Personalized hierarchical teaching method and system for higher education based on artificial intelligence

Abstract

The invention relates to a higher education personalized hierarchical teaching method and system based on artificial intelligence, and the method comprises the steps: obtaining multi-dimensional learning data, carrying out the time-space alignment processing, and generating a synchronous multi-dimensional data set; performing spatial-temporal feature fusion and time sequence modeling on the data set by using a deep neural network, and constructing a dynamic student portrait; analyzing knowledge mastery degree features in the portrait through a semantic analysis model, and generating a personalized resource recommendation sequence in combination with the knowledge graph; based on the sequence and the portrait, planning a personalized learning path by using a path reasoning algorithm; and carrying out teaching hierarchy binding on the personalized resource recommendation sequence and the learning path to form a hierarchical teaching scheme. Accurate teaching support is provided for individual differences of students, and the teaching effect and learning experience are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of smart education, and in particular relates to an artificial intelligence-based personalized stratified teaching method and system for higher education. Background Art

[0002] With the development of artificial intelligence applications in education, personalized teaching technologies are gradually emerging. Knowledge graphs and adaptive algorithms provide support for precision teaching. Traditional higher education primarily relies on teacher experience for stratified instruction, crudely categorizing students into different levels based on performance, and then matching them with fixed learning resources and pathways. These traditional approaches rely on a single stratification basis, with resource recommendations relying on simple rules or grade rankings, and learning pathway planning ignoring students' real-time learning status. For example, knowledge mastery is assessed solely through test scores, without integrating multidimensional data such as learning behavior and practice trajectories, resulting in a static student profile. Resource recommendations fail to consider knowledge point connections and cognitive differences, and pathway planning fails to account for changes in learning load, making it difficult to adapt to individual needs. Existing teaching methods present significant challenges: crude data processing, ineffective integration of multidimensional learning data, inability to dynamically update student profiles, and delayed teaching adjustments. Recommendation accuracy is insufficient, resource matching fails to deeply analyze knowledge mastery details and the logical relationships within the knowledge graph, and recommendation sequences lack scientific rationality. Furthermore, the stratification mechanism is rigid, and the binding of teaching levels to learning resources and pathways relies on manual pre-setting, failing to intelligently adjust based on students' dynamic performance. This makes it difficult to meet the personalized needs of higher education, leading to irrational allocation of teaching resources and low student learning efficiency. Summary of the Invention

[0003] Based on this, it is necessary to provide an artificial intelligence-based personalized stratified teaching method and system for higher education that can solve the above technical problems.

[0004] In the first aspect, this application provides a personalized and layered teaching method for higher education based on artificial intelligence, including:

[0005] Acquire multi-dimensional learning data and perform spatiotemporal alignment to generate synchronized multi-dimensional datasets;

[0006] Using deep neural networks to fuse spatiotemporal features and perform time series modeling on synchronized multidimensional datasets to obtain dynamic student portraits;

[0007] Use semantic analysis models to analyze the knowledge mastery characteristics of dynamic student portraits, and combine them with knowledge graphs to generate personalized resource recommendation sequences;

[0008] Based on personalized resource recommendation sequences and dynamic student portraits, a path inference algorithm is used to generate personalized learning paths;

[0009] Bind the personalized resource recommendation sequence and personalized learning path to the teaching level to generate a tiered teaching plan.

[0010] In one embodiment, a semantic analysis model is used to analyze the knowledge mastery characteristics of dynamic student portraits, and combined with the knowledge graph to generate a personalized resource recommendation sequence, including:

[0011] Use semantic analysis models to analyze the knowledge mastery characteristics of dynamic student portraits and generate weak knowledge point information;

[0012] Match target learning resources associated with weak knowledge points from the pre-set learning resource library;

[0013] The target learning resources are sorted using a preset priority sorting algorithm to generate a personalized resource recommendation sequence.

[0014] In one embodiment, the path inference algorithm is constructed using the following formula:

[0015]

[0016] Where μ represents the cognitive load weight coefficient, C(t) = w1·e -t / τ +w2·sigmoid(ΔK t ) represents the cognitive load value at time t, w1 and w2 represent the weights of time decay and knowledge point transition, τ represents the learning fatigue time constant, ΔK t represents the jump distance between adjacent knowledge points in the knowledge graph, C max represents the preset maximum tolerable cognitive load, v represents the knowledge span penalty coefficient, dist(k t , k t-1 ) represents the cognitive span between knowledge points, λ represents the knowledge coverage reward coefficient, S g (π) represents the coverage score of level g, and G represents the set of teaching objective levels in the knowledge graph.

[0017] In one embodiment, the personalized resource recommendation sequence and the personalized learning path are bound to the teaching level to generate a tiered teaching plan, including:

[0018] Based on the multi-level node structure of the knowledge graph, the learning nodes in the personalized learning path are divided into levels, and the resources in the personalized resource recommendation sequence are mapped to the corresponding levels;

[0019] Generate hierarchical teaching plans based on the teaching objectives of each teaching level;

[0020] Integrate hierarchical teaching plans, stratified personalized learning paths, and mapped personalized resources to generate stratified teaching plans.

[0021] In one embodiment, the method further comprises:

[0022] Based on the temporal changes in the characteristics of knowledge mastery, calculate the deviation from the teaching objectives;

[0023] When the deviation exceeds the preset threshold, the node topology relationship of the teaching level is recalculated in the knowledge graph, and the sorting priority of the personalized resource recommendation sequence is updated;

[0024] Adjust the cognitive load weight coefficient μ and the knowledge coverage reward coefficient λ in the path reasoning algorithm according to the deviation direction of the deviation degree;

[0025] Based on the updated sorting priorities and the adjusted path reasoning algorithm parameters, a new hierarchical teaching plan is iteratively generated.

