used to convert unstructured data into a unified JSON object containing logical structure and spatial coordinates; the nursing documents include nursing guidelines, SOPs, ward round records, and patient education materials; the document structure of the nursing documents includes tables, flowcharts, and path diagrams.

[0007] In some embodiments, the knowledge slicing and governance module utilizes RAGFlow's document understanding module to perform semantic slicing on the parsing results, generate metadata, and execute version control and deduplication algorithms to eliminate duplicate and expired content. The metadata includes content fragment ID, topic tag, timestamp, source document ID, page coordinates, and version number. The version control and deduplication algorithm is used to attach version number and validity period tag to knowledge units, calculate the semantic similarity between new knowledge units and existing knowledge units to identify duplicates, and mark duplicate or expired knowledge units according to preset version age and source rules so that the knowledge units are given priority in the retrieval results.

[0008] In some embodiments, when performing a retrieval, the indexing and retrieval module uses a hybrid index comprising dense vector index and sparse inverted index, and employs a learned reordering model and a rule-based reordering strategy to comprehensively rank the retrieval results. The dense vector index is used for retrieval based on semantic vector similarity. The sparse inverted index is an index that calculates the matching score between query keywords and documents, and integrates word frequency and inverse document frequency information. The learned reordering model receives query information and candidate documents, and calculates the semantic similarity score and keyword matching score of candidate knowledge fragments. The rule-based reordering strategy includes prioritizing the institution's SOP and prioritizing the latest version.

[0009] In some embodiments, the query understanding and task decomposition module performs intent recognition and named entity recognition on the user-input query, extracts entity data, and decomposes the query into several execution sub-tasks; the entity data includes disease type, age group, and medication method; the query understanding and task decomposition module is also used to perform context weight adjustment based on the retrieval results of the index and retrieval module.

[0010] In some embodiments, the RAG generation and evidence alignment module includes an evidence gating mechanism; The evidence gating mechanism is used to perform consistency checks when generating question-and-answer results, so that the output results include conclusions and evidence pairs; the evidence pairs include the document ID, page coordinates, and version number of the source knowledge fragment.

[0011] In some embodiments, the preset nursing roles in the intelligent agent orchestration module include instructors, quality control personnel, and education personnel; the tools in the intelligent agent orchestration module include an education sheet generator, a dosage calculator, and a quality control checklist checker; the target tasks include education sheet generation, dosage calculation, and quality control checklist verification.

[0012] In some embodiments, the system further includes a compliance and audit module: The compliance and audit module is connected to the RAG generation and evidence alignment module and is used to record audit logs based on access control tags. The audit logs include the operator, time, input question, referenced document anchor, and version number.

[0013] In some embodiments, the compliance and auditing module further includes a data watermarking mechanism; the data watermarking mechanism is used to securely mark the output question-and-answer results so as to leave a trace of the process and comply with medical information security compliance requirements.

[0014] In some embodiments, the system further includes a user interaction and feedback module; The user interaction and feedback module is used to provide feedback channels to update the learning and reordering model weights and rule reordering strategies.

[0015] To achieve the above objectives, another aspect of this application proposes a nursing knowledge management method based on RAGflow, which is implemented through the aforementioned nursing knowledge management system based on RAGflow. The method includes: Nursing knowledge management is performed through modules for data access and parsing, knowledge slicing and governance, indexing and retrieval, query understanding and task decomposition, RAG generation and evidence alignment, and intelligent agent orchestration.

[0016] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0018] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0019] The embodiments of this application include at least the following beneficial effects: This application provides a nursing knowledge management system, method, electronic device, storage medium, and program product based on RAGflow. This application includes: a data access and parsing module for structuring nursing documents and outputting parsing results in a unified format; a knowledge slicing and governance module for semantically slicing the parsing results, generating metadata, and eliminating duplicate content; an indexing and retrieval module for constructing and maintaining a hybrid index containing dense vector indexes and sparse inverted indexes; a query understanding and task decomposition module for identifying query questions, extracting entity data, and decomposing the query question into execution sub-tasks; a RAG generation and evidence alignment module for generating question-and-answer results with evidence snapshots; the evidence snapshot includes the source document and the document content location; and an intelligent agent orchestration module for calling preset task templates and tools according to nursing roles to execute preset target tasks. This application can achieve the interpretability and traceability of question-and-answer results. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the RAGflow-based nursing knowledge management system provided in an embodiment of this application; Figure 2 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0023] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0024] 1) RAGflow, a workflow framework or platform for enhancing retrieval generation.

[0025] 2) RAG, Retrieval Enhanced Generation, is a technological paradigm that combines external knowledge retrieval with large language model generation.

[0026] 3) OCR, Optical Character Recognition, is a technology that converts text in images or scanned documents into computer-editable and processable text characters.

[0027] 4) JSON, JavaScript Object Notation, is a lightweight data exchange format.

