Intelligent management method and system applied to medical education platform

Abstract

The invention relates to the field of medical education management, in particular to an intelligent management method and system applied to a medical education platform, and the method comprises the following steps: collecting multi-source data related to medical education, and forming a structured initialization data set through data cleaning processing; based on the structured initialization data set, constructing a four-dimensional intelligent adaptation model, establishing a mapping relation, and generating a personalized permission configuration result and a data view; relevant data of students, departments and a rotation template are collected, intelligent linkage of rotation shift arrangement, teaching activities and assessment evaluation is achieved through a fusion algorithm, and service data are output; generating a personalized learning resource pushing result through feature extraction and model training; and extracting related data of trainees' employment, verifying employment conditions and outputting a standard state result. According to the invention, by constructing the four-dimensional intelligent adaptation model and fusing multi-algorithm and full-process data processing, intelligent linkage and data closed loop of medical education services are realized.

Description

Technical Field

[0001] This invention relates to the field of medical education management, and in particular to an intelligent management method and system applied to a medical education platform. Background Technology

[0002] Currently, various digital management systems have emerged in the field of medical education management, covering basic functional modules such as user file maintenance, rotation scheduling, teaching activity organization, assessment and evaluation implementation, and graduation review. Some systems introduce single algorithms for optimization of specific processes, such as basic scheduling algorithms, fixed permission allocation rules, and simple test paper generation logic. These systems can complete basic management operations in the medical education process, providing preliminary digital support for scenarios such as standardized training of resident physicians and training of college interns, and to a certain extent replacing the traditional manual ledger management model.

[0003] However, existing technologies still have many limitations. Data between different business modules is isolated, lacking an effective data flow and linkage mechanism, and failing to form a closed loop of "data collection-processing-application-feedback," resulting in the inability to fully release the value of data. The adaptation of role permissions and data perspectives is relatively crude, only achieving basic role permission division, which is difficult to meet the personalized permission configuration and data view requirements of multiple levels and positions such as hospitals / universities, bases, departments, instructors, and students. Personalized training lacks data-driven support, and cannot accurately push suitable learning resources and assessment tasks based on full-dimensional data such as students' rotation progress, assessment results, and learning preferences. There are many manual intervention links in the core management process. Key links such as rotation scheduling conflict detection, supervision anomaly identification, and graduation condition verification rely on manual operation, which is not only inefficient but also prone to errors. Moreover, it lacks an algorithmic collaborative optimization mechanism, making it difficult to adapt to the medical education and training needs that combine standardization and personalization in multiple scenarios. Summary of the Invention

[0004] To address the technical deficiencies in the background technology, this invention proposes an intelligent management method and system for medical education platforms, which solves the aforementioned technical problems and meets practical needs. The specific technical solution is as follows: An intelligent management method for medical education platforms includes the following steps: Collect multi-source data related to medical education, and form a structured initial dataset through data cleaning and processing; Based on a structured initialization dataset, a four-dimensional intelligent adaptation model is constructed for roles, departments, bases, and levels. The mapping relationship between users, roles, permissions, and data is established, and personalized permission configuration results and data views are generated. Based on the permission configuration results, relevant data on students, departments and rotation templates are collected. Through the fusion algorithm, intelligent linkage between rotation scheduling, teaching activities and assessment is realized, and business data of each link is output. Based on business data, personalized learning resource push results are generated through feature extraction and model training. At the same time, a supervision rule base is built based on the outline standard, and anomaly detection results and rectification guidance data are output. Based on business data, extract relevant data on student graduation, verify graduation conditions and output the achievement status result, match the certificate template to generate graduation certificate and push it to the student's end.

[0005] Furthermore, the multi-source data includes user basic data, role attribute data, basic information data of departments and bases, national medical education syllabus standard data, and historical business data. The specific steps for constructing the four-dimensional intelligent adaptation model are as follows: Based on user basic data and role attribute data, the mapping relationship between users, roles, permissions and data is established through the RBAC extended algorithm, the basic permission range corresponding to each role is clarified, and the basic mapping rules are output. The user's basic data is used as the input feature, and the preset role labels are used as the output result. The input is fed into the preset multi-label classification model, which automatically assigns initial roles to users and outputs the initial role assignment results. Collect user historical operation data, including menu access frequency, button click count, and data viewing range. Combine this with basic mapping rules to dynamically adjust the permission priority of each role. Prioritize permission items with access frequency higher than a preset threshold and output the permission priority adjustment results. Collect data on the user's department, base, and level; combine the basic mapping rules and permission priority adjustment results to clarify the specific menus, buttons, and data access scope that the user can operate; and generate personalized permission configuration results. Collect historical data access records, combine them with user role attribute data to analyze user data needs and preferences, and generate personalized data views.

[0006] Furthermore, the specific process for implementing the aforementioned shift rotation schedule is as follows: Based on the permission configuration results, basic student information, department constraint information, and rotation template information are collected and integrated into basic scheduling data. Input basic scheduling data, encode the trainee rotation plan through intelligent scheduling algorithm, generate a preset number of initial scheduling scheme populations by combining department constraint information, and output the optimal intelligent template scheduling scheme through iterative optimization operation; Based on the scheduling plan, preview data is generated from multiple dimensions. When the number of scheduled personnel exceeds the department's capacity, an anomaly is marked and the scheduling preview results are output. The system collects students' entry status, manual registration data, attendance records and assessment results in real time during the rotation process. It tracks the rotation progress through a preset monitoring algorithm and generates an early warning message when the preset early warning trigger conditions are met. The message is then pushed to the corresponding instructors and management users, and the warning data is output. Based on early warning data and rotation adjustment needs, the professional matching status, teaching load and historical evaluation of teaching staff are extracted from the basic scheduling dataset. The optimal teaching staff matching result is solved by optimization algorithm. Combined with the overall arrangement of the current rotation plan and departmental constraint information, the optimal rotation adjustment path is determined by path planning algorithm. The teaching staff matching result and rotation path adjustment plan are output. The system collects data on students' completion of manuals, attendance compliance, assessment scores, and peer review results during their rotations. It then inputs this data into a multi-dimensional weighted scoring algorithm to calculate the overall academic achievement results and generates various retake adaptation plans for students who do not meet the standards.

[0007] Furthermore, the intelligent linkage process between the teaching activities and the assessment is as follows: Based on the permission configuration results, the trainees' current rotation departments and specialties data, and combined with the knowledge point tags of the National Medical Education Syllabus, the tags are matched with the teaching activity resource library to select suitable teaching activities and corresponding resources, and output suitable activity data. The task binding algorithm is used to associate the adapted activity data with the preset assessment template library, generate binding relationships, clarify the corresponding assessment type and knowledge point scope, and output linked task data. After the teaching activity ends, the bound assessment task is automatically released, the assessment data submitted by the students is collected, encrypted and stored, and the assessment data is output. Based on the assessment data, the ratio of the student's assessment score to the full score of the corresponding knowledge point is calculated to identify the knowledge points that have not been mastered. At the same time, the correlation of teaching activities is analyzed to output mastery data and related results. Based on the mastery data and related results, optimize subsequent collaborative tasks and output the task optimization results.

[0008] Furthermore, the calculation formula for the multi-dimensional weighted scoring algorithm is as follows: , in, This indicates the overall score for passing the exam. Let the weight coefficient of the i-th evaluation dimension satisfy the following condition: =1, This represents the score for the i-th evaluation dimension. The evaluation dimensions include the completion rate of the rotation manual, attendance compliance rate, theoretical assessment score, skills assessment score, and peer assessment score. The weight coefficients of each dimension are preset according to the national medical education syllabus standards and the hospital's personalized training requirements, and can be manually adjusted by the administrator through the platform.

[0009] Furthermore, the business data includes rotation management-related data, teaching activity-related data, and assessment and evaluation-related data. The specific steps for generating personalized learning resource push results are as follows: Based on business data, extract student majors, enrollment time, weak knowledge points, historical learning resource type preferences, learning duration distribution, and assessment score distribution, and output feature vectors; The feature vector is used as input, and the learning completion rate of the learner on the learning resources is used as the output label. The training set and the test set are divided. The learning preference model is constructed by logistic regression algorithm, and the model parameters are optimized by gradient descent algorithm to minimize the square error between the model prediction value and the actual value. When the accuracy of the model on the test set is higher than the threshold, the training is stopped and the trained learning preference model is output. Collect resource data from the learning resource library, filter relevant resources based on students' major data, and output the filtered resource data. Based on feature vectors and filtered resource data, combined with a learning preference model, the matching degree between learners and each resource is calculated using a matching degree calculation formula. The resources are sorted from high to low matching degree, and multiple resources are selected to generate a recommended resource list, outputting the initial recommendation results. The algorithm collects real-time data on students' actions with the initial recommendation results, inputs it into the incremental learning algorithm, updates the learning preference model parameters, re-calculates the matching degree according to a set period, dynamically adjusts the recommended resource list, and outputs optimized personalized learning resource push results.

[0010] Furthermore, the matching degree calculation formula is as follows: , in, This represents the matching degree between the k-th student and the j-th learning resource. , , These represent the knowledge point fit coefficient, difficulty suitability coefficient, and learning preference coefficient, respectively, satisfying the following conditions: , This represents the degree of fit between the weak knowledge points of the k-th student and the knowledge points covered by the j-th learning resource. It is obtained by calculating the proportion of the number of elements in the intersection of the student's weak knowledge point set and the resource's knowledge point set to the total number of elements in the student's weak knowledge point set. This represents the fit between the difficulty level of the j-th learning resource and the knowledge level of the k-th student. Both the resource difficulty level and the student's knowledge level are divided into 5 levels. The fit is 1 when the levels are exactly the same, 0.8 when the levels differ by 1 level, 0.5 when the levels differ by 2 levels, and 0.2 when the levels differ by 3 levels or more. This represents the historical learning preference coefficient of the k-th student for the j-th type of learning resources. It is obtained by calculating the weighted average of the student's historical click count, learning completion rate, and learning time percentage for this type of learning resource.

[0011] Furthermore, the specific steps for constructing the supervisory rule base are as follows: Based on the national medical education syllabus standards data, management requirement data for training bases, professional bases, departments, and trainees are extracted to construct a supervision rule base; The supervisory rule base is input into the rule engine algorithm, which transforms the rules into executable logical expressions. Business data from the platform is received in real time. When the business data meets the triggering conditions in the logical expression, supervisory task data containing the supervisory object, supervisory type, triggering reason, supervisory requirements, and completion deadline is generated and the supervisory task is output. The business data corresponding to the supervision task is transformed into a feature vector, input into the isolated forest algorithm, an isolated tree forest is constructed, the anomaly score of each data point is calculated, and when the anomaly score is higher than the threshold, it is judged as an anomaly, marked as a key supervision object, and the anomaly data is output. Anomaly data is input into a knowledge graph algorithm to construct an association graph of anomalies, rectification measures, and corresponding modules. Nodes are divided into anomaly nodes, rectification measure nodes, and system module nodes. Edges represent the relationships between nodes. The edge between anomaly nodes and rectification measure nodes represents the corresponding rectification plan. The edge between rectification measure nodes and system module nodes represents the system module that needs to be operated. The association graph data is output. Receive rectification operation data initiated by the administrator based on the association graph data, including the rectification start time and rectification operation content, track the rectification progress in real time, record the rectification completion time, and after the rectification is completed, re-input the updated business data into the isolated forest algorithm to re-detect anomalies until the anomalies are eliminated, and output the supervision rectification closed loop completion data.