[0026] In one embodiment, the method further comprises:

[0027] Based on the multi-level node structure of the knowledge graph, the personalized learning path is converted into a visual cognitive trajectory heat map;

[0028] Quantitatively model knowledge mastery characteristics and generate dynamic dashboards with historical comparisons and forecast curves;

[0029] The adaptive rendering engine is used to integrate the trajectory heat map with the dynamic dashboard in time and space to generate a visual interface for teaching decision-making.

[0030] In one embodiment, obtaining multi-dimensional learning data and performing spatiotemporal alignment to generate a synchronized multi-dimensional dataset includes:

[0031] Use data parsing technology to extract the timestamp and spatial identifier of learning data;

[0032] Perform time series calibration on the learning data according to the timestamp to obtain the time series calibrated learning data;

[0033] The time-series-calibrated learning data is spatially aligned according to the spatial identifier to generate a synchronized multidimensional dataset.

[0034] Secondly, this application also provides an artificial intelligence-based personalized tiered teaching system for higher education, including:

[0035] Data acquisition and processing module, used to acquire multi-dimensional learning data and perform spatiotemporal alignment to generate synchronized multi-dimensional data sets;

[0036] The feature fusion modeling module is used to use deep neural networks to perform spatiotemporal feature fusion and time series modeling on synchronized multidimensional data sets to obtain dynamic student portraits;

[0037] The semantic knowledge module is used to analyze the knowledge mastery characteristics of dynamic student portraits using a semantic analysis model and generate personalized resource recommendation sequences based on the knowledge graph;

[0038] The learning path planning module is used to generate personalized learning paths based on personalized resource recommendation sequences and dynamic student profiles using path reasoning algorithms;

[0039] The teaching plan generation module is used to bind the personalized resource recommendation sequence and personalized learning path to the teaching level to generate a hierarchical teaching plan.

[0040] On the third aspect, the present application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned artificial intelligence-based personalized stratified teaching method for higher education.

[0041] In a fourth aspect, the present application also provides a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the above-mentioned artificial intelligence-based personalized stratified teaching method for higher education.

[0042] The above-mentioned AI-based personalized stratification method and system for higher education, by aligning multi-dimensional learning data in time and space and using deep neural network modeling to form dynamic student portraits, can accurately portray student status in real time, solving the problems of insufficient traditional data integration and static portraits; combining semantic analysis with knowledge graphs to generate personalized resource recommendation sequences, can accurately match resources based on students' knowledge weaknesses and knowledge associations, and improve the scientific nature of recommendations; with the help of path reasoning algorithms to generate learning paths, it can balance cognitive load and knowledge span and optimize learning efficiency; binding resources and paths to knowledge graph teaching levels to realize the automatic generation of stratified teaching plans, solve the problem of traditional stratification solidification, adapt to student differences and the complexity of the knowledge system, and overall improve the accuracy of personalized teaching in higher education and resource allocation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is a flow chart of a personalized layered teaching method for higher education based on artificial intelligence of the present invention;

[0045] Figure 2 This is a structural diagram of the higher education personalized stratified teaching system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0047] The present application discloses an artificial intelligence-based personalized stratified teaching method and system for higher education. The implementation environment includes a hardware architecture consisting of a data acquisition terminal, a server, a storage device, and a visualization terminal. The hardware is connected to each other through a wired or wireless network. When used, the data acquisition terminal obtains students' multi-dimensional learning data in real time and transmits it to the server. The server calls the algorithm model through the memory to perform spatiotemporal alignment, feature fusion, and time series modeling on the data, generate dynamic student portraits, personalized resources, and learning paths, and feed back results such as stratified teaching plans, visualized cognitive trajectory heat maps, and dynamic dashboards to the visualization terminal.

[0048] In one embodiment, Figure 1 As shown, a personalized and layered teaching method for higher education based on artificial intelligence is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0049] S101, obtain multi-dimensional learning data and perform spatiotemporal alignment to generate a synchronized multi-dimensional dataset.

[0050] Multidimensional learning data can be acquired through data collection terminals such as learning platforms, online testing systems, and teaching management systems. This includes, but is not limited to, behavioral data such as learning time, performance data such as homework grades, and environmental data such as learning device type. It also includes data at different time granularities, such as real-time operation logs and periodic test results, and spatial attributes, such as different course modules and knowledge points. After spatiotemporal alignment, this data is integrated into a synchronized, spatiotemporally unified multidimensional dataset, providing the data foundation for subsequent processing.

[0051] S102, using deep neural networks to perform spatiotemporal feature fusion and time series modeling on synchronized multidimensional datasets to obtain dynamic student portraits.

[0052] Deep neural networks map the spatial identifiers of synchronized multidimensional datasets, such as knowledge points and course modules, into low-dimensional vector spaces. Graph neural networks or knowledge graph embedding techniques can be used to capture hierarchical relationships between knowledge points and predecessor-successor dependencies, such as the sequential learning relationship from calculus to linear algebra. Multi-attribute data within the same time step can be integrated through fully connected layers or convolutional neural networks, such as associating video viewing time with test scores on that knowledge point, to generate feature vectors in the spatial dimension. Sequence models such as recurrent neural networks or long-short-term memory networks can be used to process the time series data of synchronized multidimensional datasets to capture the temporal continuity of learning behavior, such as frequent recent visits to a knowledge point, reflecting weaknesses. By fusing the spatial feature vectors with the time series data, a joint spatiotemporal feature vector is generated, resulting in a dynamic student portrait that reflects knowledge mastery, learning behavior, and temporal dynamics, providing a data foundation for subsequent personalized resource recommendations and learning path planning.

