International course intelligent adaptation and quality evaluation system driven by neural network

Abstract

The invention provides an internationalized course intelligent adaptation and quality evaluation system driven by a neural network, which belongs to the technical field of education and comprises a topological feature extraction module, a space-time learning trajectory modeling module, a topological invariant prediction module, a course adaptation recommendation module, a quality evaluation module and an adaptive control module. According to the method, essential structure features of learning behaviors are extracted through a topological data analysis technology, a learning process is modeled as a trajectory on a knowledge manifold, a learning effect prediction model is constructed based on topological invariants, and personalized curriculum recommendation and multi-dimensional quality evaluation are realized. A differential geometry method is adopted to quantify learning path efficiency, a prediction framework driven by topology invariants is constructed to improve model stability, through the system, learning modes of learners with different cultural backgrounds can be deeply understood, the learning effect can be accurately predicted, learning resource configuration can be optimized, and data-driven decision support can be provided for internationalized education.

Description

Technical Field

[0001] This invention relates to the field of educational technology, and in particular to a neural network-driven intelligent adaptation and quality assessment system for internationalized curricula. Background Technology

[0002] With the deepening of globalization, international education has become an important trend in modern education development. Significant differences exist in the education systems, curriculum standards, and teaching methods of different countries and regions, making the sharing and optimal allocation of high-quality educational resources globally a pressing issue.

[0003] Traditional methods of curriculum adaptation and quality assessment rely primarily on manual analysis and experience, which suffers from high subjectivity, low efficiency, and difficulty in large-scale application. With the development of big data and artificial intelligence technologies, data-driven educational decision support systems are emerging. Existing technologies mainly evaluate learning effectiveness by analyzing learning behavior data and building simple predictive models, but these methods often only focus on surface features and fail to reveal the deep structure and essential laws of the learning process.

[0004] Existing technologies have the following problems: First, they are difficult to accurately capture the intrinsic structural characteristics of learning behavior; second, they cannot effectively model the geometric characteristics of different learning paths; third, the predictive models are not adaptable enough to changes in learning behavior; and fourth, they lack a scientific and reasonable framework for evaluating course quality. Summary of the Invention

[0005] The purpose of this invention is to provide a neural network-driven intelligent adaptation and quality assessment system for internationalized courses, aiming to solve the problems existing in traditional methods in course adaptation and quality assessment, and to achieve a deep understanding of learning behavior, accurate prediction of learning effects, and personalized course recommendations.

[0006] This invention proposes a neural network-driven intelligent adaptation and quality assessment system for internationalized courses, comprising:

[0007] The topology feature extraction module is used to collect multidimensional data of learning behavior, construct a learning behavior feature space, extract the topological structure features of learning behavior through topological data analysis technology, and generate topological feature vectors.

[0008] The spatiotemporal learning trajectory modeling module is connected to the topological feature extraction module. It is used to receive the topological feature vector, construct a knowledge state manifold, represent the learning process as a trajectory on the knowledge manifold, calculate the deviation between the actual learning trajectory and the theoretical optimal path, and extract the geometric characteristics of the learning trajectory.

[0009] The topological invariant prediction module is connected to the spatiotemporal learning trajectory modeling module. It is used to receive the geometric characteristics of the learning trajectory, construct a prediction model based on topological invariants, and generate learning effect prediction results.

[0010] The course adaptation and recommendation module is connected to the topological invariant prediction module. It is used to receive the learning effect prediction results and generate personalized course recommendations based on learner characteristics and prediction results.

[0011] The quality assessment module, connected to the topological invariant prediction module, is used to receive the learning effect prediction results, assess the correlation between course quality and learning effect, and generate a quality assessment report.

[0012] An adaptive control module is connected to the topology feature extraction module, the spatiotemporal learning trajectory modeling module, and the topology invariant prediction module, respectively, and is used to dynamically adjust model parameters and optimize prediction performance based on the deviation between the prediction results and the actual results.

[0013] Preferably, the topological feature extraction module includes:

[0014] The data acquisition unit is used to collect learning behavior data, including timestamps, learning activity types, activity durations, interaction intensity, learning content locations, and feedback results, to construct a six-dimensional learning behavior feature space.

[0015] A point cloud processing unit, connected to the data acquisition unit, is used to standardize and reduce the dimensionality of the original learning behavior data to generate learning behavior point cloud data.

[0016] The complex construction unit, connected to the point cloud processing unit, is used to construct a Vietoris-Rips complex based on a distance matrix, set multi-level distance thresholds, and construct a simple complex sequence.

[0017] The persistent homology computation unit, connected to the complex construction unit, is used to compute the persistent graph of the Betti number of each dimension as a function of a threshold, extract the statistical features of the persistent graph, and generate a topological summary vector.