[0012] Furthermore, the specific steps for verifying the graduation conditions and outputting the compliance status result are as follows: Receive the graduation task configuration data input by the administrator, including task name, task start and end time, participant screening criteria, graduation conditions and certificate template data, and output the graduation task data; Based on business data, extract students' rotation completion data, course completion scores, credit data, attendance data, and process assessment results data, and output students' graduation-related data; The algorithm inputs the student's graduation-related data and graduation task data into the rule engine algorithm, and verifies them one by one in the priority order of rotation completion rate, attendance compliance rate, process assessment results, course completion comprehensive pass score and credit data. The verification results are output, and students who meet all conditions are marked as qualified, and students who do not meet the conditions are marked as unqualified. The algorithm also lists the name of the unqualified condition, the current data value, and the qualified value. The personal information of qualified students and certificate template data are input into the template matching algorithm to generate the initial draft of the graduation certificate data. An electronic signature is added to the initial draft of the graduation certificate, and the electronic graduation certificate data is output.

[0013] An intelligent management system for a medical education platform includes a data acquisition module, a four-dimensional adaptation module, a core business module, a personalized push module, a graduation management module, a storage module, and an interaction module. The data acquisition module is used to collect relevant data from the entire medical education process and transmit it to the storage module. The four-dimensional adaptation module is used to call the initial data set of the storage module, construct a four-dimensional intelligent adaptation model, generate personalized permission configuration results and data views, and transmit them to the interaction module and core business module. The core business module is used to perform business related to rotation scheduling, teaching activities, assessment and evaluation and supervision management, and output corresponding business data. The personalized push module is used to call the business data of the core business module, process it with an algorithm to generate personalized learning resource push results and transmit them to the interaction module. The graduation management module is used to receive graduation task configuration data, call relevant business data to complete graduation condition verification and graduation certificate generation, and transmit it to the interaction module. The storage module adopts a distributed storage architecture to store various types of collected data, configuration results, business data, and supporting documents, ensuring data security and scalability. The interaction module is used to provide an operation interface, display relevant information and receive user operation instructions. After verifying the legality of the instructions based on the personalized permission configuration results, it transmits them to the corresponding module.

[0014] Compared with existing technologies, the intelligent management method and system for medical education platforms provided by this invention have the following beneficial effects: This invention constructs a four-dimensional intelligent adaptation model, integrating multiple algorithms such as RBAC extended algorithm, genetic algorithm, and rule engine with a full-process data processing mechanism to achieve intelligent linkage and closed-loop data flow among various business modules in medical education. It accurately matches the needs of multi-level roles through refined permission configuration and personalized data views, and significantly improves the efficiency and scientific nature of medical education management by leveraging core functions such as intelligent scheduling, personalized learning resource delivery, intelligent supervision and rectification, and automated graduation verification. Simultaneously, it ensures that the training process strictly adheres to national medical education syllabus standards, effectively optimizing the quality of medical talent training through full-process data traceability and precise control. Furthermore, it possesses strong scenario adaptability and practicality, and can be widely applied to various scenarios such as standardized residency training, intern training, and higher education medical education, fully releasing the value of data and providing strong support for the high-quality development of medical education. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an intelligent management method applied to a medical education platform according to the present invention.

[0016] Figure 2 This is a schematic diagram of a module of an intelligent management system applied to a medical education platform according to the present invention. Detailed Implementation

[0017] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0018] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0019] See Figure 1 This invention provides an intelligent management method for medical education platforms, comprising the following steps: Step S100: Collect multi-source data related to medical education, and form a structured initial data set through data cleaning and processing; Multi-source data refers to raw data collections from multiple independent systems in the medical education management process. It encompasses heterogeneous data types such as user information, teaching schedules, and assessment records, and can be used to provide raw input for subsequent structured processing, supporting the fundamental source of data flow throughout the entire process. The structured initialization dataset is a unified dataset after cleaning, standardization, and format conversion. It possesses consistent field structures and data types and can be used to eliminate data noise and format differences, forming a high-quality initial dataset suitable for model building and algorithm calls. Data cleaning refers to a series of processes performed on the collected raw data, including deduplication, completion, standardization, and outlier correction, to remove invalid information, standardize data formats, and ensure that data quality meets the requirements of subsequent algorithm calculations and model building.

[0020] Step S200: Based on the structured initialization data set, construct a four-dimensional intelligent adaptation model of roles, departments, bases and levels, establish the mapping relationship between users, roles, permissions and data, and generate personalized permission configuration results and data views; Roles represent the functional responsibilities of users on a medical education platform, serving as the basic unit for permission allocation. They can be used as a classification basis in the permission control system, differentiating the operational scope and data access capabilities of different positions. Departments are professionally divided business units within medical institutions, serving as the basic execution unit for rotation training and teaching organization. They can be used as the spatial carrier for rotation scheduling, teaching task allocation, and resource allocation. Hierarchy reflects the vertical hierarchical relationship of the organizational structure in the medical education management system, embodying the hierarchical attribution of management permissions. It can be used to construct vertical control logic for permission inheritance mechanisms and data visibility scope. Personalized permission configuration results are permission sets for specific users based on a four-dimensional intelligent adaptation model. This allows different users to have differentiated functional entry points and operational boundaries within the same system. Personalized data views are information display interfaces customized based on user permission configuration results. They can be used to display only the data content that the user has the right to access, improving information retrieval efficiency and privacy security.

[0021] Step S300: Based on the permission configuration results, collect relevant data on students, departments and rotation templates, and realize intelligent linkage of rotation scheduling, teaching activities and assessment through fusion algorithm, and output business data of each link; Trainees are individuals receiving medical education and training. Their learning trajectories and growth data are the core management objects and can be used as the core data source for the training process, driving the operation of scheduling, teaching, and assessment. Rotation templates are pre-set templates of trainees' rotation schedules and training requirements in various departments. They can be used as input for intelligent scheduling, ensuring that rotation plans comply with the syllabus's stipulated cycles and content. Fusion algorithms are composite optimization methods integrating multiple computational strategies, used to solve decision-making problems under complex constraints. They can be used to improve the overall coordination and conflict avoidance capabilities of rotation scheduling, teaching activity arrangements, and assessment task allocation. Rotation scheduling is the management activity of allocating trainees to different departments and their instructors according to the training plan. It can be used to ensure that trainees complete their practical training in the designated departments on time and meet the training cycle requirements. Teaching activities are teaching implementation behaviors organized at specific times and places, such as case discussions, skills training, and academic lectures. They can be used to implement the teaching plan content and promote knowledge transfer and skills mastery. Assessment and evaluation is a quantitative evaluation process of students' learning outcomes at a certain stage. It includes forms such as exams, practical assessments, and comprehensive evaluations. It can be used to verify learning effectiveness, provide feedback, and influence subsequent training path decisions. Business data refers to the process and outcome data generated during the operation of core links such as rotation scheduling, teaching activities, and assessment and evaluation. It can be used as the data foundation for subsequent feature extraction, model training, anomaly detection, and graduation verification.

[0022] Step S400: Based on business data, generate personalized learning resource push results through feature extraction and model training. At the same time, build a supervision rule base based on the outline standard and output anomaly detection results and rectification guidance data. Feature extraction is a component that identifies and extracts structural indicators with analytical value from raw business data. It can be used to transform unstructured or semi-structured behavioral data into numerical inputs that can be used for model training. Model training is an iterative computational process that uses historical data to build predictive or recommendation models. It can be used to generate learning resource matching models that reflect individual preferences. Syllabus standards are normative documents issued by the state or industry regarding the goals, content requirements, and competency indicators of medical talent training. They can be used as the fundamental basis for the formulation of supervision rules and the assessment of training quality to ensure compliance. The supervision rule base is a set of executable inspection rules based on the syllabus standards. It is used to automatically monitor the training process and can be used to achieve real-time monitoring of key indicators such as scheduling completeness, teaching frequency, and assessment coverage. Anomaly detection results are event records that deviate from standard requirements after comparison with the supervision rule base. They can be used to mark potential problem points and trigger subsequent rectification processes. Rectification guidance data provides specific corrective measures for detected anomalies. It can be used to guide relevant personnel to take correct remedial actions and form a management closed loop.

[0023] Step S500: Based on business data, extract relevant data on student graduation, verify graduation conditions and output the achievement status result, match the certificate template to generate a graduation certificate and push it to the student's end.

[0024] Graduation-related data comprises all historical and current status data used to determine whether trainees meet graduation qualifications. It serves as input for the graduation condition verification mechanism. Graduation conditions are the various indicators that trainees must meet to obtain graduation qualifications. These are typically derived from the training syllabus and can be used as the legal standard for determining whether graduation is granted. The achievement status result is a Boolean output indicating whether the trainee has met all graduation conditions, clearly indicating whether the trainee is eligible to receive a graduation certificate. The graduation certificate is an official document confirming that the trainee has completed all training requirements and possesses the corresponding qualifications. It serves as official proof of the trainee's training experience and can be used for purposes such as professional title evaluation and professional registration.

[0025] For example, in the context of teaching management for medical school interns, the application method of this invention is as follows: A medical university affiliated hospital receives 300 undergraduate clinical medicine students for internships each year. The system accesses the school's academic affairs system data, establishes a four-dimensional model, and assigns different data views to internship team leaders, supervising teachers, and teaching office administrators. Based on the teaching syllabus, rotation week templates for departments such as internal medicine and surgery are set. The fusion algorithm, considering departmental capacity and instructor workload, completes batch intelligent scheduling and simultaneously generates weekly teaching lecture plans and mid-term skills assessment tasks. The system extracts features from teaching attendance records and assessment scores, discovering that some students have low scores in obstetrics and gynecology operations, and proactively recommends virtual simulation training courses. The supervision module, based on the teaching activity frequency standards stipulated in the syllabus, immediately issues a yellow warning and prompts students to make up for missed case discussions if a department has not held such discussions for two consecutive weeks. Before the end of the internship, the system automatically summarizes the rotation completion status and assessment results of all students, generates internship evaluation certificates for those who meet the standards and pushes them to their personal accounts, and marks those who do not meet the standards as pending manual review.