[0053] S103: Use the semantic analysis model to analyze the knowledge mastery characteristics of the dynamic student portrait, and combine it with the knowledge graph to generate a personalized resource recommendation sequence.

[0054] The semantic analysis model can be a natural language processing model, which converts numerical features of knowledge mastery in the dynamic student profile, such as mastery scores and error rates, into semantic descriptions of knowledge gaps. For example, a test score of <60 on knowledge point X and errors after >3 practice attempts is marked as a weak point. Combined with learning behaviors, such as repeatedly watching a basic explanation video for knowledge point Y, it can be inferred that the student's understanding of the basic concepts of Y is weak. The knowledge graph structure can be: nodes can be knowledge points such as limits, derivatives, and integrals; learning resources such as Video A on the definition of derivatives and Exercise Book B on integral calculations; and resource types such as videos, documents, and exercises. Edges can be relationships between knowledge points, such as derivatives being a prerequisite for integrals; connections between resources and knowledge points, such as Exercise Book B containing three integral calculation knowledge points; and resource difficulty levels, such as basic / advanced / high-level. Based on a weak point, such as matrix inversion, the knowledge graph can be searched for directly related resources, such as a video demonstrating the derivation of the matrix inversion formula and 30 matrix inversion exercises, to generate a personalized resource recommendation sequence.

[0055] S104, based on the personalized resource recommendation sequence and dynamic student portrait, a personalized learning path is generated using a path reasoning algorithm.

[0056] The essence of the path inference algorithm is to plan an optimal path from students' current knowledge state to the teaching objectives. This can be achieved by controlling cognitive load to avoid learning fatigue caused by overload. Knowledge span optimization avoids skipping learning paths, ensuring a learning path that focuses on the basics before advancing. Knowledge coverage rewards encourage learning paths that cover both the basics and applications, rather than simply repeating basic content, to form an orderly resource learning sequence. For example: 1. Review matrix definitions: low-load, consolidate the foundation; 2. Watch a video explaining the basics of matrix inversion: connect to current weaknesses, and learn a moderate learning span; 3. Complete simple matrix inversion exercises: immediate practice to maintain a balanced learning load; 4. Learn advanced techniques for block matrix inversion: advanced content with a slightly higher level of difficulty. This creates a personalized learning path, guiding students on what to learn first and what to learn next. Each step is tailored to the student's ability and status, making learning more efficient and less stressful.

[0057] S105: Bind the personalized resource recommendation sequence and the personalized learning path to the teaching level to generate a tiered teaching plan.

[0058] Among them, the generated personalized resource recommendation sequence and learning path are integrated and bound according to the teaching hierarchy structure of the knowledge system, such as basic, advanced, and high-level, to form a tiered teaching plan for different student levels. This aligns the personalized learning content with the hierarchical goals of the knowledge system, solving the problem of extensive hierarchical division and disconnected resource paths in traditional tiered teaching. Hierarchical division can be carried out according to the nodes in the knowledge graph, and each level corresponds to a clear teaching goal, such as the basic level requires understanding of concepts, the advanced level requires proficient application, and the high-level level requires comprehensive innovation. The personalized resource recommendation sequence and personalized learning path are matched to the corresponding level, and a corresponding teaching plan is formulated for each level, which is integrated to generate an overall learning plan. This preserves the individual differences of students and ensures that the teaching content conforms to the scientific hierarchy of the knowledge system.

[0059] The AI-based personalized stratified teaching method for higher education acquires multidimensional learning data and performs spatiotemporal alignment processing to generate synchronized multidimensional datasets. This effectively integrates multi-source heterogeneous data, such as learning behaviors and practice trajectories, addressing the issues of extensive data processing and ineffective integration of multidimensional data in traditional methods. Deep neural networks are used to fuse spatiotemporal features and perform time series modeling on the dataset to construct dynamic student profiles, accurately depicting the dynamic changes in students' knowledge mastery, learning behaviors, and cognitive states. A semantic analysis model is used to analyze the knowledge mastery features in the profiles. Combined with the knowledge graph, related learning resources are matched and sorted to generate personalized resource recommendation sequences, improving the scientific nature and accuracy of resource recommendations. Based on the recommendation sequences and dynamic profiles, a path inference algorithm is used to plan personalized learning paths, optimize learning efficiency, and adapt to individual needs. The personalized resource recommendation sequences are bound to the learning paths at the teaching level. Learning node levels are divided and resources are mapped based on the multi-level node structure of the knowledge graph to generate stratified teaching plans, achieving intelligent and dynamic matching of teaching levels with learning resources and paths. This addresses the rigidity and reliance on manual pre-setting of traditional stratification mechanisms, provides precise teaching support tailored to individual student differences, and effectively improves teaching effectiveness and learning experience.

[0060] In one embodiment, a semantic analysis model is used to analyze the knowledge mastery characteristics of dynamic student portraits, and combined with the knowledge graph to generate a personalized resource recommendation sequence, including:

[0061] S201, using the semantic analysis model to analyze the knowledge mastery characteristics of the dynamic student portrait and generate weak knowledge point information;

[0062] S202, matching target learning resources associated with weak knowledge point information from a preset learning resource library;

[0063] S203: Rank the target learning resources using a preset priority ranking algorithm to generate a personalized resource recommendation sequence.