[0018] The feature mapping unit, connected to the persistent cohomology computing unit, is used to fuse topological features with traditional learned features, establish a feature importance scoring mechanism, and generate the final topological feature vector.

[0019] Preferably, the spatiotemporal learning trajectory modeling module includes:

[0020] The knowledge state assessment unit is used to assess learners’ mastery of knowledge points based on test, assignment and interaction data, and to construct a time series knowledge state vector, which is represented as a point in the knowledge state space.

[0021] Knowledge manifold construction unit, used to define Riemannian metrics based on knowledge point association networks, construct the geometric structure of the knowledge space, and establish the local coordinate system of the knowledge manifold;

[0022] The geodesic calculation unit, connected to the knowledge manifold construction unit, is used to define the initial and target knowledge states, apply the geodesic algorithm to calculate the optimal learning path, and generate a reference geodesic trajectory.

[0023] The trajectory feature extraction unit is connected to the knowledge state evaluation unit and the geodesic calculation unit, respectively, and is used to calculate the geometric characteristics of the actual learning trajectory, measure the deviation between the actual trajectory and the geodesic, and extract the trajectory feature vector.

[0024] Preferably, the topological invariant prediction module includes:

[0025] The feature fusion unit receives topological feature vectors and trajectory feature vectors, applies nonlinear transformations to map the features to the latent space, and generates a comprehensive feature representation.

[0026] A topological invariant processing unit, connected to the feature fusion unit, is used to extract topological invariants from features, construct an invariant weight allocation mechanism, and generate a weighted feature representation.

[0027] The temporal dependency modeling unit, connected to the topological invariant processing unit, is used to process the temporal dependency of learning behavior sequences, capture long and short-term learning patterns, and generate context-enhanced feature representations.

[0028] A multi-level prediction generation unit, connected to the temporal dependency modeling unit, is used to perform multi-level predictions from the behavior level to the knowledge level and then to the ability level, and generate learning effect prediction results.

[0029] The prediction quality assessment unit, connected to the multi-level prediction generation unit, is used to calculate prediction errors and confidence intervals, evaluate topological stability indicators, and generate prediction quality reports.

[0030] Preferably, the course adaptation recommendation module includes:

[0031] The learner profiling unit is used to build a multi-dimensional feature model of learners based on learning behavior data and knowledge status data, and to identify learning styles, knowledge base and learning goals.

[0032] The course feature analysis unit is used to extract course content features, structural features, and difficulty features, and to construct a course feature vector.

[0033] The matching degree calculation unit, connected to the learner profiling unit and the course feature analysis unit, is used to calculate the degree of matching between learner features and course features and generate a matching degree score.

[0034] The recommended path generation unit, connected to the matching degree calculation unit and the topology invariant prediction module, is used to generate personalized learning paths and course recommendations based on the matching degree score and learning effect prediction results.

[0035] Preferably, the quality assessment module includes:

[0036] The multi-dimensional indicator collection unit is used to collect multi-dimensional evaluation indicators such as course content quality, instructional design quality, interactive experience quality, and learning support quality.

[0037] The correlation analysis unit, connected to the multi-dimensional indicator acquisition unit and the topological invariant prediction module, is used to analyze the correlation between quality indicators and learning effect prediction results, and to identify key influencing factors.

[0038] The quality scoring unit, connected to the correlation analysis unit, is used to quantitatively score the quality of various aspects of the course based on the correlation analysis results and generate a comprehensive quality score.

[0039] An improvement suggestion unit, connected to the quality scoring unit, is used to generate targeted improvement suggestions based on the quality score and correlation analysis results, forming a quality assessment report.

[0040] Preferably, the adaptive control module includes:

[0041] The performance comparison unit is used to compare the predicted learning performance with the actual learning performance, calculate the prediction error, and identify model deficiencies.

[0042] The parameter adjustment unit, connected to the effect comparison unit, is used to dynamically adjust the parameters of the topology feature extraction module, the spatiotemporal learning trajectory modeling module, and the topology invariant prediction module based on the prediction error, thereby optimizing the model performance.

[0043] The strategy optimization unit, connected to the parameter adjustment unit, is used to optimize the feature selection strategy, trajectory modeling strategy, and prediction algorithm strategy to improve the overall system performance.

[0044] Preferably, the persistent coherence calculation unit includes:

[0045] The boundary operator construction module is used to construct the boundary operator matrix of the simple complex, which records the inclusion relationship between simplexes of different dimensions;

[0046] The matrix reduction module, connected to the boundary operator construction module, is used to perform row reduction operations on the boundary operator matrix and identify the generators of the homology group.

[0047] A persistent barcode generation module, connected to the matrix reduction module, is used to record the birth and death scales of topological features and generate persistent barcode representations.

[0048] The statistical feature extraction module, connected to the persistent barcode generation module, is used to extract statistical features from the persistent barcode, including lifecycle distribution, persistence index, and topological complexity measure.