[0028] 5) SOP, Standard Operating Procedure, in the field of medical and nursing care, specifically refers to a detailed and standardized set of steps to ensure the safe and standardized completion of a nursing procedure.

[0029] 6) TF-IDF, Term Frequency-Inverse Document Frequency, is a statistical method used to assess the importance of a word to a document in a set of documents or a corpus.

[0030] 7) BM25, a more advanced probabilistic relevance scoring function than TF-IDF, is used for information retrieval.

[0031] 8) FAQ, Frequently Asked Questions, is a pre-organized collection of questions and answers. In traditional systems, the default answer is usually returned by directly matching the user's question with the questions in the FAQ library.

[0032] 9) RBAC / ABAC, role-based access control / attribute-based access control, are two mainstream access control models.

[0033] 10) API, Application Programming Interface, is a set of predefined functions and rules that allow different software applications to communicate with each other and exchange data.

[0034] 11) NER, Named Entity Recognition, is a natural language processing technique used to identify and classify entities with specific meanings from text, such as names of people, places, and organizations.

[0035] 12) cross-encoder, a deep neural network model used for text pair matching.

[0036] This invention takes into account that while traditional architectures can work with small amounts of information, they suffer from the following problems when faced with large-scale, heterogeneous, and unstructured nursing knowledge (such as tables, flowcharts, photocopies, path diagrams, etc.): ① The search results rely on keyword matching and lack semantic understanding capabilities; ② It does not support structured parsing of complex documents (such as nursing pathway diagrams, tables, charts, etc.); ③ Lack of version control, traceable evidence, and explainable mechanisms; ④ The generated results often contain misleading information (incorrect answers) or lack cited sources; ⑤ Unable to provide differentiated task support for nursing roles (such as teaching, quality control, and education).

[0037] In view of this, this application provides a nursing knowledge management system based on RAGflow. This solution uses the RAGFlow framework and semantic understanding technology, deep document parsing and cross-document anchoring technology, evidence gating and versioned knowledge governance mechanism, and role-based intelligent agent orchestration mechanism to achieve accurate semantic retrieval of nursing knowledge, traceable and credible question answering, compliant knowledge management, and intelligent collaboration for clinical and teaching scenarios, thereby building an intelligent, interpretable, secure and compliant nursing knowledge management platform.

[0038] The RAGflow-based nursing knowledge management system provided in this application relates to the field of nursing information system technology. The RAGflow-based nursing knowledge management method provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, or desktop computer, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network.

[0039] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0040] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0041] Figure 1 This is an optional schematic diagram of the RAGflow-based nursing knowledge management system provided in the embodiments of this application. The RAGflow-based nursing knowledge management system includes a data access and parsing module, a knowledge slicing and governance module, an indexing and retrieval module, a query understanding and task decomposition module, a RAG generation and evidence alignment module, and an intelligent agent orchestration module. The data access and parsing module is used to perform structured processing on nursing documents and output parsing results in a unified format; The knowledge slicing and governance module is connected to the data access and parsing module. The knowledge slicing and governance module is used to perform semantic slicing on the parsing results, generate metadata, and eliminate duplicate content. The indexing and retrieval module is connected to the knowledge slicing and governance module. The indexing and retrieval module is used to build and maintain a hybrid index that includes dense vector indexes and sparse inverted indexes. The query understanding and task decomposition module is used to identify query questions, extract entity data, and decompose query questions into sub-tasks to be executed. The RAG generation and evidence alignment module is connected to the indexing and retrieval module. The RAG generation and evidence alignment module is used to generate question-and-answer results with evidence snapshots; the evidence snapshots include the source document and the document content location. The intelligent agent orchestration module is connected to the query understanding and task decomposition module. The intelligent agent orchestration module is used to call preset task templates and tools according to the nursing role and execute preset target tasks.

[0042] Optionally, in this embodiment, the data access and parsing module performs unified structured processing and parsing of multi-source heterogeneous nursing documents; the knowledge slicing and governance module performs semantic slicing, metadata annotation, and knowledge deduplication governance on the parsing results to form manageable knowledge units; the indexing and retrieval module establishes and maintains a hybrid index to support semantic and keyword fusion retrieval; when a user initiates a query, the query understanding and task decomposition module performs intent recognition and entity extraction, and decomposes complex queries into sub-tasks; then the RAG generation and evidence alignment module generates answers based on the retrieval results, and forces the attachment of traceable evidence snapshots through an evidence gating mechanism; the intelligent agent orchestration module calls the corresponding task templates and toolsets according to the identified nursing roles and task types to execute automated nursing tasks such as education and quality control, thereby completing the entire closed loop from document access to intelligent task execution.

[0043] In some embodiments, the data access and parsing module performs structured processing on nursing documents using OCR and layout parsing algorithms, identifies text blocks, table boundaries and image elements, and outputs parsing results in a unified format; the layout parsing algorithm is used to convert unstructured data into a unified JSON object containing logical structure and spatial coordinates; nursing documents include nursing guidelines, SOPs, ward round records, and patient education materials; the document structure of nursing documents includes tables, flowcharts and path diagrams.