[0026] This invention first unifies heterogeneous data from multiple sources to form a high-quality initial dataset; then, under an access control framework, it enables differentiated access to functions and data for users at different levels; furthermore, it uses a fusion algorithm to drive the coordinated execution of scheduling, teaching, and assessment, ensuring consistency in time, resources, and tasks; then, it relies on business data to support the precise delivery of personalized learning resources and standard-based full-process supervision and monitoring; finally, it completes the automated verification of graduation requirements and the generation and delivery of certificates; the entire process forms a full-cycle management chain from data collection, access control configuration, business linkage, intelligent feedback to graduation closure, with each module sharing the same business data foundation, and the algorithm components operating collaboratively under the constraints of a four-dimensional access control model, which not only ensures the compliance and traceability of medical education management, but also improves resource allocation efficiency, teaching adaptation accuracy, and student service response level.

[0027] In one embodiment of the present invention, the multi-source data includes user basic data, role attribute data, basic information data of departments and bases, national medical education syllabus standard data, and historical business data. The specific steps for constructing the four-dimensional intelligent adaptation model are as follows: Step S201: Based on user basic data and role attribute data, establish the mapping relationship between users, roles, permissions and data through the RBAC extended algorithm, clarify the basic permission range corresponding to each role, and output the basic mapping rules; Based on user base data (including user type, job title, job content, department, and management scope) and role attribute data (including role name, job description, and permission requirements), an extended RBAC algorithm is used to construct a four-layer mapping logic of "user-role-permission-data". The algorithm clarifies the basic permission boundaries for each role, specifically covering access permissions for two-level menus and operation permissions for core function buttons under the menus (such as add, edit, delete, and query). It also defines the data scope boundaries accessible to each role, ultimately outputting a structured basic mapping rule document that provides underlying logical support for subsequent permission configuration.

[0028] The RBAC extended algorithm is an extension of the traditional RBAC (Role-Based Access Control) model. It adds a "data" dimension to the three-layer relationship of "user-role-permission", forming a four-layer mapping of "user-role-permission-data". It can accurately define the access permissions of roles to specific data and is suitable for the multi-level and multi-scenario permission management needs of medical education platforms.

[0029] Step S202: Input user basic data as input features and preset role labels as output results into a preset multi-label classification model to automatically assign initial roles to users and output the initial role assignment results; The multi-label classification model is a machine learning model that can assign multiple labels (role labels in this invention) to samples based on multiple input features. After training with historical role assignment data, it can automatically identify the matching relationship between users and each role, realize the intelligent allocation of initial roles, and reduce manual intervention.

[0030] The core input features are job title, job content, department, and management scope from the user's basic data. Pre-defined role labels such as top administrator, base director, department secretary, instructor, and student are used as the model's output targets and input into a multi-label classification model trained on historical role assignment data. The model automatically assigns an initial role to each user by calculating the matching degree between the input features and each role label, outputting a list of initial role assignment results including user ID, assigned role name, and matching confidence score.

[0031] Step S203: Collect user historical operation data, including menu access frequency, button click count, and data viewing range. Combined with basic mapping rules, dynamically adjust the permission priority of each role, increase the priority of permission items with access frequency higher than the preset threshold, and output the permission priority adjustment result. The system automatically collects users' historical operation data for the past 30 days from system logs, including the access frequency of each menu, the number of clicks on function buttons, and the specific scope of data viewed. This data is then matched with the basic mapping rules output in the first step. A permission access frequency threshold is set (e.g., ≥10 visits per month), and a weighted adjustment algorithm is used to dynamically optimize the permission priority of each role. Permission items with access frequencies exceeding the threshold are given higher display and response priorities on the user interface. The output is a table containing the adjustment results for each role's permission items and their corresponding priorities.

[0032] Step S204: Collect data on the user's department, base, and level; combine the basic mapping rules and permission priority adjustment results to clarify the specific menus, buttons, and data access scope that the user can operate; and generate personalized permission configuration results. Permission priority is the priority of permission usage set according to the frequency of user operations. Permissions that are accessed frequently are displayed first in the operation interface and have a faster response time, which can optimize the user operation process and improve the efficiency of core functions.

[0033] The system extracts user department, training base, and hierarchical information from the structured initial dataset. Combining this with the basic mapping rules from step one and the permission priority adjustment results from step three, a secondary filtering and definition process is performed using a refined permission algorithm. This clarifies the specific menu names, function button identifiers, and department / training base scope that each user can access. For example, it may allow only the department's teaching secretary to access menus related to rotation scheduling and performance evaluation within their department. A personalized permission configuration matrix indexed by user ID is then generated to ensure precise permission allocation down to the individual user.

[0034] Step S205: Collect historical data access records, combine user role attribute data to analyze user data needs and preferences, and generate personalized data views.

[0035] Extracting users' historical data access records for the past 90 days, including the names of accessed modules, data types, dwell time, and query frequency, and combining this with users' role attribute data (such as the management responsibilities of the base director and the learning needs of students), a demand preference identification algorithm is used to analyze users' core data needs. For example, for instructors, the algorithm prioritizes identifying their needs for data such as the learning progress and assessment results of their students, automatically generating personalized data views that include core data modules, statistical dimensions, and display styles, allowing users to manually adjust the module layout later.

[0036] In this embodiment, for new users without historical operation data, the system matches the historical average data demand preference vector of users with the same role, level, and professional base based on their four-dimensional attributes of role, department, base, and level, and generates a standardized initial personalized data view. After the user generates 30 consecutive days of valid operation data, the preference vector is automatically updated based on their actual access frequency, data viewing duration, and other behavioral characteristics, and the view is switched to a fully personalized data view, taking into account both the initial user experience of new users and the personalized needs of old users.

[0037] In this embodiment, the four-dimensional intelligent adaptation model of roles, departments, bases, and levels adopts a hierarchical nested tree logical structure. With the training base as the root node, it sequentially associates professional bases, clinical departments, and user roles. The upper-level node automatically inherits all data access permissions of the lower-level node, while the lower-level node only has the corresponding permissions of its own level and the levels below it. At the same time, permission blocking rules can be configured through the RBAC extended algorithm to block the default inherited permissions of the upper-level node for specific confidential departments or sensitive data. Finally, a four-dimensional permission matrix is ​​constructed to realize the rapid verification and fine-grained control of permissions at all levels.

[0038] This invention upgrades the system from "general permission allocation" to "personalized and precise adaptation." By combining the RBAC extended algorithm with a multi-label classification model, it ensures the accuracy of the initial matching of roles and permissions. A dynamic permission priority adjustment mechanism optimizes the permission usage experience to align with users' actual operating habits. Through three-dimensional data constraints of departments, bases, and levels, it achieves minimal and precise permission configuration, enhancing system data security. Personalized data views are generated based on user needs and preferences, reducing invalid data interference and improving user operational efficiency. The personalized permission configuration results and data views generated in this step provide precise support for permission verification and data retrieval in subsequent core business modules such as rotation scheduling and teaching assessments, ensuring the smoothness and security of the entire business process.

[0039] In one embodiment of the present invention, the specific process of implementing the rotating shift schedule is as follows: Step S301: Based on the permission configuration results, collect trainee basic information, department constraint information and rotation template information, and integrate them into basic scheduling data; Based on personalized permission configuration results, the boundaries of data collection permissions are clearly defined, and basic scheduling data is collected in a targeted manner: basic student information includes core fields such as the number of students, their major, enrollment time, and departments with rotation restrictions (if any); department constraint information includes the maximum number of students each department can accommodate, the number and specialization of supervising teachers, and the monthly acceptable rotation cycle; rotation template information includes national standard rotation templates and hospital-customized templates (clearly defining the rotation order of each specialty and the rotation duration requirements for each department). These three types of data are integrated into a structured scheduling basic dataset, and the data format is unified (such as department codes and standardized time formats) to provide standardized data support for subsequent scheduling algorithm input.

[0040] Step S302: Input basic scheduling data, encode the trainee rotation plan through intelligent scheduling algorithm, generate a preset number of initial scheduling scheme populations by combining department constraint information, and output the optimal intelligent template scheduling scheme through iterative optimization operation; Intelligent scheduling algorithms, also known as genetic algorithms, simulate selection, crossover, and mutation processes in biological evolution to find the optimal rotation scheduling scheme under constraints (such as department capacity and professional suitability). They are suitable for complex scheduling scenarios with multiple objectives and constraints. The initial scheduling scheme population is the initial input set of the genetic algorithm, containing multiple randomly generated valid scheduling schemes (without obvious conflicts). The population size needs to be reasonably set according to the number of trainees (50-200), directly affecting the algorithm's optimization efficiency and the quality of the optimal scheme.

[0041] The basic scheduling dataset is input into the intelligent scheduling algorithm (preferably a genetic algorithm). First, the rotation plan of each trainee is encoded into chromosomes in the format of "department code - rotation duration", with the chromosome length consistent with the number of rotation cycles. Based on the department constraint information, the initial population size (usually 50-200) is set, generating an initial scheme containing different scheduling combinations. The optimal scheme is selected by roulette wheel selection, and the parent chromosome gene segments are exchanged by single-point crossover. Gene mutation is performed with a mutation probability of 0.01-0.05. After 100-200 iterations, the iteration stops when the change in the optimal fitness value of the population is less than 0.001 for 10 consecutive generations. The optimal intelligent template scheduling scheme that takes into account the balance of department capacity and the professional matching of trainees is output.

[0042] Step S303: Generate preview data based on the scheduling plan from multiple dimensions. When the number of scheduled personnel exceeds the department's capacity, mark the exception and output the scheduling preview results. The multi-dimensional generated preview data is a collection of data that displays the scheduling results from different perspectives, including the department dimension (the distribution of the number of people on duty in each department) and the student dimension (the rotation plan of an individual student). It supports switching between three time dimensions: month, week, and day, making it easy for administrators to quickly verify the rationality of the scheduling.

[0043] Based on the optimal scheduling scheme, the system automatically generates scheduling preview data in three dimensions: monthly, weekly, and daily. The department-level preview data displays the list of students and the number of students scheduled for each department within each time period. The student-level preview data displays the rotation department, start and end time, and mentor assignment for each student. By comparing the real-time capacity of each department with the maximum acceptance capacity threshold, when the number of students scheduled for a certain time period exceeds the threshold, the time period and the corresponding department are highlighted in red, generating a scheduling preview result with an anomaly marker, which is then simultaneously pushed to users with department management permissions.

[0044] Step S304: Collect student enrollment status, manual registration data, attendance records and assessment results in real time during the student rotation process. Track the rotation progress through a preset monitoring algorithm. When the preset early warning trigger conditions are met, generate early warning information, push it to the corresponding instructors and management users, and output early warning data. The warning trigger conditions are preset rules for triggering rotation abnormality warnings. They are set based on medical education and training requirements. Common conditions include a rotation manual completion rate below a threshold, an excessive number of attendance abnormalities, and failure to meet the standards for a periodic assessment, ensuring that abnormal situations are detected in a timely manner.