[0049] Preferably, the trajectory feature extraction unit includes:

[0050] The curvature calculation module is used to calculate the curvature of the learning trajectory on the knowledge manifold, quantifying the degree of bending of the learning path;

[0051] The velocity field analysis module is used to calculate the speed at which the learning trajectory progresses across different knowledge points and to identify points where the learning speed changes.

[0052] The geodesic deviation quantification module is used to calculate the geodesic distance between the actual learning trajectory and the theoretical optimal path, thereby quantifying the learning efficiency.

[0053] The feature fusion module is connected to the curvature calculation module, the velocity field analysis module, and the geodesy deviation quantification module, respectively, and is used to integrate various geometric features to generate a comprehensive trajectory feature vector.

[0054] Preferably, the topological invariant processing unit includes:

[0055] The feature space transformation module is used to map topological features and trajectory features to a latent space with topology preservation properties, thus preserving the topological structure of the data.

[0056] An invariant identification module, connected to the feature space transformation module, is used to identify topological invariants in the feature representation, including connectivity, ring structures, and high-dimensional holes.

[0057] The stability assessment module, connected to the invariant identification module, is used to assess the stability of topological features to disturbances and calculate the topological stability index.

[0058] The weight optimization module is connected to the invariant identification module and the stability evaluation module, respectively, and is used to assign dynamic weights to topological features based on the importance and stability index of invariants to generate weighted feature representations.

[0059] This invention, by introducing cutting-edge mathematical methods such as topological data analysis and differential geometry, constructs a complete technical system from topological feature extraction and learning trajectory modeling to adaptive prediction, which has the following beneficial effects:

[0060] 1. Through topological data analysis technology, we can capture the essential structural characteristics of learning behavior and discover learning patterns and knowledge structures that traditional methods cannot identify;

[0061] 2. A learning trajectory modeling method based on differential geometry represents the learning process as a trajectory on a knowledge manifold, enabling a scientific analysis of the optimality and efficiency of the learning path;

[0062] 3. The topological invariant-driven prediction framework improves the model's adaptability to different learning behaviors and its prediction stability;

[0063] 4. Personalized course recommendations based on learning outcome prediction significantly improve the accuracy of learning resource allocation and learning effectiveness;

[0064] 5. A multi-dimensional quality assessment system provides data-driven decision support for curriculum improvement and teaching innovation. Attached Figure Description

[0065] Figure 1 This is a diagram of the overall system architecture of the present invention;

[0066] Figure 2 This is a schematic diagram of the topology feature extraction module;

[0067] Figure 3 A schematic diagram of the spatiotemporal learning trajectory modeling module;

[0068] Figure 4 This is a schematic diagram of the topological invariant prediction module;

[0069] Figure 5 A structural diagram of the recommended module adapted for the course;

[0070] Figure 6 This is a structural diagram of the quality assessment module;

[0071] Figure 7 This is a schematic diagram of the adaptive control module.

[0072] Figure 8 A schematic diagram of the structure of a persistent cohomology computing unit;

[0073] Figure 9 This is a schematic diagram of the trajectory feature extraction unit.

[0074] Figure 10 This is a schematic diagram of the topological invariant processing unit. Detailed Implementation

[0075] Please refer to Figures 1-10 The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0076] like Figure 1As shown, the neural network-driven intelligent adaptation and quality assessment system for internationalized courses provided by this invention includes a topological feature extraction module 1, a spatiotemporal learning trajectory modeling module 2, a topological invariant prediction module 3, a course adaptation recommendation module 4, a quality assessment module 5, and an adaptive control module 6.

[0077] The topology feature extraction module 1 is used to collect multidimensional data on learning behaviors, construct a learning behavior feature space, and extract the topological structural features of learning behaviors through topological data analysis techniques to generate topological feature vectors. In international curriculum teaching, this module can capture the differences in learning patterns among learners from different cultural backgrounds and discover potential patterns in learning behaviors.

[0078] The spatiotemporal learning trajectory modeling module 2 is connected to the topological feature extraction module 1. It receives topological feature vectors, constructs a knowledge state manifold, represents the learning process as a trajectory on the knowledge manifold, calculates the deviation between the actual learning trajectory and the theoretical optimal path, and extracts the geometric characteristics of the learning trajectory. This module can analyze the efficiency and characteristics of learners' knowledge acquisition paths under different international curriculum standards.

[0079] The topological invariant prediction module 3 is connected to the spatiotemporal learning trajectory modeling module 2. It receives the geometric characteristics of the learning trajectory, constructs a prediction model based on topological invariants, and generates learning outcome predictions. This module can predict learners' knowledge acquisition and learning outcomes for curriculum content within different national education systems.