[0064] Specifically, the semantic analysis model extracts students' mastery deficiencies in specific knowledge points from portrait data containing multi-dimensional features such as knowledge mastery, learning behavior, and cognitive load. Structured data containing information such as the name of the weak knowledge point, mastery score, and associated error type is generated to form weak knowledge point information. Using the weak knowledge point information as an index, all directly or indirectly matching related learning resources in the pre-set learning resource library are retrieved. For example, if the weak knowledge point is matrix inversion, the resources marked as matrix inversion in the knowledge graph are matched. At the same time, through prerequisite relationships such as matrix definition is the basis of matrix inversion and extended relationships such as the application of matrix inversion in engineering, the matching of related pre-review resources and post-application resources is expanded to form target learning resources. The priority ranking algorithm comprehensively considers multiple personalized factors to weight candidate resources and scores them. These factors may include: knowledge relevance: calculating the semantic match between resource content and weak knowledge points, such as videos specifically explaining the steps of matrix inversion being more relevant than linear algebra overview documents; cognitive adaptability: adjusting resource difficulty based on the current knowledge level of the student profile, such as reducing the weight of advanced resources for students with weak foundations; learning preferences: combining historical student behavior data, such as preferences for video resources, to increase the ranking priority of video resources; efficiency indicators: referencing the historical learning effects of similar resources, such as the average mastery improvement rate of a particular exercise set. The priority ranking algorithm calculates and ranks candidate resources, generating a resource recommendation sequence that matches individual student characteristics, such as prioritizing the learning order of basic explanation videos → corresponding exercises → advanced cases. This ensures that resource matching accurately targets weak points while also meeting students' learning habits and efficiency needs.

[0065] In one embodiment, at S301, the path inference algorithm is constructed using the following formula:

[0066]

[0067] Where μ represents the cognitive load weight coefficient, C(t) = w1·e -t / τ +w2·sigmoid(ΔK t ) represents the cognitive load value at time t, w1 and w2 represent the weights of time decay and knowledge point transition, τ represents the learning fatigue time constant, ΔK t represents the jump distance between adjacent knowledge points in the knowledge graph, C max represents the preset maximum tolerable cognitive load, v represents the knowledge span penalty coefficient, dist(k t , k t-1 ) represents the cognitive span between knowledge points, λ represents the knowledge coverage reward coefficient, S g (π) represents the coverage score of level g, and G represents the set of teaching objective levels in the knowledge graph.

[0068] Exemplarily, the cognitive load weight coefficient μ is an empirical parameter preset based on the cognitive load theory in educational psychology. It can be calibrated through the correlation analysis of load changes and learning efficiency in historical learning data to adjust the importance of cognitive load in the objective function. The larger μ is, the more priority is given to avoiding high-load learning, such as increasing this coefficient for fatigue-sensitive students. The time decay weight w1 fits the decay law that the longer the student's learning time, the higher the fatigue level. It can be determined through regression analysis of historical learning time and error rate, and is used to characterize the natural accumulation of fatigue over time. The knowledge point jump weight w2 is based on the association relationship of knowledge points in the knowledge graph. It quantifies the impact of knowledge point jumps on cognitive load through students' performance data across knowledge points, such as the decline in accuracy when learning across chapters. The learning fatigue time constant τ is a fatigue cycle parameter obtained through group learning data statistics, which is used to control the time decay rate. The preset maximum tolerable cognitive load C max As the safety upper limit of cognitive load, it is used to force the insertion of low-load learning content to reduce the load when the cognitive load value C(t) is close. The knowledge span penalty coefficient v penalizes the cognitive span dist(k t , k t-1 ) Ensure that the learning path follows the logic of first basic and then advanced. The jump distance between adjacent knowledge points ΔK t Quantify the logical span of knowledge point transitions, which affects the knowledge point transition load in C(t). That is, the larger the jump distance, the higher the load increment. The knowledge coverage reward coefficient λ is preset according to the importance of each level of goals in the syllabus, which encourages the learning path to cover more teaching goal levels and avoids the path being limited to a single difficulty level. The coverage score S of level g g (π) is calculated based on the number and importance of nodes at the level g contained in the learning path π. This is used to measure the path's coverage of each teaching objective level in the knowledge graph, guiding the generation of a step-by-step path from basic to advanced to high-level. The set of teaching objective levels G in the knowledge graph is predefined based on the multi-level node structure of the knowledge graph. This serves as the target space for path planning, ensuring that the generated learning path conforms to the hierarchical structure of the knowledge system. This generates personalized learning paths with moderate load, logical coherence, and complete objectives, tailored to the individual student's cognitive characteristics and the knowledge graph structure.

[0069] In one embodiment, the personalized resource recommendation sequence and the personalized learning path are bound to the teaching level to generate a tiered teaching plan, including:

[0070] S401, based on the multi-level node structure of the knowledge graph, divide the learning nodes in the personalized learning path into levels, and map the resources in the personalized resource recommendation sequence to the corresponding levels;

[0071] S402, generating a level-by-level teaching plan based on the teaching objectives of each level;

[0072] S403, integrating the hierarchical teaching plan, the stratified personalized learning path and the mapped personalized resources to generate a stratified teaching plan.