[0045] The system collects real-time data on trainees' rotation process: Departmental entry status includes actual entry time and explanations for absence; manual registration data covers the completion status of bed assignments, disease registration, and operation records; attendance records include daily check-in time, number of days off, and number of late arrivals / early departures; assessment results include periodic quiz scores and operational skills evaluations; a sliding window algorithm (window length set to 7 days) is used to monitor the data in real time, setting early warning trigger conditions (such as manual completion rate below 80% within the window or attendance abnormalities exceeding 3 times). When the conditions are met, an early warning message is automatically generated, clearly indicating the trainee ID, warning type, and current data status, and pushed to the corresponding supervising teacher and department secretary, outputting structured early warning data for archiving.

[0046] Step S305: Based on the early warning data and rotation adjustment needs, extract the professional matching status, teaching load and historical evaluation of teaching staff from the basic scheduling dataset. Solve the optimal teaching staff matching result through optimization algorithm. Combine the overall arrangement of the current rotation plan and departmental constraint information, determine the optimal rotation adjustment path through path planning algorithm, and output the teaching staff matching result and rotation path adjustment plan. The optimization algorithm (Hungarian algorithm) is a combinatorial optimization algorithm used to solve the problem of optimal matching between instructors and students. Its goal is to establish an optimal correspondence between instructors and students, achieving the highest professional matching degree and the most balanced teaching workload. The path planning algorithm is used to adjust student rotation paths. Its core objective is to minimize the impact of rotation adjustments on the overall scheduling plan while satisfying departmental constraints, and simultaneously shorten students' commuting costs across departments and the total rotation time.

[0047] Based on early warning data and the rotation adjustment requests submitted by trainees, relevant data on mentors are extracted from the basic scheduling dataset: professional matching is quantified by the degree of fit between the mentor's professional direction and the trainee's major (0-1 points); teaching load is represented by the proportion of the current number of trainees being taught to the maximum teaching limit; historical evaluation is the comprehensive score given by trainees over the past year (0-100 points). The above data is input into the Hungarian algorithm to solve for the optimal matching result between mentors and trainees, ensuring the rational allocation of teaching resources; at the same time, combined with the current overall rotation plan and departmental constraints, the shortest path algorithm is used to plan the rotation adjustment path, with the goal of "shortest total rotation time after adjustment, lowest cross-departmental commuting cost, and minimal impact on the scheduling of other trainees", outputting a mentor matching list and rotation path adjustment plan.

[0048] Step S306: Collect the completion status of the manual, attendance compliance, various assessment scores and peer evaluation results of students during their rotation period, input the multi-dimensional weighted scoring algorithm, calculate the comprehensive achievement results, and generate multiple retake adaptation plans for students who do not meet the standards.

[0049] The multi-dimensional weighted scoring algorithm is a quantification of course completion status. By assigning weights to multiple evaluation dimensions such as rotation manual completion rate and attendance compliance rate, it calculates a comprehensive completion score, achieving objectivity and standardization in course completion determination and avoiding subjective biases from manual judgment. The retraining adaptation plan is a supplementary rotation plan designed for trainees who have not met the course completion standards. It includes three types: continuing rotations in the current department, adding to the end of the schedule, and random insertion. The choice can be flexibly made based on department capacity, trainee's remaining rotation time, and other actual circumstances, ensuring that trainees address their skill gaps.

[0050] Collect comprehensive evaluation data during trainees' rotations: the rotation manual completion rate is calculated as "actual number of completed registration items / number of registration items to be completed × 100"; the attendance compliance rate is calculated as "compliance check-in days / total number of check-in days to be completed × 100"; the theoretical assessment score and skills assessment score are the standardized scores (0-100 points) of the corresponding assessment stages; and the mentor-trainee mutual evaluation score is the average of the two-way evaluations from the mentor and trainee (0-100 points). Input the above data into a multi-dimensional weighted scoring algorithm (the weight coefficients are preset according to the national medical education syllabus standards and can be manually adjusted) to calculate the departmental comprehensive pass score. When the score is lower than the preset pass threshold (e.g., 60 points), three retake adaptation plans are automatically generated: continue rotation in the current department for supplementary training, add the corresponding department rotation to the end of the existing rotation plan, or randomly insert an idle period in the current rotation plan for the administrator to choose from.

[0051] This invention utilizes a genetic algorithm to optimize scheduling, significantly reducing departmental capacity conflict rates (which can be controlled within 5%) and improving scheduling efficiency and rationality. Through real-time data acquisition and a sliding window monitoring algorithm, it enables timely early warning and rapid response to rotation anomalies, ensuring the continuity of the training process. The Hungarian algorithm and shortest path algorithm optimize instructor matching and rotation adjustments, ensuring optimal resource allocation with minimal impact on the overall plan. A multi-dimensional weighted scoring algorithm provides objective quantification of department exit determination, combined with a standardized retake plan, ensuring controllable quality of department exits. The scheduling plan, early warning data, and instructor matching results output in this step provide precise data support for subsequent modules such as teaching activity arrangement and assessment implementation, promoting the upgrade of rotation management from "manual coordination" to "intelligent closed loop," and significantly improving the refinement and management efficiency of medical education rotation training.

[0052] In one embodiment of the present invention, the intelligent linkage process between teaching activities and assessment is as follows: Step S307: Based on the permission configuration results, the trainee's current rotation department and specialty data, and combined with the knowledge point tags of the National Medical Education Syllabus, perform tag matching with the teaching activity resource library, filter out suitable teaching activities and corresponding resources, and output suitable activity data; The Teaching Activity Resource Repository is a structured database storing teaching activities and supporting resources related to medical education. Resource types cover various formats such as mini-lectures, case discussions, and teaching rounds. Each resource is associated with standardized metadata such as knowledge point tags, applicable specialties, and rotation stages. It supports filtering suitable resources by tag matching and keyword search. Knowledge point tags are standardized keywords or phrases that categorize and identify knowledge points, core content of teaching activities, and assessment focuses in the national medical education syllabus. They possess unified coding rules and a hierarchical system, serving as the core link for accurately matching teaching activities, assessment tasks, and student needs.

[0053] Based on the personalized permission configuration results output by the four-dimensional adaptation module, core data such as the trainee's current rotation department and specialty are collected in a targeted manner. At the same time, the knowledge point system in the National Medical Education Syllabus is extracted and standardized and tagged (e.g., the internal medicine knowledge point tags include "diagnosis and treatment of cardiovascular diseases" and "differentiation of common digestive system diseases"). The trainee data is associated and matched with the tagged syllabus knowledge points, and then compared with the tag similarity of the teaching activity resource library that stores resources such as mini-lectures, case discussions, and teaching rounds. Teaching activities and supporting shared materials (documents, audio and video, etc.) that are highly consistent with the trainee's rotation stage and specialty are selected. Finally, structured adaptation activity data containing activity name, adapted knowledge points, resource list, and participation requirements is output.

[0054] Step S308: Using a task binding algorithm, the adapted activity data is associated with the preset assessment template library to generate a binding relationship, clarify the corresponding assessment type and knowledge point scope, and output the linked task data. The task binding algorithm is used to establish a correspondence between teaching activities and assessment tasks. By analyzing the similarity between the two in dimensions such as knowledge point coverage, applicable scenarios, and evaluation objectives, it generates a stable and accurate mapping relationship, ensuring that the assessment task can effectively test the learning outcomes of the corresponding teaching activities. The assessment template library is a database storing standardized templates for various assessment tasks. These templates contain fixed configurations such as assessment type (theory, skills, etc.), knowledge point scope, question type, score, answering time, and evaluation criteria. They can be directly called or slightly modified for use in issuing assessment tasks, improving the efficiency of assessment organization.

[0055] A task binding algorithm is used to establish a mapping relationship between adapted activity data and a preset assessment template library. The assessment template library covers various types such as theoretical assessment, skills operation assessment, and case analysis assessment. Each template clearly defines the corresponding knowledge point coverage, question type distribution, and score setting. The algorithm identifies the matching degree between the core knowledge points of the teaching activity and the knowledge points of the assessment template, determines one or more adapted assessment types, clarifies the specific knowledge point boundaries and evaluation criteria of the assessment, generates a binding relationship table of "teaching activity-assessment task", and outputs linked task data containing activity ID, assessment template ID, knowledge point range, and assessment trigger conditions.

[0056] In this embodiment, the task binding algorithm adopts the weighted Jaccard tag similarity calculation method, assigning 2 times the weight to core knowledge points of the National Medical Education Syllabus and 1 times the weight to ordinary knowledge points, and setting a similarity threshold of 0.6. When the weighted similarity between the knowledge point tags of teaching activities and the knowledge point tags of assessment templates reaches the threshold, a binding relationship is automatically established. When multiple assessment templates meet the threshold, they are sorted and selected according to the similarity from high to low. A three-level processing rule is adopted for binding conflicts. The binding relationship manually specified by the administrator has the highest priority. If not manually specified, it is sorted according to the weighted similarity. When the similarity is the same, it is filtered according to the type matching degree of teaching activities and assessment templates. Unresolved conflicts are automatically pushed to the teaching administrator for manual confirmation.

[0057] Step S309: After the teaching activity ends, the bound assessment task will be automatically released, the assessment data submitted by the students will be collected, encrypted and stored, and the assessment data will be output. After the teaching activities conclude at the preset time, the system automatically triggers the assessment task release process bound in the linked task data, notifying participating students via SMS, platform push notifications, and other means. After students complete the assessment and submit their answers within the specified time, the system receives the assessment data submitted by the students (including answers to objective questions, answers to subjective questions, skill operation videos, etc.), encrypts the assessment data using the AES symmetric encryption algorithm, and stores it in a distributed database to ensure data security and integrity. At the same time, it outputs an assessment data list containing the student ID, assessment task ID, submission time, and the encrypted assessment data index.

[0058] Step S310: Based on the assessment data, calculate the ratio of the student's assessment score to the full score of the corresponding knowledge point, identify the knowledge points that have not been mastered, and at the same time, combine the data from teaching activities to analyze the correlation and output the mastery data and correlation results. Mastery data is a set of data that quantitatively reflects students' mastery of each knowledge point. The core indicator is the ratio of the score for a knowledge point to the full score for that knowledge point. It also includes derivative information such as a list of knowledge points not mastered and the level of mastery. It is the core basis for evaluating teaching effectiveness and optimizing subsequent training programs.

[0059] Extract the scores of each student for each knowledge point from the encrypted assessment data, calculate the ratio of the score for a single knowledge point to the full score for that knowledge point (e.g., if a knowledge point has a full score of 20 points and the student scores 10 points, the ratio is 0.5), set a ratio below 0.6 as the threshold for not mastering the knowledge, and filter out knowledge points below this threshold as those not mastered by the student; at the same time, collect student participation data in the corresponding teaching activities (e.g., attendance status, participation time, number of interactive speeches, etc.), and analyze the correlation between participation data and not mastered knowledge points through association rule algorithms (e.g., students who participated for less than 30 minutes had a significantly higher rate of not mastering a certain type of knowledge point than students who participated for sufficient time), and output mastery data and association results including a list of not mastered knowledge points, the mastery ratio for each knowledge point, and a correlation analysis report of participation data and assessment results.