[0044] Optionally, the data access and parsing module in this embodiment uses deep parsing technology to uniformly transform unstructured and diverse nursing documents (such as images and complex PDF formats) into machine-readable, understandable, and precisely located structured data, thereby laying a solid data foundation for subsequent knowledge slicing, precise retrieval, and evidence tracing. Recognizing text blocks involves identifying continuous text areas as logical units. For example, identifying a complete "operation steps" paragraph in a guide as a text block. Recognizing table boundaries refers to accurately detecting the borders of a table and defining the position of each cell. For example, recognizing the outer border and each row and column within a "Drug Incompatibilities Table." Recognizing image elements refers to recognizing non-text areas such as images, charts, stamps, and handwritten signatures in a document. Flowcharts and path diagrams are typical forms for describing nursing procedures, clinical pathways, and emergency procedures (such as a "CPR Path Diagram").

[0045] In some embodiments, the knowledge slicing and governance module utilizes RAGFlow's document understanding module to perform semantic slicing on the parsing results, generate metadata, and execute version control and deduplication algorithms to eliminate duplicate and expired content. The metadata includes content fragment ID, topic tag, timestamp, source document ID, page coordinates, and version number. The version control and deduplication algorithm is used to attach version number and validity period tag to knowledge units, calculate the semantic similarity between new knowledge units and existing knowledge units to identify duplicates, and mark duplicate or expired knowledge units according to preset version age and source rules so that knowledge units are given priority in the retrieval results.

[0046] Optionally, the knowledge slicing and governance module in this embodiment uses RAGFlow's document understanding module to perform semantic slicing on the parsing results. Based on the semantic coherence and topical completeness of the content, rather than a fixed number of words or punctuation, long documents are cut into independent knowledge fragments. For example, a "Pressure Ulcer Care Guide" can be cut into multiple semantically complete knowledge units such as "Risk Assessment (Braden Score)," "Prevention Measures (Relief and Turning)," and "Stage Treatment Plan (Stages I-IV)." This makes each slice a knowledge package that can independently answer questions. Metadata is structured information describing the attributes of knowledge units, including content fragment ID (unique identifier), topic tags (such as ["Pressure Ulcer," "Turning," "Q2h"], used for quick classification and filtering), timestamp (knowledge effective or database entry time), source document ID, page coordinates, and version number (e.g., source: SOP_DOC_007, page number: P5; combined with the coordinate information from Module 1, the specific cell can be located; the version number is used to distinguish between new and old). Version control in this embodiment can establish a new-old association of knowledge by attaching version numbers and expiration date tags to knowledge units. When a new version of the guide is imported, the system automatically identifies and updates the corresponding knowledge units, marking the old version as a "historical version" or "expired." The deduplication algorithm intelligently identifies entries with duplicate or highly similar content by calculating the semantic similarity between new and existing knowledge units (e.g., comparing their vector representations). For example, the descriptions of "hand hygiene" in this institute's SOP and those in the superior guide may be semantically highly similar; the system can associate or merge them. Priority marking automatically adjusts the weight of knowledge units in the search based on preset version and source rules (e.g., "latest version prioritized," "institute regulations prioritized"). For example, for the same operation, even if the text matching degree of the old guide is slightly higher, the system will prioritize the new version content due to the version number update, avoiding contamination of search results by old standards from the source.

[0047] In some embodiments, when performing a retrieval, the indexing and retrieval module uses a hybrid index that includes dense vector indexes and sparse inverted indexes, and employs a learned reordering model and a rule-based reordering strategy to comprehensively rank the retrieval results. The dense vector index is used for retrieval based on semantic vector similarity; the sparse inverted index is an index that calculates the matching score between query keywords and documents, and integrates term frequency and inverse document frequency information; the learned reordering model receives query information and candidate documents, and calculates the semantic similarity score and keyword matching score of candidate knowledge fragments; the rule-based reordering strategy includes prioritizing the institute's SOP and prioritizing the latest version.