[0073] Specifically, based on the multi-level node structure of the knowledge graph, such as the preset three-level architecture of basic layer → advanced layer → high layer, each learning node in the personalized learning path, such as a specific knowledge point or learning task, is mapped to the corresponding teaching level. If the matrix definition node in the knowledge graph belongs to the basic layer, the block matrix inversion node belongs to the advanced layer, and the matrix application node in machine learning belongs to the high layer, then mastering the matrix definition in the learning path corresponds to the basic layer, and practicing block matrix inversion corresponds to the advanced layer. The basis for division is: the hierarchical label of the node in the knowledge graph, that is, it is determined by the difficulty level, prerequisite relationship and other attributes pre-defined when the graph is constructed. For example: the basic layer is entry-level knowledge points such as core concepts and formula derivation; the advanced layer is intermediate difficulty knowledge points such as method application and comprehensive calculation; the high layer is high-level knowledge points such as cross-domain expansion and innovative practice. Each resource, such as a video, exercise, or document, is extracted from the personalized resource recommendation sequence and mapped to the corresponding knowledge point's hierarchy through the resource-knowledge point relationship in the knowledge graph. The resource's hierarchical label inherits the hierarchy of its associated knowledge point, ensuring strict alignment between resource difficulty and knowledge point hierarchy. This transforms scattered learning nodes and resources into hierarchically ordered teaching elements. The teaching objectives for each level can be: The goal for the basic level is to understand core concepts, master basic formulas, and simple applications; the goal for the advanced level is to master the application of methods to solve moderately complex problems and form a knowledge network; and the goal for the advanced level is to achieve comprehensive cross-disciplinary application and innovative solutions to practical problems. Specific implementation strategies are developed for each level's teaching objectives. For example, the learning strategy for the basic level utilizes theoretical explanations and low-level exercises; for the advanced level, case analysis and specialized training; and for the advanced level, project-based practice and research-based learning. Time management for the basic level is 40-60 minutes per knowledge point, with no more than three new knowledge points added daily. For the advanced level, the learning time for each knowledge point is 60-90 minutes, with a 20-minute summary. For the basic level, concept mastery is assessed by accuracy of ≥60%, for the advanced level, application ability is assessed by completion of comprehensive questions of ≥70%, and for the advanced level, overall proficiency is assessed through project reports / innovation solutions. The personalized learning paths, mapped personalized resources, and tiered teaching plans are integrated according to the tiered dimensions to form a tiered teaching plan with clear objectives, specific strategies, and manageable progress.

[0074] In one embodiment, the method further comprises:

[0075] S501, calculating the deviation from the teaching objective based on the temporal changes of the knowledge mastery characteristics;

[0076] S502: When the deviation exceeds a preset threshold, the node topology relationship of the teaching level is recalculated in the knowledge graph, and the sorting priority of the personalized resource recommendation sequence is updated;

[0077] S503, adjusting the cognitive load weight coefficient μ and the knowledge coverage reward coefficient λ in the path reasoning algorithm according to the deviation direction of the deviation degree;

[0078] S504 , iteratively generating a new hierarchical teaching plan based on the updated sorting priority and the adjusted path reasoning algorithm parameters.

[0079] Specifically, the student's knowledge mastery at different time points is compared with the teaching objectives. The difference between the two is calculated using some quantitative method to obtain the deviation. The degree of deviation can be measured using methods such as difference or ratio. For example, if the teaching objective requires a student to achieve 80% mastery of a certain knowledge point, and the student's actual mastery is 60%, the deviation can be expressed as 20%. A preset threshold is set based on teaching experience or experimental data. When the deviation exceeds the preset threshold, it indicates that the student's learning progress has significantly deviated from expectations, and the teaching hierarchy arrangement needs to be reviewed. By recalculating the node topology of the teaching hierarchy in the knowledge graph, a learning path and knowledge point sequence that is more suitable for the student's current situation can be found. For example, if a student's mastery of a certain foundational knowledge point is weak, the learning sequence of subsequent knowledge points may need to be adjusted to strengthen the study of related foundational knowledge points. When a student's learning progress deviates significantly, the original resource recommendations may no longer be applicable. Based on the recalculated node topology, the resources in the personalized resource recommendation sequence are reordered. This will help students fill knowledge gaps and achieve teaching objectives by placing resources at the top, thereby improving the relevance of resource recommendations. Deviation refers to whether a student's knowledge mastery is above or below the teaching objective, requiring different instructional adjustment strategies. The cognitive load weight coefficient controls the degree to which cognitive load influences path planning during the learning process. If a student's knowledge mastery is below the teaching objective, the cognitive load weight coefficient may need to be appropriately lowered to allow them to focus more on addressing their knowledge gaps and avoid poor learning outcomes due to excessive learning pressure. If a student's knowledge mastery is above the teaching objective, the cognitive load weight coefficient can be appropriately increased to increase the learning challenge and promote further improvement. The knowledge coverage bonus coefficient is used to incentivize learning paths to cover more levels of the teaching objective. When a student's knowledge mastery is below the teaching objective, the knowledge coverage bonus coefficient can be increased to encourage them to expand their learning scope while addressing their current knowledge gaps, accelerating progress towards the teaching objective. When a student's knowledge mastery is above the teaching objective, the knowledge coverage bonus coefficient can be lowered to encourage them to focus more on in-depth learning and expanding their knowledge. This approach takes into account the student's current learning status, the recalculated node topology, and the adjusted algorithm parameters. By continuously iteratively updating the learning plan based on the student's learning progress, a new hierarchical teaching plan can be generated to better adapt to individual student differences and improve teaching effectiveness.