[0060] Association rule algorithms are used to uncover potential associations between teaching activity participation data and assessment results. By analyzing the distribution patterns of assessment scores corresponding to different participation behaviors (such as participation duration and interaction frequency), they identify key factors affecting learning outcomes and provide data support for optimizing teaching activities and assessment tasks.

[0061] Step S311: Based on the mastery data and correlation results, optimize subsequent linkage tasks and output the task optimization results.

[0062] Based on the knowledge points not mastered and their related results identified in the mastery data, targeted optimizations are made to subsequent collaborative tasks: for teaching activities with a high concentration of knowledge points not mastered, a more suitable assessment template is used (e.g., increasing the proportion of case analysis questions); for cases where there is a strong correlation between participation data and assessment results, the participation requirements for teaching activities are adjusted (e.g., extending the minimum participation time) or the assessment triggering time is adjusted (e.g., completing the assessment within 24 hours after the end of the teaching activity), generating task optimization results that include the direction of task adjustment, the selection of new assessment templates, and suggestions for optimizing participation rules, providing a dynamic adjustment basis for the linkage between subsequent teaching activities and assessment tasks.

[0063] It should be noted that the calculation formula for the multi-dimensional weighted scoring algorithm is as follows: , in, The overall score for graduation is the core quantitative indicator for determining whether a trainee meets the graduation requirements. The score ranges from 0 to 100, with higher scores indicating better performance. A score of 60 is usually preset as the threshold for graduation (which can be adjusted by the hospital according to the training requirements). Let the weight coefficient of the i-th evaluation dimension satisfy the following condition: =1, This represents the score for the i-th evaluation dimension, ranging from 0 to 100. The score for each dimension is calculated as follows: When i=1, S1 is the completion rate of the rotation manual, and the calculation logic is "the number of registration items actually completed by the trainee in the rotation department, such as bed management, disease types, operations, emergency, and outpatient services ÷ the number of registration items required to be completed by the department × 100". When i=2, S2 is the attendance compliance rate, and the calculation logic is "the number of compliant attendance days of the trainee in the rotation department (including normal attendance and approved supplementary attendance days) ÷ the total number of attendance days in the rotation cycle × 100". When i=3, S3 is the theoretical examination score, which is obtained by standardizing the student's original theoretical examination score in the department (when the original score exceeds 0-100 points, it is mapped to the target range by "(original score - minimum score) ÷ (maximum score - minimum score) × 100"). When i=4, S4 is the skills assessment score, which is obtained by standardizing the student's original score of the skills operation test in the department (the processing method is the same as the theoretical assessment score). When i=5, S5 is the peer review score, which is the average of the mentor's score on the student's professional ability, learning attitude, communication skills and the student's score on the mentor's teaching quality. When i≥6, S For scores of custom evaluation dimensions, the calculation method is preset by the administrator according to the dimension type (such as quantitative data, level rating).

[0064] The evaluation dimensions include the completion rate of the rotation manual, attendance compliance rate, theoretical assessment score, skills assessment score, and peer assessment score. The weight coefficients of each dimension are preset according to the national medical education syllabus standards and the hospital's personalized training requirements, and can be manually adjusted by the administrator through the platform.

[0065] In one embodiment of the present invention, the business data includes rotation management-related data, teaching activity-related data, and assessment and evaluation-related data. The specific steps for generating personalized learning resource push results are as follows: Step S401: Based on business data, extract students' majors, training time, weak knowledge points, historical learning resource type preferences, learning duration distribution, and assessment score distribution, and output feature vectors; The system retrieves rotation management data, teaching activity data, and assessment data from the core business modules and extracts core student characteristic dimensions: student major (standardized professional classifications such as internal medicine, surgery, and obstetrics and gynecology), enrollment time (accurate to year, month, and day), weak knowledge points (determined based on the list of knowledge points not mastered in the assessment), historical learning resource type preference (such as the access ratio of videos, documents, and exercises), learning time distribution (daily / weekly learning time statistics), and assessment score distribution (scores for each knowledge point and assessment type). The extracted feature data is then standardized (such as unifying the format of time data, labeling categorical data, and normalizing numerical data to the [0,1] interval) to form a fixed-dimensional structured feature vector, providing standardized input for subsequent model training.

[0066] Feature vectors are structured numerical vectors that transform a student's multi-dimensional features, such as major, training time, and weak knowledge points. Each feature corresponds to one dimension of the vector. The vector format is uniform and can be directly input into the algorithm model for calculation. It is the core data carrier for realizing accurate profiling of student needs.

[0067] Step S402: Take the feature vector as input, and the learning completion rate of the students on the learning resources as the output label. Divide the training set and the test set, construct the learning preference model through the logistic regression algorithm, optimize the model parameters through the gradient descent algorithm, minimize the square error between the model prediction value and the actual value, stop training when the accuracy of the model on the test set is higher than the threshold, and output the trained learning preference model. Logistic regression is a machine learning algorithm used for binary classification or probability prediction. It's used to build learning preference models by learning the mapping relationship between learners' historical characteristics and resource completion rates, predicting learners' willingness to learn new resources. It features simple models, efficient training, and strong interpretability, making it suitable for resource delivery needs in medical education scenarios. Gradient descent is an iterative algorithm used to optimize model parameters. By calculating the gradient direction of the loss function, it gradually adjusts the weight parameters of the learning preference model to minimize the error between the model's predicted and actual values. It is the core algorithm for parameter optimization during model training.

[0068] The feature vector is used as the model input variable, and the learning completion rate of students for historical learning resources (number of completed resources / total number of pushed resources × 100%) is used as the output label. The dataset is randomly divided in a 7:3 ratio, with 70% used as the training set for model parameter learning and 30% used as the test set for model performance validation. A learning preference model is constructed using the logistic regression algorithm, and the model weight parameters are initialized with random values. The parameters are iteratively optimized using the gradient descent algorithm. The loss function is the sum of the squared errors between the model's predicted value and the actual learning completion rate. The loss value is calculated and the weights are adjusted after each iteration. Training stops when the model's prediction accuracy on the test set is higher than 85% and the change in the loss value is less than 0.001 for 5 consecutive iterations. The trained learning preference model and the corresponding optimal parameter set are output.

[0069] In this embodiment, the learning completion rate adopts differentiated quantitative standards for different types of learning resources. For video resources, the rate is calculated as the ratio of actual effective viewing time to the total resource time. Effective viewing requires that the speed increase does not exceed 1.5 times and the single viewing time is not less than 80% of the total time. For document resources, the rate is calculated as the ratio of actual reading time to standard reading time. The standard reading time is estimated at 1000 words / 10 minutes and page scrolling behavior is verified to exclude idle time. For exercise resources, the rate is calculated as the ratio of the number of correct answers to the total number of questions. The completion rate of all resources is uniformly quantified as a value of 0-100%, which is used as the training output label of the learning preference model.

[0070] Step S403: Collect resource data from the learning resource library, filter relevant resources based on the students' professional data, and output the filtered resource data; The system accesses the full resource data of the learning resource library, which covers five standardized resource categories: videos, audio, images, documents, and exercises. Each resource category is associated with metadata such as professional tags, knowledge point tags, difficulty level tags (levels 1-5), and resource format. Based on the student's professional characteristics extracted in step 1, a tag matching algorithm is used to filter out resources whose professional tags are consistent with or highly related to the student's major. Duplicate resources and resources with incompatible formats are removed. The filtered resources are then categorized and organized according to knowledge point tags, and a structured set of filtered resource data is output to ensure the basic compatibility of resources with the student's major.

[0071] Step S404: Based on the feature vector and the filtered resource data, combined with the learning preference model, calculate the matching degree between the learner and each resource using the matching degree calculation formula, sort them from high to low matching degree, select multiple resources to generate a recommended resource list, and output the initial recommendation result; The matching degree calculation formula is a mathematical expression used to quantify the degree of fit between learners and learning resources. By integrating three core factors—knowledge point fit, difficulty fit, and learning preference coefficient—it achieves accurate quantification of resource fit and provides an objective basis for recommending and ranking resources.

[0072] The feature vector and the filtered resource data are input into the trained learning preference model. The matching degree is calculated one by one with the matching degree between the kth student and the jth learning resource. The resources are sorted from high to low according to the matching degree. The top 10 resources are selected to form a recommended resource list and the initial recommendation result is output.

[0073] Step S405: Collect student operation data on the initial recommendation results in real time, input it into the incremental learning algorithm, update the learning preference model parameters, re-execute the matching degree calculation according to the set period, dynamically adjust the recommended resource list, and output the optimized personalized learning resource push results.

[0074] Incremental learning algorithms are algorithms that update model parameters without retraining the entire dataset, using only new data (such as data on students' actions on recommended resources). In this invention, they are used to dynamically optimize the learning preference model, enabling it to quickly adapt to changes in students' learning needs while reducing the computational and time costs of model updates.

[0075] The system logs collect real-time user behavior data regarding the initial recommendations, including resource click counts, actual learning duration, learning completion status (completed / incomplete), subjective evaluation scores (1-5 points), and instructions to mark uninterested resource types. This data is then used as incremental data to feed into the learning preference model, which updates the model's weight parameters using an incremental learning algorithm, eliminating the need to retrain the entire dataset and improving update efficiency. The matching degree calculation process is re-executed every 7 days, dynamically adjusting the sorting and content of the recommended resource list. For resource types marked as uninteresting by users, their corresponding Dkj coefficient is reduced by 50%, decreasing the frequency of similar resource recommendations. Finally, the optimized personalized learning resource recommendations are output and synchronized to the user's device.

[0076] In this embodiment, incremental learning uses the stochastic gradient descent algorithm to dynamically fine-tune the model weights. Every 7 days, student behavior data on recommended resources is collected as the incremental training set. Only the weight parameters related to the incremental data are updated, without the need for full retraining. A weight decay coefficient of 0.99 is introduced to weaken the influence of historical data on the model. For resource types that students mark as uninteresting, their corresponding preference coefficients are reduced by 50% to reduce the frequency of similar resource pushes. After each incremental update, the model accuracy is verified through the test set. When it is below 80%, full retraining is triggered.

[0077] It should be noted that the matching degree calculation formula is as follows: , in, This represents the matching degree between the k-th student and the j-th learning resource. α represents the knowledge point fit coefficient, which is used to weigh the importance of matching the student's weak knowledge points with the resource knowledge point coverage. The value range is 0-1, and the default value in this invention is 0.4. It can be dynamically adjusted according to the requirements and weights of knowledge points in the national medical education syllabus. The higher the requirements of knowledge points, the larger the value of α.