[0048] Optionally, the indexing and retrieval module in this embodiment of the application achieves intelligent and compliant retrieval capabilities for the nursing knowledge base through a "hybrid retrieval + intelligent reordering" technical architecture. Dense vector indexing converts text into high-dimensional semantic vectors, enabling meaning-based retrieval. For example, when a nurse queries "IV drip," the index can match semantically similar but literally different terms in the knowledge base, such as "intravenous infusion" and "intravenous transfusion," solving the problem of synonymous terms and diverse expressions. Sparse inverted indexing establishes an inverted table of "keyword -> document" and uses algorithms such as BM25 to calculate matching scores, achieving precise matching based on "words." For example, for the exact drug name "cefotaxime sodium for injection" or the professional abbreviation "CVP (central venous pressure)," it can quickly and accurately recall documents containing these precise terms. The learning reordering model can use deep models such as Cross-Encoder, which performs deep interactive calculations by simultaneously encoding queries and candidate documents to obtain a more accurate relevance score. It is better at understanding complex semantic relationships than simple vector cosine similarity. Prioritizing our hospital's SOPs means that when searching for similar general guidelines and our hospital's SOPs, our hospital's SOPs will be prioritized in the ranking. For example, regarding "medical waste classification," even if the guidelines are more detailed, our hospital's specific classification list will be displayed first to ensure compliance. Prioritizing the latest version means that when searching for information on the same topic, the latest version will be prioritized in the ranking. For example, regarding "CPR compression depth," the 2023 guidelines recommend "5-6 cm," while the older version recommends "at least 5 cm." The system will prioritize the newer version to ensure the information is up-to-date.

[0049] In some embodiments, the query understanding and task decomposition module performs intent recognition and named entity recognition on the user-input query, extracts entity data, and decomposes the query into several execution sub-tasks; the entity data includes disease type, age group, and medication method; the query understanding and task decomposition module is also used to perform context weight adjustment based on the retrieval results of the indexing and retrieval module.

[0050] Optionally, the query understanding and task decomposition module in this embodiment of the application, through "deep understanding of user intent + dynamic planning of task flow," transforms the user's vague and complex natural language requests into structured instructions and task sequences that the machine can accurately execute. It can also intelligently optimize based on the retrieved context information, thereby enabling the system to leap from a "passive question-answering machine" to an "active task collaborator," improving the efficiency and intelligence level of human-computer interaction. Intent recognition can determine the fundamental purpose of the user's question. For example, "Tell me how to administer intravenous fluids to a heart failure patient" is recognized as "Intent: Operation Query"; "Make a discharge education form for a diabetic patient" is recognized as "Intent: Task Execution (Education)." Named entity recognition can extract key structured information fields from the query text. For example, it can extract the disease type: such as "Disease: Heart Failure" from "How to use diuretics for heart failure patients"; extract the age group: such as "Age group: Newborn" from "What should be paid attention to in the care of neonatal pneumonia"; and extract the method of administration: such as "Method of administration: Intravenous injection, oral administration" from "Is this medicine administered intravenously or orally." Decomposing a query into several sub-tasks involves breaking down a complex, multi-step request into a series of ordered, automatically or semi-automatically executed atomic operations. For example, a user querying "develop a post-operative rehabilitation plan for an elderly patient who has just undergone knee replacement surgery" can be broken down into sub-task 1: retrieving general rehabilitation guidelines for "post-knee replacement surgery"; sub-task 2: adjusting the rehabilitation intensity and precautions based on the characteristics of "elderly patient"; sub-task 3: retrieving and integrating knowledge from key modules such as "pain management," "prevention of deep vein thrombosis," and "functional exercises"; and sub-task 4: using a "educational sheet generator" tool to format the above content into a personalized plan. Context weight adjustment is a dynamic feedback optimization mechanism. After initially understanding the query, the module uses the initially retrieved knowledge fragments as "context" to correct or deepen its understanding of the original query. By analyzing the distribution and relevance of the search results, it dynamically adjusts the importance of different entities in the query. For example, when searching for "diet and exercise for patients with hypertension," the initial search reveals that guidelines on "hypertension exercise" heavily emphasize "heart rate control" and "avoiding isometric exercise." Therefore, the weighting is adjusted: the system will automatically increase the weight of the "exercise" related parts of the query, and when generating answers or breaking down tasks, it will focus more on presenting sub-tasks and content related to "exercise safety," making the output more targeted and in-depth.

[0051] In some embodiments, the RAG generation and evidence alignment module includes an evidence gating mechanism; Evidence gating mechanisms are used to perform consistency checks when generating question-and-answer results, so that the output includes conclusions and evidence pairs; the evidence pairs include the document ID, page coordinates, and version number of the source knowledge fragment.

[0052] Optionally, the RAG generation and evidence alignment module in this application embodiment performs consistency checks when generating question-and-answer results through an evidence gating mechanism. The consistency check compares the semantic consistency between the text (or attention focus) to be generated by the model and the cited retrieval fragment (evidence) to prevent the model from distorting facts. For example, if the retrieved evidence is "a positive penicillin skin test is characterized by a local erythema diameter greater than 1 cm," the correct generation (passing the check) is: the model generates "a erythema in the skin test area exceeding 1 cm is positive." The incorrect generation (blocked by the gating mechanism) is: if the model attempts to generate "a erythema greater than 0.5 cm is positive," the gating mechanism will prevent this conclusion from being output due to semantic contradiction, or force it to revert to a statement consistent with the evidence. Conclusion and evidence pair: refers to the final output format of the system being a bundle of "conclusion paragraph + accompanying standardized citation information." For example, conclusion: For patients in shock, central venous pressure (CVP) should be monitored and recorded hourly; evidence pair: [Based on: Critical Care Operating Procedures (V4.2), page 15].