[0080] In one embodiment, the method further comprises:

[0081] S601, based on the multi-level node structure of the knowledge graph, converts the personalized learning path into a visual cognitive trajectory heat map;

[0082] S602: Quantitative modeling of knowledge mastery characteristics is performed to generate a dynamic dashboard including historical comparison and prediction curves;

[0083] S603: Using an adaptive rendering engine, the trajectory heat map and the dynamic dashboard are temporally and spatially integrated to generate a visual interface for teaching decision making.

[0084] For example, the horizontal axis could be the knowledge graph's hierarchical structure, and the vertical axis could be time. The color of each coordinate point represents the student's learning intensity at that level, such as the number of practice attempts, error rate, and duration of study. Heatmap algorithms, such as Gaussian blur, are used to render node density, creating learning hotspots—for example, concentrated learning trajectories of foundational knowledge points within a specific time period. This visually illustrates the student's progress through the knowledge graph, allowing for rapid identification of high-frequency learning areas (i.e., strengths) and low-frequency areas (i.e., weaknesses), providing a spatially intuitive reference for instructional interventions. Knowledge mastery characteristics, such as knowledge point accuracy, cognitive load, and knowledge network completeness, are extracted from dynamic student profiles and converted into quantifiable metrics, such as a scale of 0 to 100. A historical comparison curve plots mastery changes over time: for example, mastery of a linear algebra module increased from 60% to 75% over the past seven days, with different colors used to distinguish different knowledge points or modules. A prediction curve uses a time series model to predict future mastery trends and annotate warning thresholds, such as displaying a yellow alert when predicted mastery falls below the teaching target. This transforms abstract knowledge states into trackable, quantifiable data, enabling real-time monitoring of learning progress. Learning patterns can be discovered through historical comparisons, and potential risks can be identified through predictions. For example, at the current pace, students may not meet the final target. This integration integrates spatiotemporal correlations. Temporal correlation aligns the timestamps of learning trajectories in the heatmap with the dashboard's timeline. Clicking a foundational layer node at a specific time point in the heatmap displays detailed mastery of the foundational layer knowledge point during that period. For example, clicking the foundational layer in Week 3 displays a 35% error rate for exercises. Spatial correlation links the knowledge graph's hierarchical structure with the dashboard's module categories. For example, clicking an advanced module in the dashboard highlights all learning node trajectories corresponding to that module in the heatmap. The adaptive rendering engine dynamically adjusts the display granularity based on user role, such as teacher / student. For teachers, a heatmap overlay is displayed for the entire class, highlighting common areas of weakness. For students, the focus is on individual trajectories, with nodes associated with personalized resource recommendations marked, such as "Recommended Video A" next to a red node. Historical trajectories can be viewed by dragging the timeline, and detailed learning records, such as specific incorrect questions and resource access time, can be viewed by clicking a node. Through the three steps of trajectory visualization → state quantification → space-time fusion, the accuracy and decision-making efficiency of personalized teaching can be improved.

[0085] In one embodiment, obtaining multi-dimensional learning data and performing spatiotemporal alignment to generate a synchronized multi-dimensional dataset includes:

[0086] S701, extracting the timestamp and spatial identifier of the learning data using data parsing technology;

[0087] S702, performing time sequence calibration on the learning data according to the timestamp to obtain time sequence calibrated learning data;

[0088] S703 , performing spatial calibration on the time-series-calibrated learning data according to the spatial identifier to generate a synchronized multidimensional data set.

[0089] Specifically, data parsing techniques are used to extract key metadata from learning data from various sources, such as platform logs, operating system data, and test scores. Timestamps can be extracted from log entries or operation records, representing the precise time the data was generated, such as 2025-04-25-15:30:00. Spatial identifiers can be used to label the location of the knowledge system or teaching scenario to which the data belongs, such as Course ID = 202301 - Chapter = 3 - Knowledge Point = Matrix Inversion - Resource Type = Lab Course - Module = Circuit Analysis. This establishes dual time-space anchors for multidimensional data, enabling spatiotemporal localization of fragmented data. Through time format unification, missing data processing, and time series sorting, temporal alignment is performed, transforming fragmented time point data into continuous time series, supporting subsequent analysis of temporal dependencies in learning behavior. Based on the pre-defined knowledge graph structure (e.g., course → chapter → knowledge point), spatial identifiers are associated with specific nodes. For example, if the spatial identifier is Course A - Chapter 2 - Knowledge Point B, it is mapped to the Knowledge Point B node in the knowledge graph and inherits its hierarchical attributes. Transform data from disordered scenes into ordered nodes in the knowledge network, so that multi-dimensional data has semantic associations. For example, the operation data of the circuit analysis module in the experimental class is associated with the knowledge points of Ohm's law in the theoretical class, so that multi-dimensional learning data can be effectively processed by deep neural networks, supporting the operation of the entire personalized layered teaching system.