[0078] This represents the difficulty adaptation coefficient, used to adjust the weight of the resource difficulty level and the student's knowledge level. The value range is 0-1, with a default value of 0.3. For students in the basic stage, the β value can be increased (e.g., 0.4) to ensure that the resource difficulty matches the student's ability to accept it, avoiding the inefficiency caused by resources that are too difficult or too easy.

[0079] γ represents the learning preference coefficient, used to reflect the weight of the influence of students' historical learning habits on resource recommendations. Its value ranges from 0 to 1, with a default value of 0.3. The richer the accumulated learning behavior data of students, the more significant the moderating effect of γ, which can improve students' acceptance of recommended resources and their learning enthusiasm, thus satisfying [the desired outcome].

[0080] This represents the degree of fit between the weak knowledge points of the k-th student and the knowledge points covered by the j-th learning resource. It is obtained by calculating the proportion of the number of elements in the intersection of the student's weak knowledge point set and the resource's knowledge point set to the total number of elements in the student's weak knowledge point set, and the value ranges from 0 to 1.

[0081] This represents the fit between the difficulty level of the j-th learning resource and the knowledge level of the k-th student. Both the difficulty level of the resource and the knowledge level of the student are divided into 5 levels. The fit is 1 when the levels are exactly the same, 0.8 when the levels differ by 1 level, 0.5 when the levels differ by 2 levels, and 0.2 when the levels differ by 3 levels or more. The value range is 0-1.

[0082] This represents the historical learning preference coefficient of the k-th student for the j-th type of learning resource. It is obtained by calculating the weighted average of the student's historical click count, learning completion rate, and learning time percentage for this type of learning resource. The value ranges from 0 to 1. = Click count percentage × 0.3 + Learning completion rate × 0.4 + Learning time percentage × 0.3, where the learning completion rate has the highest weight (0.4), highlighting the core role of effective learning in preference determination. A value of 1 indicates that the learner has a strong preference for and willingness to complete this type of learning resource.

[0083] In one embodiment of the present invention, the specific steps for constructing the supervisory rule base are as follows: Step S406: Based on the national medical education syllabus standard data, extract the management requirement data of training bases, professional bases, departments and trainees, and construct a supervision rule base; The system retrieves national medical education syllabus standard data stored in the storage module and extracts management requirements at four levels: training base, professional base, department, and trainee. At the training base level, it focuses on extracting quantitative indicators such as the threshold for fluctuations in the number of trainees and the upper limit for the number of rotation anomalies. At the professional base level, it focuses on core requirements such as the compliance rate of rotation cycles and the completion rate of process assessments for each specialty. At the department level, it clarifies management standards such as the ratio of supervising teachers and the participation rate in teaching activities. At the trainee level, it defines specific specifications such as the completion rate of the manual and the passing score for assessments. The extracted requirement data is structured and formatted as "Rule ID-Level-Indicator Name-Quantitative Standard-Trigger Condition-Responsible Entity," integrating them to form a supervisory rule base covering the entire process and multiple levels, ensuring that the rules are completely consistent with the syllabus standards and are executable.

[0084] Step S407: Input the supervision rule base into the rule engine algorithm, convert the rules into executable logical expressions, receive business data from the platform in real time, and when the business data meets the triggering conditions in the logical expression, generate supervision task data containing the supervision object, supervision type, triggering reason, supervision requirements, and completion time limit, and output the supervision task. A rule engine algorithm is an algorithm that can convert natural language rules into computer-executable logical expressions, and perform rule matching and automatically trigger corresponding tasks based on business data. In this invention, it is used to realize the automated generation of supervision tasks, and can complete the entire process from rule matching to task issuance without manual intervention.

[0085] The completed supervision rule base data is input into the rule engine algorithm. The algorithm uses semantic parsing to transform each rule described in natural language into a computer-executable logical expression (e.g., "Monthly change in the number of trainees at the training base > 10% → trigger compliance supervision"). The system receives full-process business data such as rotation management, teaching activities, and assessment and evaluation output from the core business modules in real time. The business data is compared and matched with the logical expressions one by one. When the business data meets the triggering conditions in the logical expression, standardized supervision task data is automatically generated. This data includes core fields such as the supervision object (specific base, department, or trainee), supervision type (compliance supervision / quality supervision), triggering reason (e.g., "the frequency of teaching activities conducted by the department does not reach the lower limit"), supervision requirements (e.g., "submit a rectification plan within 5 working days"), and completion deadline, and is simultaneously pushed to the platform account of the corresponding responsible entity.

[0086] Step S408: Convert the business data corresponding to the supervision task into feature vectors, input them into the isolated forest algorithm, construct an isolated tree forest, calculate the anomaly score of each data point, and when the anomaly score is higher than the threshold, it is determined to be an anomaly, marked as a key supervision object, and the anomaly data is output. The Isolation Forest algorithm is an anomaly detection algorithm based on ensemble learning. It quickly isolates anomalous data points by constructing multiple isolation trees and converts the isolated path length of the data points into anomaly scores. It is suitable for anomaly identification in medical education business data and features fast detection speed, high accuracy, and sensitivity to anomalous data.

[0087] For the generated supervision tasks, the corresponding business data (such as student handbook registration data and attendance data that trigger the "Student Handbook Completion Rate Not Meeting Standard" supervision task) is extracted. Key indicators (such as handbook completion rate, number of incomplete items, and number of overdue days) are extracted using feature engineering algorithms to form feature vectors. The feature vectors are then input into the isolated forest algorithm. The algorithm constructs multiple isolated trees to form an isolated forest by randomly selecting features and split points. The average isolated path length of each data point in the forest is calculated and converted into an anomaly score in the range of 0-1. When the anomaly score is higher than a preset threshold (such as 0.7), the data point is determined to be an anomaly and marked as a key supervision target. The anomaly data containing the anomaly ID, corresponding original business data, anomaly score, and associated supervision task ID is output, which helps administrators quickly locate core issues.

[0088] Step S409: Input the anomaly data into the knowledge graph algorithm to construct an association graph of anomalies, rectification measures and corresponding modules. The nodes are divided into anomaly nodes, rectification measure nodes and system module nodes. The edges represent the association between nodes. The edge between anomaly nodes and rectification measure nodes represents the corresponding rectification plan. The edge between rectification measure nodes and system module nodes represents the system module to be operated. Output the association graph data. Knowledge graphs are visual semantic networks centered on nodes and edges. In this invention, nodes are divided into anomaly nodes, rectification measure nodes, and system module nodes. Edges represent the relationships between nodes (such as the correspondence between anomalies and rectification measures), which are used to intuitively present the rectification path and reduce the threshold for rectification operations.

[0089] Anomaly data is input into a knowledge graph algorithm. The algorithm first defines three types of nodes: anomaly nodes (e.g., "Student manual completion rate is below 60%", including anomaly indicators, current values, and achievement requirements), rectification measure nodes (e.g., "Mentor provides one-on-one supervision for supplementation", including rectification steps, responsible personnel, and completion standards), and system module nodes (e.g., "Rotation management module - manual management function", including operation paths and permission requirements). Then, the algorithm mines the relationships between nodes. The edges between anomaly nodes and rectification measure nodes are labeled with specific rectification processes (e.g., "1. Mentor reminds 2. Student supplements 3. Department secretary reviews"), and the edges between rectification measure nodes and system module nodes are labeled with detailed operation paths (e.g., "Login platform - rotation management - manual management - supplementation function"). This constructs a visual "anomaly - rectification measure - corresponding module" relationship graph, outputting relationship graph data that includes node attributes, edge relationship descriptions, and operation guidelines.

[0090] Step S410: Receive rectification operation data initiated by the administrator based on the association graph data, including the rectification start time and rectification operation content, track the rectification progress in real time, record the rectification completion time, and after the rectification is completed, re-input the updated business data into the isolated forest algorithm to re-detect anomalies until the anomalies are eliminated, and output the supervision rectification closed loop completion data.

[0091] The closed-loop supervision and rectification refers to the complete management process from anomaly identification, supervision task generation, rectification measure implementation, rectification progress tracking to anomaly elimination. Through the "detection-rectification-re-inspection" cycle mechanism, it ensures that every anomaly can be effectively resolved, achieving closed-loop control of supervision and management.

[0092] The system receives real-time rectification operation data initiated by the administrator based on the association graph data, including the rectification start time, specific operation content (such as issuing supplementary information notices, adjusting the work schedule), and information on the personnel involved in the implementation. It automatically generates a rectification tracking ledger and dynamically records the rectification progress (not started / in progress / completed). After the rectification is completed, the updated business data (such as the manual data supplemented by trainees) is re-entered into the isolated forest algorithm to recalculate the anomaly score. If the score is lower than the preset threshold, the anomaly is determined to be eliminated, and the data on the completion of the supervision and rectification loop is output. If the score is still higher than the threshold, the system returns to step four to rematch the appropriate rectification measures and continues to track until the anomaly is completely eliminated, ensuring that the supervision and rectification form a complete closed loop.

[0093] In one embodiment of the present invention, the specific steps for verifying the graduation conditions and outputting the achievement status result are as follows: Step S501: Receive the graduation task configuration data input by the administrator, including task name, task start and end time, participant screening criteria, graduation conditions and certificate template data, and output the graduation task data. The interaction module receives the graduation task configuration data input by the administrator. This data must include the task name (e.g., "2024 Annual Standardized Residency Training Graduation Assessment"), task start and end times (accurate to year, month, day, hour, and minute), participant screening criteria (supporting a combination of screening by training base, specialty, and training time interval), graduation conditions (clearly defining quantitative standards such as 100% rotation completion rate, comprehensive exit score ≥60 points, cumulative credits ≥50 points, attendance compliance rate ≥90%, and all process assessments passed), and certificate template data (the template includes fixed elements such as hospital logo, certificate name, issuing unit, and issuance date, and allows administrators to customize text format and element positions through a visual editor). The system performs format verification and structured processing on the configuration data, generating standardized graduation task data in the format of "Task ID - Configuration Item - Parameter Value", which is then stored in the storage module and synchronized to the graduation management module.

[0094] Step S502: Based on business data, extract students' rotation completion data, course completion scores, credit data, attendance data, and process assessment results data, and output students' graduation-related data. The graduation management module calls business data stored in the core business module through a data interface and uses an SQL query algorithm to accurately extract graduation-related data for trainees: rotation completion data includes the name of each rotation department, actual rotation duration, number of rotation departments, and rotation completion rate; the comprehensive exit score is the result of a multi-dimensional weighted scoring algorithm; credit data includes the cumulative value of course learning credits, teaching activity participation credits, and assessment achievement credits; attendance data includes the total number of required check-in days, compliant check-in days, and attendance compliance rate; process assessment results include the pass / fail status of four types of tasks: handwritten medical records, skills operation practice, case analysis, and daily assessment. The extracted data is integrated in the format of "trainee ID-data type-value-unit" to output a structured trainee graduation-related dataset.