[0053] In some embodiments, the preset nursing roles in the intelligent agent orchestration module include instructors, quality control personnel, and education personnel; the tools in the intelligent agent orchestration module include an education sheet generator, a dosage calculator, and a quality control checklist checker; the target tasks include education sheet generation, dosage calculation, and quality control checklist verification.

[0054] Optionally, the intelligent agent orchestration module in this embodiment integrates "role-based intelligent agents and dedicated toolsets" to transform general knowledge question-and-answer capabilities into automated productivity tools that fit specific nursing job positions and solve pain points in actual workflows. This improves work efficiency, standardization, and intelligence in clinical, teaching, and quality control scenarios, achieving a paradigm shift from a "knowledge query platform" to a "workflow intelligent partner." Personalized workflows are achieved through preset nursing roles (instructors, quality control personnel, and health educators). Instructors need to quickly generate teaching cases, training plans, and assessment points; quality control personnel need to efficiently complete standardized checks, identify problems, and generate reports; and health educators need to quickly produce personalized health education materials tailored to different diseases and patient characteristics. This embodiment performs atomic operations by calling preset tools (health education sheet generator, dosage calculator, and quality control checklist checker). The health education sheet generator automatically generates clearly structured and easy-to-understand health education materials by combining patient information and disease knowledge and applying preset graphic templates. For example, by entering "Patient: Zhang, Diagnosis: Type 2 Diabetes," the tool automatically generates a PDF educational leaflet containing dietary charts, exercise recommendations, and blood glucose monitoring methods. The dosage calculator automatically performs complex medication calculations using built-in pharmaceutical calculation rules and formulas. For example, by entering "Doctor's order: Ceftriaxone 2g q12h, Specification: 1g / bottle, Solvent: 0.9% NS 100ml," the tool automatically calculates and outputs: "Two bottles are required, dissolved together in 100ml of normal saline, final concentration 20mg / ml." The quality control checklist checker achieves digital verification by loading structured quality control items and comparing them with actual data (or manual selection). For example, based on the "Intravenous Infusion Safety Checklist," it presents items such as "Whether double-checked" and "Whether the infusion label is complete," supporting online selection and problem recording.

[0055] In some embodiments, the RAGflow-based nursing knowledge management system also includes a compliance and audit module: The compliance and audit module, connected to the RAG generation and evidence alignment module, is used to record audit logs based on access control tags. The audit logs include the operator, time, input question, referenced document anchor, and version number.

[0056] Optionally, this application embodiment achieves access control and behavior traceability by recording audit logs based on access control tags. An access control tag is a permission identifier, typically generated based on RBAC (Role-Based Access Control) or ABAC (Attribute-Based Access Control) models. It defines "who," "under what conditions," and "what operation" can perform on "what resources." For example, the tag could be set as {Role: Nurse, Department: Cardiology, Permission: Query, Resource Scope: Departmental SOPs and General Guidelines}, meaning a respiratory nurse would not be able to query the cardiology-specific procedures. The audit log is an automatically generated, chronologically recorded, and tamper-proof record of operations.

[0057] In some embodiments, the compliance and audit module also includes a data watermarking mechanism; the data watermarking mechanism is used to securely mark the output question and answer results so as to leave a trace of the process and comply with medical information security compliance requirements.

[0058] Optionally, the data watermarking mechanism in this application embeds specific identification information (watermark) into a digital carrier (such as text or a document) without affecting the normal use and appearance of the carrier. Security marking involves subtly modifying the encoding of the output content (such as generated answer text or educational leaflet PDF) to embed key audit information (such as operator ID, timestamp, or session number). For example, performing text watermarking involves embedding information covertly by adjusting the Unicode encoding of specific characters in a sentence, adjusting space width, or replacing synonyms. Performing document watermarking involves embedding information in the generated PDF or Word document using invisible layers, metadata, or fine-tuning font attributes.

[0059] In some embodiments, the RAGflow-based nursing knowledge management system also includes a user interaction and feedback module; The user interaction and feedback module provides a feedback channel to update the learning and reordering model weights and rule reordering strategies.

[0060] Optionally, the feedback channel provided by the user interaction and feedback module in this application embodiment can be an interactive function designed on the product interface for users to evaluate the system output. For example, explicit feedback: setting a "helpful / not helpful" button next to each question-and-answer result or search result, or a five-star rating for "accuracy" or "relevance". The feedback channel can also be implicit feedback: user behavior data automatically recorded in the background, such as: the user ultimately clicked the third result in the ranking instead of the first result; the user modified the search keywords and searched again; the user stayed for a long time or quickly closed the page after reading an answer. Through these channels, the user's subjective judgment and real behavior are transformed into quantifiable training data.