[0090] The above-mentioned personalized stratified education method for higher education based on artificial intelligence obtains multi-dimensional learning data and performs spatiotemporal alignment to generate a synchronous multi-dimensional data set, effectively integrating multi-source heterogeneous data such as learning behavior and practice trajectories, and solving the problems of extensive data processing and ineffective integration of multi-dimensional data in traditional methods; uses deep neural networks to fuse spatiotemporal features and time series modeling to construct dynamic student portraits, accurately depicting the dynamic changes of students' knowledge mastery, learning behavior and cognitive status in real time, avoiding the defects of static student portraits and lagging teaching adjustments; uses semantic analysis models to analyze knowledge mastery characteristics and combines knowledge graphs to generate personalized resource recommendation sequences, deeply analyzes knowledge mastery details and knowledge graph logical relationships, and improves the scientificity and accuracy of resource recommendations; uses path reasoning algorithms based on recommendation sequences and dynamic portraits to plan personalized learning paths, fully considering Taking into account changes in learning load and the cognitive span between knowledge points, this approach optimizes learning efficiency and adapts to individual needs. It also binds personalized resource recommendation sequences to learning paths at the teaching level. Based on the multi-level node structure of the knowledge graph, it divides learning node levels and maps resources to generate tiered teaching plans, achieving intelligent and dynamic matching of teaching levels with learning resources and paths. This addresses the rigidity and reliance on manual pre-setting of traditional tiered mechanisms. By calculating deviations and adjusting parameters based on the temporal changes in knowledge mastery, a dynamic feedback mechanism is constructed to iteratively optimize teaching plans, ensuring continuous alignment between the learning process and teaching objectives. The multi-level node structure of the knowledge graph is used to transform learning paths into visual cognitive trajectory heat maps. Combined with dynamic dashboards generated by quantitative modeling of knowledge mastery, this approach uses an adaptive rendering engine to integrate spatiotemporal correlations to create a visual interface for teaching decision-making, providing intuitive data support for teaching adjustments. This approach provides precise teaching support tailored to individual student differences, effectively improving teaching effectiveness and learning experience, and achieving intelligent, dynamic, and efficient personalized tiered teaching in higher education.

[0091] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0092] Based on the same inventive concept, the embodiments of the present application also provide an AI-based personalized stratified teaching system for higher education, which is used to implement the aforementioned AI-based personalized stratified teaching method for higher education. The implementation solution provided by this system is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the AI-based personalized stratified teaching system for higher education provided below can be found in the above-mentioned limitations of the AI-based personalized stratified teaching method for higher education, and will not be repeated here.

[0093] In an exemplary embodiment, Figure 2 As shown, a personalized and hierarchical teaching system for higher education based on artificial intelligence is provided, including:

[0094] The data acquisition and processing module 11 is used to acquire multi-dimensional learning data and perform spatiotemporal alignment to generate a synchronized multi-dimensional dataset;

[0095] A feature fusion modeling module 12 is used to perform spatiotemporal feature fusion and time series modeling on the synchronized multidimensional dataset using a deep neural network to obtain a dynamic student portrait;

[0096] Semantic knowledge module 13, used to analyze the knowledge mastery characteristics of dynamic student portraits using a semantic analysis model, and generate a personalized resource recommendation sequence based on the knowledge graph;

[0097] The learning path planning module is used to generate personalized learning paths based on personalized resource recommendation sequences and dynamic student profiles using path reasoning algorithms;

[0098] The teaching plan generation module is used to bind the personalized resource recommendation sequence and personalized learning path to the teaching level to generate a hierarchical teaching plan.

[0099] In one embodiment, the semantic knowledge module 13 is further configured to:

[0100] Use semantic analysis models to analyze the knowledge mastery characteristics of dynamic student portraits and generate weak knowledge point information;

[0101] Match target learning resources associated with weak knowledge points from the pre-set learning resource library;

[0102] The target learning resources are sorted using a preset priority sorting algorithm to generate a personalized resource recommendation sequence.

[0103] In one embodiment, the learning path planning module 14 is further configured to construct the path reasoning algorithm using the following formula:

[0104]

[0105] Where μ represents the cognitive load weight coefficient, C(t) = w1·e -t / τ +w2·sigmoid(ΔK t ) represents the cognitive load value at time t, w1 and w2 represent the weights of time decay and knowledge point transition, τ represents the learning fatigue time constant, ΔK t represents the jump distance between adjacent knowledge points in the knowledge graph, C max represents the preset maximum tolerable cognitive load, v represents the knowledge span penalty coefficient, dist(k t , k t-1 ) represents the cognitive span between knowledge points, λ represents the knowledge coverage reward coefficient, S g (π) represents the coverage score of level g, and G represents the set of teaching objective levels in the knowledge graph.

[0106] In one embodiment, the teaching plan generating module 15 is further configured to:

[0107] Based on the multi-level node structure of the knowledge graph, the learning nodes in the personalized learning path are divided into levels, and the resources in the personalized resource recommendation sequence are mapped to the corresponding levels;

[0108] Generate hierarchical teaching plans based on the teaching objectives of each teaching level;

[0109] Integrate hierarchical teaching plans, stratified personalized learning paths, and mapped personalized resources to generate stratified teaching plans.

[0110] In one embodiment, the system further includes an optimization module for:

[0111] Based on the temporal changes in the characteristics of knowledge mastery, calculate the deviation from the teaching objectives;

[0112] When the deviation exceeds the preset threshold, the node topology relationship of the teaching level is recalculated in the knowledge graph, and the sorting priority of the personalized resource recommendation sequence is updated;

[0113] Adjust the cognitive load weight coefficient μ and the knowledge coverage reward coefficient λ in the path reasoning algorithm according to the deviation direction of the deviation degree;

[0114] Based on the updated sorting priorities and the adjusted path reasoning algorithm parameters, a new hierarchical teaching plan is iteratively generated.

[0115] In one embodiment, the system further includes a visualization module for:

[0116] Based on the multi-level node structure of the knowledge graph, the personalized learning path is converted into a visual cognitive trajectory heat map;

[0117] Quantitatively model knowledge mastery characteristics and generate dynamic dashboards with historical comparisons and forecast curves;

[0118] The adaptive rendering engine is used to integrate the trajectory heat map with the dynamic dashboard in time and space to generate a visual interface for teaching decision-making.