[0095] Step S503: Input the student graduation-related data and graduation task data into the rule engine algorithm, and verify them one by one according to the priority order of rotation completion rate, attendance compliance rate, process assessment results, course completion comprehensive pass score and credit data. Output the verification results. Students who meet all conditions are marked as qualified, and students who do not meet the conditions are marked as unqualified. List the name of the unqualified condition, the current data value, and the qualified value. The graduation management module inputs the graduation-related data of students and the graduation conditions in the graduation task data into the rule engine algorithm. The algorithm executes the verification logic in the priority order of "rotation completion rate → attendance compliance rate → process assessment results → course completion score → credit data": First, it verifies whether the rotation completion rate reaches 100%. If it does not reach the standard, the corresponding shortcoming item is directly marked. After reaching the standard, it verifies whether the attendance compliance rate is ≥90%, whether all process assessments are qualified, whether the course completion score is ≥60 points, and whether the cumulative credits are ≥50 points. The data comparison results are recorded at each step of the verification. Finally, the verification results are output: students who meet all conditions are marked as "qualified" and students who do not meet the conditions are marked as "unqualified". The module also lists the unqualified shortcoming items in detail (including the name of the unqualified condition, the current data value, and the required value, such as "credit data - current 42 points - required ≥50 points").

[0096] Step S504: Input the personal information of qualified students and certificate template data into the template matching algorithm to generate the initial draft of the graduation certificate data, add an electronic signature to the initial draft of the graduation certificate, and output the electronic graduation certificate data.

[0097] For trainees who have achieved the "qualification" status, the graduation management module extracts their personal information (name, gender, ID number, enrollment date, graduation date, major, and training base name), and inputs it along with the certificate template data from the graduation task data into the template matching algorithm. The algorithm fills in the data according to the correspondence between the preset fields of the template and the trainee's personal information, generating a draft of the graduation certificate (PDF format). Using an electronic signature algorithm based on digital signature technology, the module calls the hospital's electronic seal registered in the storage module to add an electronic signature with a timestamp and digital certificate to the draft of the graduation certificate, ensuring that the certificate content is tamper-proof and the source is traceable, and outputting electronic graduation certificate data that meets the electronic archive standards.

[0098] The graduation management module pushes electronic graduation certificate data to trainees through both platform in-site messages and SMS. Trainees can view and download PDF certificate documents online through the interactive module, and can also print them out using a connected printer. At the same time, the system synchronizes the electronic graduation certificate data to the hospital's file management system through an interface, completes the electronic file archiving according to the directory structure of "year-base-specialty-trainee ID", generates an archiving log (including archiving time, file path, and operator), and outputs a certificate push success indicator and archiving completion data.

[0099] The interaction module receives remedial course application data (including the application reason and expected remedial course time) initiated by students in the "Unsatisfied" status. After the administrator reviews the application through the platform (the review result is pass / rejection, and the reason for rejection must be stated), the graduation management module generates targeted remedial course plan data based on the unsatisfied shortcomings: if the credits are insufficient, the system specifies the courses / teaching activities to be remediated and the corresponding credit values; if the process assessment is unsatisfactory, the system specifies the remedial assessment items and completion deadlines. The remedial course plan data is pushed to the student's end. After the student completes the remedial course, the system automatically collects the updated business data and re-executes the process steps S502-S504 until the student meets all graduation requirements, generates an electronic graduation certificate, and outputs the remedial course closed-loop completion data.

[0100] This invention constructs a full-process medical education management system. It achieves multi-level precise permission and data view matching through a four-dimensional intelligent adaptation model. By leveraging multi-algorithm collaboration, it breaks down the isolation of module data and realizes intelligent optimization of rotation scheduling, teaching-assessment linkage, personalized resource push, intelligent supervision, and automated graduation verification. This significantly reduces manual intervention, improves management efficiency and decision-making scientificity, ensures that the training process conforms to the national medical education syllabus standards, optimizes the quality of medical talent training through precise full-process control, and is adaptable to various medical education scenarios.

[0101] See Figure 2 The present invention also provides an intelligent management system for a medical education platform, including a data acquisition module, a four-dimensional adaptation module, a core business module, a personalized push module, a graduation management module, a storage module, and an interaction module; The data acquisition module is used to collect relevant data from the entire medical education process and transmit it to the storage module. As the core data input of the intelligent medical education management system, it is responsible for comprehensively collecting multi-source data related to the entire medical education process, including user basic data, role attribute data, department and base basic information, national medical education syllabus standard data, historical business data, and real-time business data such as rotation, teaching, and assessment. After data aggregation is completed through standardized data acquisition interfaces, the data is uniformly transmitted to the storage module for archiving, providing complete and standardized basic data support for the operation of all subsequent business modules and ensuring the comprehensiveness and accuracy of the data source.

[0102] The four-dimensional adaptation module is used to call the initialization data set of the storage module, construct a four-dimensional intelligent adaptation model, generate personalized permission configuration results and data views, and transmit them to the interaction module and core business module. It undertakes the core responsibility of intelligent adaptation of permissions and data views. By calling the preprocessed structured initialization data set in the storage module, it constructs a four-dimensional intelligent adaptation model of roles, departments, bases, and levels. With the help of RBAC extended algorithms, multi-label classification algorithms, etc., it generates personalized permission configuration results that accurately match each user. At the same time, it generates personalized data views based on user data needs and preferences. The two types of results are transmitted to the interaction module and the core business module respectively, which not only provides users with customized data display, but also provides a basis for permission verification of core business.

[0103] The core business module is used to execute business related to rotation scheduling, teaching activities, assessment and evaluation, and supervision management, and outputs corresponding business data. It is the business execution center of the system, focusing on the core management scenarios of medical education. Specifically, it is responsible for key business such as intelligent planning and dynamic adjustment of rotation scheduling, organization and resource matching of teaching activities, automated implementation and result analysis of assessment and evaluation, and rule implementation and anomaly rectification of supervision management. It completes the business process closed loop through multi-algorithm collaborative processing, and outputs full business data such as rotation data, teaching data, assessment data, and supervision data, providing core data support for subsequent modules such as personalized push and graduation management.

[0104] The personalized push module is used to call business data from the core business module, process it with algorithms to generate personalized learning resource push results, and transmit them to the interaction module. Focusing on empowering students' personalized learning, it calls business data such as rotation, teaching, and assessment output from the core business module through data interfaces, extracts multi-dimensional features of students, and processes them with algorithms such as logistic regression and incremental learning to build a learning preference model and calculate resource matching degree. It generates personalized learning resource push results that accurately meet the needs of students and transmits them synchronously to the interaction module to be displayed to students, realizing the precise supply of "data-driven" learning resources.

[0105] The graduation management module receives graduation task configuration data, calls relevant business data to complete graduation condition verification and graduation certificate generation, and transmits the data to the interaction module. It manages the entire graduation process for students, first receiving graduation task configuration data input by the administrator through the interaction module, then calling relevant business data such as rotation completion data, assessment scores, and credit data stored in the core business module. Through a rule engine algorithm, it automatically verifies graduation conditions, generates electronic graduation certificates for qualified students, supports remedial procedures for unqualified students, and finally transmits the verification results and electronic graduation certificates to the interaction module, achieving a standardized and automated closed-loop graduation management system.

[0106] The storage module adopts a distributed storage architecture to store various types of collected data, configuration results, business data, and supporting documents, ensuring data security and scalability. This distributed architecture undertakes the storage and management of all system data, specifically storing various types of collected data, personalized permission configuration results, business processing data, learning resource data, graduation certificates, etc. The distributed architecture design not only ensures secure storage, fast retrieval, and efficient reading and writing of massive amounts of data, but also possesses strong horizontal scalability, allowing for flexible expansion according to business growth, providing reliable data storage support for stable system operation.

[0107] The storage module adopts a distributed storage system with a master-slave architecture. Data is sharded according to the hash rule of "training base ID + student ID" and evenly distributed to multiple storage nodes. Each shard is configured with 1 master node and 2 slave nodes. The master node is responsible for data reading and writing, and the slave nodes are responsible for data backup. Automatic failover occurs when the master node fails. The Paxos consensus algorithm is used to ensure data consistency between master and slave nodes. Full data backup is performed daily and incremental data backup is performed hourly. Backup data is stored on an off-site disaster recovery node. At the same time, the module integrates the permission configuration of the four-dimensional adaptation module, which allows only authorized users to access the corresponding data, ensuring high availability, consistency and security of data storage.

[0108] The interaction module provides an interface to display relevant information and receive user commands. After verifying the validity of the commands based on personalized permission configuration results, it transmits them to the corresponding modules. As the system's entry point for interaction with users, it provides a visual interface. On one hand, it displays personalized data views, business processing results, learning resource push content, graduation certificates, and other information to different roles such as administrators, instructors, and students. On the other hand, it receives command instructions from various users (such as configuring graduation tasks, initiating supplementary course applications, and adjusting work schedules). Based on the personalized permission configuration results output by the four-dimensional adaptation module, it verifies the validity of the commands and only transmits valid commands to the corresponding business modules for execution, ensuring operational compliance and system security.

[0109] The system forms a closed-loop data flow across the entire chain. The data acquisition module standardizes the collected multi-source data and transmits it to the storage module. The four-dimensional adaptation module generates personalized permission configurations and data views based on the user and role data in the storage module, which are then transmitted to the interaction module and the core business module, respectively. The core business module completes business processing such as rotation, teaching, assessment, and supervision based on permission constraints. The output business data is stored in the storage module and supports the operation of the personalized push module and the graduation management module. User operation instructions received by the interaction module are transmitted to the corresponding business module for execution after being verified by the four-dimensional adaptation module. The storage module provides data storage, retrieval, and read / write services for all modules in the entire system.

[0110] This invention, by constructing a full-link system architecture, achieves collaborative linkage and closed-loop data flow among various modules, effectively solving the pain points of isolated business modules, poor data flow, and insufficient permission and interaction adaptation in existing technologies. Its core technical effects are reflected in: clear functional positioning of each module and efficient data flow. It ensures data comprehensiveness through a data acquisition module, achieves precise adaptation of permissions and views through a four-dimensional adaptation module, automates the entire business process through core business modules, provides precise learning empowerment through a personalized push module, achieves a standardized closed loop for graduation management through a graduation management module, and ensures stable system operation and convenient operation through storage and interaction modules. The overall architecture ensures the standardization and compliance of medical education management while improving management efficiency and personalized adaptation capabilities, adapting to various medical education scenarios and providing solid system support for intelligent management throughout the entire process.

[0111] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent management method applied to a medical education platform, characterized in that, Includes the following steps: Collect multi-source data related to medical education, and form a structured initial dataset through data cleaning and processing; Based on a structured initialization dataset, a four-dimensional intelligent adaptation model is constructed for roles, departments, bases, and levels. The mapping relationship between users, roles, permissions, and data is established, and personalized permission configuration results and data views are generated. Based on the permission configuration results, relevant data on students, departments and rotation templates are collected. Through the fusion algorithm, intelligent linkage between rotation scheduling, teaching activities and assessment is realized, and business data of each link is output. Based on business data, personalized learning resource push results are generated through feature extraction and model training. At the same time, a supervision rule base is built based on the outline standard, and anomaly detection results and rectification guidance data are output. Based on business data, extract relevant data on student graduation, verify graduation conditions and output the achievement status result, match the certificate template to generate graduation certificate and push it to the student's end.