[0061] This application addresses the following issues in the relevant technology: (1) Lack of semantic understanding and intelligent retrieval capabilities: Traditional systems rely on keyword matching (TF-IDF, BM25) for retrieval, which fails to understand the semantic relationships or contextual meanings of nursing terms. For example, "intravenous drip" and "IV drip" cannot be identified as synonyms, leading to insufficient recall and incomplete answer fragments.

[0062] This invention improves upon existing technologies by utilizing RAGFlow's semantic vector retrieval and knowledge slicing techniques, combined with entity standardization and multi-vector hybrid rearrangement, to achieve semantic-level knowledge recall and matching. It can understand concepts such as "administration method" and "disease type / age group" at the semantic level, thereby significantly improving retrieval accuracy and recall.

[0063] (2) Unable to handle complex document structures and cross-document references: The nursing knowledge base for related technologies is primarily designed for plain text or simple structured documents, making it difficult to parse complex documents such as tables, flowcharts, and path diagrams. This results in the system being unable to extract key knowledge points from table cells or path nodes.

[0064] Improvements of this invention: This invention utilizes "deep document parsing + cross-document anchor point generation" technology to slice complex document content into indexable knowledge units and establish cross-document anchor points (including page numbers, table cells, and figure caption coordinates). This supports accurate retrieval and evidence backtracking of multi-source heterogeneous knowledge.

[0065] (3) The answer lacks traceability and explainability of evidence: Traditional systems only provide text summaries or document links when returning answers, lacking citation information, page numbers, and version information, making it impossible for users to verify the source and accuracy of the answers.

[0066] Improvements of this invention: This invention introduces an evidence alignment mechanism and "evidence gating" control during the RAG generation stage, forcing the generation of answers to include corresponding evidence snapshots, document IDs, page numbers, and version numbers, thereby achieving interpretability and traceability of the question-and-answer results. This improvement is particularly important in nursing quality control and teaching scenarios.

[0067] (4) Lagging version updates and knowledge governance: The system based on the relevant technology cannot automatically identify and replace updated nursing guidelines and regulations, which poses a risk of relying on outdated knowledge.

[0068] Improvements of this invention: This system introduces version identification and validity period management mechanisms into the knowledge governance module to achieve deduplication, disambiguation, and timeliness control of knowledge units. Simultaneously, a rule-based reordering strategy prioritizes the presentation of the "latest version and the institute's regulations," ensuring the compliance and timeliness of the output results.

[0069] (5) Lack of role-based intelligent collaboration in nursing care: The related technologies are designed for a single search function and cannot provide differentiated workflows based on nursing roles (such as teaching, quality control, and health education), resulting in a low level of intelligence.

[0070] Improvements of this invention: This invention introduces an intelligent agent orchestration module, which provides task templates and tool calls (such as education sheet generator, dosage calculator, and quality control checklist checker) for different nursing roles, enabling multi-step reasoning and task collaboration, and significantly improving the level of intelligence in clinical and teaching scenarios.

[0071] (6) Lack of security compliance and auditing mechanisms: Most traditional systems lack access control and operation auditing functions, posing risks of information leakage and difficulty in tracing responsibility.

[0072] Improvements of this invention: The system incorporates access control tags, data watermarking, and audit log mechanisms to record the operator, time, and evidence anchor points, thereby achieving secure and controllable nursing knowledge and full-process traceability, in compliance with medical information security requirements.

[0073] Related technologies generally suffer from insufficient semantic understanding, weak document structure parsing capabilities, uninterpretable answers, lagging version management, and a lack of intelligent collaboration. This invention systematically improves these deficiencies through key technologies such as RAG-enhanced retrieval, cross-document anchoring, evidence alignment, and intelligent agent orchestration, thereby enabling the intelligent, reliable, and localized deployment capabilities of a nursing knowledge management platform.

[0074] Compared to related nursing knowledge management systems, this invention adopts an integrated technical path at the architectural level, encompassing "data understanding—knowledge governance—hybrid retrieval—evidence alignment—agent orchestration—compliance auditing." While traditional systems focus on static document retrieval, this invention utilizes the RAGFlow retrieval enhancement generation mechanism and agent collaboration mechanism to achieve a closed-loop process encompassing semantic understanding, automated knowledge reasoning, and traceable question answering. See Table 1 for a comparison of the differences.

[0075] Table 1. Comparison of Differences

[0076] As an optional implementation method, embodiments of this application include: (1) Overall system structure: The overall structure of the nursing knowledge management system includes: ① Data access and parsing module; ② Knowledge Slicing and Governance Module; ③ Indexing and retrieval module; ④ Query understanding and task decomposition module; ⑤ RAG generation and evidence alignment module; ⑥ Intelligent agent orchestration module; ⑦ Compliance and Audit Module; ⑧ User interaction and feedback module.