[0119] In one embodiment, the data acquisition and processing module 11 is further configured to:

[0120] Perform time series calibration on the learning data according to the timestamp to obtain the time series calibrated learning data;

[0121] The time-series-calibrated learning data is spatially aligned according to the spatial identifier to generate a synchronized multidimensional dataset.

[0122] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the aforementioned personalized stratified teaching method for higher education based on artificial intelligence are implemented.

[0123] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0124] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0125] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A personalized and layered teaching method for higher education based on artificial intelligence, characterized by: The method comprises: Acquire multi-dimensional learning data and perform spatiotemporal alignment to generate synchronized multi-dimensional datasets; Using a deep neural network to perform spatiotemporal feature fusion and time series modeling on the synchronized multidimensional dataset to obtain a dynamic student portrait; Utilize a semantic analysis model to analyze the knowledge mastery characteristics of the dynamic student portrait, and generate a personalized resource recommendation sequence based on the knowledge graph; Based on the personalized resource recommendation sequence and dynamic student portrait, a personalized learning path is generated using a path reasoning algorithm; The personalized resource recommendation sequence and the personalized learning path are bound to the teaching level to generate a layered teaching plan.

2. The method according to claim 1, characterized in that The method of using a semantic analysis model to analyze the knowledge mastery characteristics of the dynamic student portrait and generating a personalized resource recommendation sequence in combination with the knowledge graph includes: Utilizing a semantic analysis model to analyze the knowledge mastery characteristics of the dynamic student portrait and generate weak knowledge point information; Matching target learning resources associated with the weak knowledge point information from a preset learning resource library; The target learning resources are sorted using a preset priority sorting algorithm to generate the personalized resource recommendation sequence.

3. The method according to claim 1, characterized in that The path inference algorithm is constructed using the following formula: Where μ represents the cognitive load weight coefficient, C(t) = w1·e -t / τ +w2·sigmoid(ΔK t ) represents the cognitive load value at time t, w1 and w2 represent the weights of time decay and knowledge point transition, τ represents the learning fatigue time constant, ΔK t represents the jump distance between adjacent knowledge points in the knowledge graph, C max represents the preset maximum tolerable cognitive load, v represents the knowledge span penalty coefficient, dist(k t , k t-1 ) represents the cognitive span between knowledge points, λ represents the knowledge coverage reward coefficient, S g (π) represents the coverage score of level g, and G represents the set of teaching objective levels in the knowledge graph.

4. The method according to claim 1, wherein Binding the personalized resource recommendation sequence and the personalized learning path to the teaching level to generate a tiered teaching plan includes: Based on the multi-level node structure of the knowledge graph, the learning nodes in the personalized learning path are divided into levels, and the resources in the personalized resource recommendation sequence are mapped to the corresponding levels; Generate a level teaching plan according to the teaching objectives of each teaching level; The hierarchical teaching plan, the stratified personalized learning path and the mapped personalized resources are integrated to generate a stratified teaching plan.

5. The method according to claim 3, characterized in that The method further comprises: Calculating the degree of deviation from the teaching objective based on the temporal changes of the knowledge mastery characteristics; When the deviation exceeds a preset threshold, the node topology relationship of the teaching level is recalculated in the knowledge graph, and the sorting priority of the personalized resource recommendation sequence is updated; Adjust the cognitive load weight coefficient μ and the knowledge coverage reward coefficient λ in the path reasoning algorithm according to the deviation direction of the deviation degree; Based on the updated sorting priorities and the adjusted path reasoning algorithm parameters, a new hierarchical teaching plan is iteratively generated.

6. The method according to claim 1, characterized in that The method further comprises: Based on the multi-level node structure of the knowledge graph, the personalized learning path is converted into a visual cognitive trajectory heat map; Quantitatively model the knowledge mastery characteristics and generate a dynamic dashboard containing historical comparisons and forecast curves; The trajectory heat map is temporally and spatially integrated with the dynamic dashboard using an adaptive rendering engine to generate a visual interface for teaching decision-making.

7. The method according to claim 1, characterized in that The method of acquiring multi-dimensional learning data and performing spatiotemporal alignment to generate a synchronized multi-dimensional dataset includes: Extracting the timestamp and spatial identifier of the learning data using data parsing technology; Performing time sequence calibration on the learning data according to the timestamp to obtain time sequence calibrated learning data; The time-series-calibrated learning data is spatially calibrated according to the spatial identifier to generate the synchronized multidimensional dataset.

8. An artificial intelligence-based personalized stratified teaching system for higher education, characterized by: The system comprises: Data acquisition and processing module, used to acquire multi-dimensional learning data and perform spatiotemporal alignment to generate synchronized multi-dimensional data sets; A feature fusion modeling module is used to use a deep neural network to perform spatiotemporal feature fusion and time series modeling on the synchronized multidimensional dataset to obtain a dynamic student portrait; A semantic knowledge module is used to analyze the knowledge mastery characteristics of the dynamic student profile using a semantic analysis model, and generate a personalized resource recommendation sequence in combination with the knowledge graph; A learning path planning module is used to generate a personalized learning path using a path reasoning algorithm based on the personalized resource recommendation sequence and dynamic student profile; The teaching plan generation module is used to bind the personalized resource recommendation sequence and the personalized learning path to the teaching level to generate a hierarchical teaching plan.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images