2. The intelligent management method applied to a medical education platform according to claim 1, characterized in that, The multi-source data includes user basic data, role attribute data, basic information data of departments and bases, national medical education syllabus standard data, and historical business data. The specific steps for constructing the four-dimensional intelligent adaptation model are as follows: Based on user basic data and role attribute data, the mapping relationship between users, roles, permissions and data is established through the RBAC extended algorithm, the basic permission range corresponding to each role is clarified, and the basic mapping rules are output. The user's basic data is used as the input feature, and the preset role labels are used as the output result. The input is fed into the preset multi-label classification model, which automatically assigns initial roles to users and outputs the initial role assignment results. Collect user historical operation data, including menu access frequency, button click count, and data viewing range. Combine this with basic mapping rules to dynamically adjust the permission priority of each role. Prioritize permission items with access frequency higher than a preset threshold and output the permission priority adjustment results. Collect data on the user's department, base, and level; combine the basic mapping rules and permission priority adjustment results to clarify the specific menus, buttons, and data access scope that the user can operate; and generate personalized permission configuration results. Collect historical data access records, combine them with user role attribute data to analyze user data needs and preferences, and generate personalized data views.

3. The intelligent management method for a medical education platform according to claim 1, characterized in that, The specific process for implementing the aforementioned shift rotation schedule is as follows: Based on the permission configuration results, basic student information, department constraint information, and rotation template information are collected and integrated into basic scheduling data. Input basic scheduling data, encode the trainee rotation plan through intelligent scheduling algorithm, generate a preset number of initial scheduling scheme populations by combining department constraint information, and output the optimal intelligent template scheduling scheme through iterative optimization operation; Based on the scheduling plan, preview data is generated from multiple dimensions. When the number of scheduled personnel exceeds the department's capacity, an anomaly is marked and the scheduling preview results are output. The system collects students' entry status, manual registration data, attendance records and assessment results in real time during the rotation process. It tracks the rotation progress through a preset monitoring algorithm and generates an early warning message when the preset early warning trigger conditions are met. The message is then pushed to the corresponding instructors and management users, and the warning data is output. Based on early warning data and rotation adjustment needs, the professional matching status, teaching load and historical evaluation of teaching staff are extracted from the basic scheduling dataset. The optimal teaching staff matching result is solved by optimization algorithm. Combined with the overall arrangement of the current rotation plan and departmental constraint information, the optimal rotation adjustment path is determined by path planning algorithm. The teaching staff matching result and rotation path adjustment plan are output. The system collects data on students' completion of manuals, attendance compliance, assessment scores, and peer review results during their rotations. It then inputs this data into a multi-dimensional weighted scoring algorithm to calculate the overall academic achievement results and generates various retake adaptation plans for students who do not meet the standards.

4. The intelligent management method applied to a medical education platform according to claim 1, characterized in that, The intelligent linkage process between teaching activities and assessment is as follows: Based on the permission configuration results, the trainees' current rotation departments and specialties data, and combined with the knowledge point tags of the National Medical Education Syllabus, the tags are matched with the teaching activity resource library to select suitable teaching activities and corresponding resources, and output suitable activity data. The task binding algorithm is used to associate the adapted activity data with the preset assessment template library, generate binding relationships, clarify the corresponding assessment type and knowledge point scope, and output linked task data. After the teaching activity ends, the bound assessment task is automatically released, the assessment data submitted by the students is collected, encrypted and stored, and the assessment data is output. Based on the assessment data, the ratio of the student's assessment score to the full score of the corresponding knowledge point is calculated to identify the knowledge points that have not been mastered. At the same time, the correlation of teaching activities is analyzed to output mastery data and related results. Based on the mastery data and related results, optimize subsequent collaborative tasks and output the task optimization results.

5. The intelligent management method for a medical education platform according to claim 3, characterized in that, The calculation formula for the multi-dimensional weighted scoring algorithm is as follows: ， in, This indicates the overall score for passing the exam. Let the weight coefficient of the i-th evaluation dimension satisfy the following condition: =1, This represents the score for the i-th evaluation dimension. The evaluation dimensions include the completion rate of the rotation manual, attendance compliance rate, theoretical assessment score, skills assessment score, and peer assessment score. The weight coefficients of each dimension are preset according to the national medical education syllabus standards and the hospital's personalized training requirements, and can be manually adjusted by the administrator through the platform.

6. The intelligent management method applied to a medical education platform according to claim 1, characterized in that, The business data includes rotation management-related data, teaching activity-related data, and assessment and evaluation-related data. The specific steps for generating personalized learning resource push results are as follows: Based on business data, extract student majors, enrollment time, weak knowledge points, historical learning resource type preferences, learning duration distribution, and assessment score distribution, and output feature vectors; The feature vector is used as input, and the learning completion rate of the learner on the learning resources is used as the output label. The training set and the test set are divided. The learning preference model is constructed by logistic regression algorithm, and the model parameters are optimized by gradient descent algorithm to minimize the square error between the model prediction value and the actual value. When the accuracy of the model on the test set is higher than the threshold, the training is stopped and the trained learning preference model is output. Collect resource data from the learning resource library, filter relevant resources based on students' major data, and output the filtered resource data. Based on feature vectors and filtered resource data, combined with a learning preference model, the matching degree between learners and each resource is calculated using a matching degree calculation formula. The resources are sorted from high to low matching degree, and multiple resources are selected to generate a recommended resource list, outputting the initial recommendation results. The algorithm collects real-time data on students' actions with the initial recommendation results, inputs it into the incremental learning algorithm, updates the learning preference model parameters, re-calculates the matching degree according to a set period, dynamically adjusts the recommended resource list, and outputs optimized personalized learning resource push results.

7. The intelligent management method for a medical education platform according to claim 6, characterized in that, The matching degree calculation formula is as follows: ， in, This represents the matching degree between the k-th student and the j-th learning resource. , , These represent the knowledge point fit coefficient, difficulty suitability coefficient, and learning preference coefficient, respectively, satisfying the following conditions: , This represents the degree of fit between the weak knowledge points of the k-th student and the knowledge points covered by the j-th learning resource. It is obtained by calculating the proportion of the number of elements in the intersection of the student's weak knowledge point set and the resource's knowledge point set to the total number of elements in the student's weak knowledge point set. This represents the fit between the difficulty level of the j-th learning resource and the knowledge level of the k-th student. Both the resource difficulty level and the student's knowledge level are divided into 5 levels. The fit is 1 when the levels are exactly the same, 0.8 when the levels differ by 1 level, 0.5 when the levels differ by 2 levels, and 0.2 when the levels differ by 3 levels or more. This represents the historical learning preference coefficient of the k-th student for the j-th type of learning resources. It is obtained by calculating the weighted average of the student's historical click count, learning completion rate, and learning time percentage for this type of learning resource.

8. The intelligent management method applied to a medical education platform according to claim 2, characterized in that, The specific steps for constructing the supervision rule base are as follows: Based on the national medical education syllabus standards data, management requirement data for training bases, professional bases, departments, and trainees are extracted to construct a supervision rule base; The supervisory rule base is input into the rule engine algorithm, which transforms the rules into executable logical expressions. Business data from the platform is received in real time. When the business data meets the triggering conditions in the logical expression, supervisory task data containing the supervisory object, supervisory type, triggering reason, supervisory requirements, and completion deadline is generated and the supervisory task is output. The business data corresponding to the supervision task is transformed into a feature vector, input into the isolated forest algorithm, an isolated tree forest is constructed, the anomaly score of each data point is calculated, and when the anomaly score is higher than the threshold, it is judged as an anomaly, marked as a key supervision object, and the anomaly data is output. Anomaly data is input into a knowledge graph algorithm to construct an association graph of anomalies, rectification measures, and corresponding modules. Nodes are divided into anomaly nodes, rectification measure nodes, and system module nodes. Edges represent the relationships between nodes. The edge between anomaly nodes and rectification measure nodes represents the corresponding rectification plan. The edge between rectification measure nodes and system module nodes represents the system module that needs to be operated. The association graph data is output. Receive rectification operation data initiated by the administrator based on the association graph data, including the rectification start time and rectification operation content, track the rectification progress in real time, record the rectification completion time, and after the rectification is completed, re-input the updated business data into the isolated forest algorithm to re-detect anomalies until the anomalies are eliminated, and output the supervision rectification closed loop completion data.

9. The intelligent management method applied to a medical education platform according to claim 1, characterized in that, The specific steps for verifying the graduation conditions and outputting the compliance status result are as follows: Receive the graduation task configuration data input by the administrator, including task name, task start and end time, participant screening criteria, graduation conditions and certificate template data, and output the graduation task data; Based on business data, extract students' rotation completion data, course completion scores, credit data, attendance data, and process assessment results data, and output students' graduation-related data; The algorithm inputs the student's graduation-related data and graduation task data into the rule engine algorithm, and verifies them one by one in the priority order of rotation completion rate, attendance compliance rate, process assessment results, course completion comprehensive pass score and credit data. The verification results are output, and students who meet all conditions are marked as qualified, and students who do not meet the conditions are marked as unqualified. The algorithm also lists the name of the unqualified condition, the current data value, and the qualified value. The personal information of qualified students and certificate template data are input into the template matching algorithm to generate the initial draft of the graduation certificate data. An electronic signature is added to the initial draft of the graduation certificate, and the electronic graduation certificate data is output.

10. An intelligent management system applied to a medical education platform, characterized in that, It includes a data acquisition module, a 4D adaptation module, a core business module, a personalized push module, a graduation management module, a storage module, and an interaction module; The data acquisition module is used to collect relevant data from the entire medical education process and transmit it to the storage module. The four-dimensional adaptation module is used to call the initial data set of the storage module, construct a four-dimensional intelligent adaptation model, generate personalized permission configuration results and data views, and transmit them to the interaction module and core business module. The core business module is used to perform business related to rotation scheduling, teaching activities, assessment and evaluation and supervision management, and output corresponding business data. The personalized push module is used to call the business data of the core business module, process it with an algorithm to generate personalized learning resource push results and transmit them to the interaction module. The graduation management module is used to receive graduation task configuration data, call relevant business data to complete graduation condition verification and graduation certificate generation, and transmit it to the interaction module. The storage module adopts a distributed storage architecture to store various types of collected data, configuration results, business data, and supporting documents, ensuring data security and scalability. The interaction module is used to provide an operation interface, display relevant information and receive user operation instructions. After verifying the legality of the instructions based on the personalized permission configuration results, it transmits them to the corresponding module.

Images