[0077] Module connection relationship description: The ① data access and parsing module and the ② knowledge slicing and governance module are directly connected via data flow for document input and structuring. After parsing, a structured data stream (including text, tables, and image annotations) is output.

[0078] ② The knowledge governance module and ③ the index module are bidirectionally connected, which is used to refresh the index in real time when knowledge is updated or the version changes.

[0079] ③ The indexing and retrieval module outputs candidate fragments to the ⑤ RAG generation module, and simultaneously passes the retrieval score to the ④ task decomposition module for context weight adjustment.

[0080] The ④ query understanding and task decomposition module communicates bidirectionally with the ⑥ intelligent agent orchestration module to decompose complex nursing problems into executable task flows.

[0081] The ⑤RAG generation and evidence alignment module is connected to the ⑦compliance audit module to generate Q&A results with evidence snapshots and trace logs.

[0082] ⑧ The user interaction module receives results unidirectionally from other modules and provides a human feedback channel to update model weights and reordering strategies.

[0083] (2) Step-by-step instructions: S1. Data Access and Parsing: The system accesses nursing-related documents (guidelines, SOPs, ward round records, patient education materials) via API or batch import. It uses OCR and layout parsing algorithms to recognize text blocks, table boundaries, and image elements, outputting a standardized JSON object.

[0084] S2. Knowledge Slicing and Governance: The parsing results are semantically sliced ​​using RAGFlow's document understanding module. Each slice contains metadata such as content fragment ID, topic tag, timestamp, source document ID, page coordinates, and version number. Duplicate and outdated content is eliminated through version control and deduplication algorithms.

[0085] S3. Establish a hybrid index and rule-based reordering model: The system simultaneously constructs a dense vector index (semantic layer) and a sparse inverted index (keyword layer) in the database, and uses a lightweight reordering model (such as cross-encoder) to sort by confidence and source priority. The system supports rule weights for "priority to our institute's SOP" and "priority to the latest version".

[0086] S4. User Query Understanding and Task Decomposition: When a user enters a question, the system first performs intent recognition and named entity recognition (NER) to extract key entities such as disease type, age group, and medication method, and then generates a query plan. For complex questions, the system breaks them down into several executable subtasks.

[0087] S5. Hybrid retrieval recall and candidate rearrangement: The system retrieves Top-K candidate fragments from the vector index and sparse index respectively, and after fusion, generates the optimal result set according to the rearrangement model and business rules.

[0088] S6. RAG generation and evidence alignment: The RAG generation module receives candidate results and executes an "evidence gating mechanism" to ensure that each conclusion is accompanied by its source evidence (including document ID, page number, and version number). The output includes a "conclusion-evidence pair".

[0089] S7. Agent Task Execution and Result Integration: The agent module executes subsequent tasks based on the role template, such as generating educational leaflets, calculating usage, and quality control verification. The results of each task, along with the RAG Q&A results, are output to the front-end interface.

[0090] S8. Audit Records and Feedback Updates: The system automatically generates audit logs, including: user ID, operation time, input question, referenced document anchor, version number, and returned content. These logs are used for compliance tracking and model iteration optimization.

[0091] The technical effects achieved by the embodiments of this application include: ①Semantic-level precise retrieval and cross-document reasoning; ② Explainable answer generation and evidence tracing; ③ Multi-step task automation for nursing roles; ④ Versioned knowledge governance and local security compliance.

[0092] The key technical points of this invention are shown in Table 2.

[0093] Table 2. Key Technologies

[0094] This application also provides a RAGflow-based nursing knowledge management method, which is implemented using the aforementioned RAGflow-based nursing knowledge management system. The method includes: Nursing knowledge management is performed through modules for data access and parsing, knowledge slicing and governance, indexing and retrieval, query understanding and task decomposition, RAG generation and evidence alignment, and intelligent agent orchestration.

[0095] It is understood that the content of the above method embodiments is applicable to this method embodiment. The specific functions implemented in this method embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0096] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0097] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0098] Please see Figure 2 , Figure 2 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 201 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 202 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 202 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 202 and is called and executed by the processor 201 using the methods described in the embodiments of this application. Input / output interface 203 is used to implement information input and output; The communication interface 204 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 205 transmits information between various components of the device (e.g., processor 201, memory 202, input / output interface 203, and communication interface 204); The processor 201, memory 202, input / output interface 203 and communication interface 204 are connected to each other within the device via bus 205.

[0099] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0100] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0101] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0102] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0103] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0104] The RAGflow-based nursing knowledge management system, method, electronic device, storage medium, and program products provided in this application construct a structured knowledge base through "deep document parsing and semantic slicing," achieve accurate knowledge retrieval through "hybrid indexing and rule rearrangement," ensure the traceability and credibility of answers through "evidence gating and RAG generation," and drive the automation of nursing tasks through "role-based intelligent agent orchestration." Ultimately, it realizes intelligent management of nursing knowledge, credible question answering, compliant governance, and scenario-based collaboration, providing accurate, efficient, and secure decision support and workflow empowerment for clinical nursing, teaching and training, and quality control.

[0105] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0106] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0107] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0108] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0109] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0110] It should be understood that in this application, "at least one (item)" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0112] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0113] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0114] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A nursing knowledge management system based on RAGflow, characterized in that, The system includes a data access and parsing module, a knowledge slicing and governance module, an indexing and retrieval module, a query understanding and task decomposition module, a RAG generation and evidence alignment module, and an agent orchestration module. The data access and parsing module is used to perform structured processing on nursing documents and output parsing results in a unified format. The knowledge slicing and governance module is connected to the data access and parsing module. The knowledge slicing and governance module is used to perform semantic slicing on the parsing results, generate metadata, and eliminate duplicate content. The indexing and retrieval module is connected to the knowledge slicing and governance module. The indexing and retrieval module is used to construct and maintain a hybrid index containing dense vector indexes and sparse inverted indexes. The query understanding and task decomposition module is used to identify the query question, extract entity data, and decompose the query question into execution sub-tasks. The RAG generation and evidence alignment module is connected to the indexing and retrieval module. The RAG generation and evidence alignment module is used to generate question-and-answer results with evidence snapshots. The evidence snapshots include the source document and the document content location. The intelligent agent orchestration module is connected to the query understanding and task decomposition module. The intelligent agent orchestration module is used to call preset task templates and tools according to the nursing role and execute preset target tasks.

2. The system according to claim 1, characterized in that, The data access and parsing module performs structured processing on nursing documents using OCR and layout parsing algorithms, recognizing text blocks, table boundaries, and image elements, and outputting parsing results in a unified format. The layout parsing algorithm is used to convert unstructured data into a unified JSON object containing logical structure and spatial coordinates. The nursing documents include nursing guidelines, SOPs, ward round records, and patient education materials. The document structure of the nursing documents includes tables, flowcharts, and path diagrams.

3. The system according to claim 1, characterized in that, The knowledge slicing and governance module utilizes RAGFlow's document understanding module to perform semantic slicing on the parsing results, generate metadata, and execute version control and deduplication algorithms to eliminate duplicate and expired content. The metadata includes content fragment ID, topic tag, timestamp, source document ID, page coordinates, and version number. The version control and deduplication algorithm is used to attach version number and validity period tag to knowledge units, calculate the semantic similarity between new knowledge units and existing knowledge units to identify duplicates, and mark duplicate or expired knowledge units according to preset version age and source rules so that the knowledge units are given priority in the search results.

4. The system according to claim 1, characterized in that, When performing a search, the indexing and retrieval module uses a hybrid index that includes dense vector indexes and sparse inverted indexes, and employs a learned reordering model and rule-based reordering strategy to comprehensively rank the search results; the dense vector index is used for retrieval based on semantic vector similarity. The A sparse inverted index is an index that calculates the matching score between query keywords and documents, taking into account term frequency and inverse document frequency information; The learning-based reordering model receives query information and candidate documents, and calculates the semantic similarity score and keyword matching score of the candidate knowledge fragments; the rule-based reordering strategy includes prioritizing the institute's SOP and prioritizing the latest version.

5. The system according to claim 1, characterized in that, The query understanding and task decomposition module performs intent recognition and named entity recognition on the user-input query, extracts entity data, and decomposes the query into several execution sub-tasks; the entity data includes disease type, age group, and medication method; the query understanding and task decomposition module is also used to perform context weight adjustment based on the retrieval results of the index and retrieval module.

6. The system according to claim 1, characterized in that, The RAG generation and evidence alignment module includes an evidence gating mechanism; The evidence gating mechanism is used to perform consistency checks when generating question-and-answer results, so that the output results include conclusions and evidence pairs; the evidence pairs include the document ID, page coordinates, and version number of the source knowledge fragment.

7. The system according to claim 1, characterized in that, The preset nursing roles in the intelligent agent orchestration module include instructors, quality control personnel, and education personnel; the tools in the intelligent agent orchestration module include an education sheet generator, a dosage calculator, and a quality control checklist checker; the target tasks include education sheet generation, dosage calculation, and quality control checklist verification.

8. The system according to claim 1, characterized in that, The system also includes a compliance and audit module: The compliance and audit module is connected to the RAG generation and evidence alignment module and is used to record audit logs based on access control tags. The audit logs include the operator, time, input question, referenced document anchor, and version number.

9. The system according to claim 8, characterized in that, The compliance and auditing module also includes a data watermarking mechanism; the data watermarking mechanism is used to securely mark the output question and answer results so that the process is traceable and complies with medical information security compliance requirements.

10. The system according to claim 1, characterized in that, The system also includes a user interaction and feedback module; The user interaction and feedback module is used to provide feedback channels to update the learning and reordering model weights and rule reordering strategies.

Images