[0080] The course adaptation and recommendation module 4 connects to the topological invariant prediction module 3 to receive learning outcome prediction results and generate personalized course recommendations based on learner characteristics and prediction results. For international courses, this module can recommend the most suitable course content and learning path based on the learner's cultural background and learning style.

[0081] The quality assessment module 5 is connected to the topological invariant prediction module 3. It receives learning outcome prediction results, evaluates the correlation between course quality and learning outcomes, and generates a quality assessment report. This module can analyze the applicability and effectiveness of internationalized courses in different cultural contexts.

[0082] The adaptive control module 6 is connected to the topology feature extraction module 1, the spatiotemporal learning trajectory modeling module 2, and the topology invariant prediction module 3, respectively. It dynamically adjusts model parameters and optimizes prediction performance based on the deviation between the prediction results and the actual results. This module ensures that the system can continuously adapt to the changes and diversity of the international education environment.

[0083] like Figure 2 As shown, the topology feature extraction module 1 includes a data acquisition unit 11, a point cloud processing unit 12, a complex construction unit 13, a persistent homology calculation unit 14, and a feature mapping unit 15.

[0084] The data acquisition unit 11 is used to collect learning behavior data, including timestamps, learning activity types, activity durations, interaction intensity, learning content locations, and feedback results, constructing a six-dimensional learning behavior feature space. In one embodiment of the present invention, the learning behavior feature space can be represented as:

[0085] ,

[0086] in, It represents a timestamp, accurate to the second, recording the specific time point when the learner engages in a learning activity; The learning activity type is represented by 15 standardized activity types, such as video viewing, reading, practice, and discussion, coded by integers 1-15. This indicates the duration of the activity, in seconds, and records the duration of each learning activity. This represents the intensity of the interaction, with a discrete value ranging from 0 to 10. The larger the value, the deeper the interaction. This indicates the location of the learning content, i.e., the node ID in the course structure, used to locate the position of the learning content within the course; This represents the feedback result, with a discrete value ranging from -5 to +5. Negative values ​​indicate negative feedback, and positive values ​​indicate positive feedback.

[0087] For example, in an international economics course, the system might record learning behavior data points like: {2023-10-05 14:30:45, 3 (video viewing), 350, 7, module2-lecture3, 4}, indicating that the learner watched the video content of module 2-lecture3 at that time, lasting for 350 seconds, with an interaction intensity of 7 (possibly pausing, rewinding, etc.) and a feedback result of 4 (indicating a relatively positive learning experience).

[0088] Data acquisition employs a dual-mode approach, combining activity-triggered and fixed-interval (5-minute) methods to ensure timeliness and continuity. Data storage utilizes a hybrid architecture of time-series and relational databases, establishing relationships through learner IDs and session IDs.

[0089] Point cloud processing unit 12 is connected to data acquisition unit 11 and is used to standardize and reduce the dimensionality of the original learning behavior data to generate learning behavior point cloud data. Standardization involves converting data of different dimensions to a uniform scale to eliminate the influence of unit dimensions. For timestamp data, a relative time calculation method is used; for discrete data, such as learning activity types, one-hot encoding is used to convert them into numerical representations. Dimensionality reduction uses principal component analysis (PCA) to retain 95% of the information, typically reducing the original six-dimensional data to three to five dimensions. In practical applications, for learners from different countries, it may be necessary to adjust the standardization parameters to accommodate cultural differences to ensure fairness in data processing.

[0090] The complex construction unit 13 is connected to the point cloud processing unit 12 and is used to construct a Vietoris-Rips complex based on a distance matrix, set multi-level distance thresholds, and construct a simple complex sequence. In this invention, the Vietoris-Rips complex is defined as:

[0091] ,

[0092] in, The distance threshold parameter controls the scale of complex construction; Representing a simplex The maximum distance between any two points in the set is used to determine whether a set of points forms a simplex; Representing a simplex It is a learning behavior feature space A subset of.

[0093] Distance threshold parameter The threshold settings employ a multi-level strategy, typically selecting 10–20 uniformly distributed thresholds within the range of [0.01, 0.5]. This range is chosen based on statistical experimental data; 0.01 captures the smallest local structures, while 0.5 is large enough to form globally connected complexes. In international courses, learners from different cultural backgrounds may exhibit different learning behavior patterns, and multi-level thresholds can capture these pattern characteristics at different scales.

[0094] The persistent cohomology computation unit 14 is connected to the complex construction unit 13 and is used to compute the persistent graph with varying Betti numbers across dimensions as a function of a threshold, extract statistical features of the persistent graph, and generate a topological summary vector. The core of persistent cohomology computation is tracking the appearance and disappearance of topological features (such as connected components, cycles, and holes) at different scales. In practice, for the analysis of internationalized course learning behavior, the zero-dimensional Betti number (… The one-dimensional Betty number reflects the clustering pattern of learning behavior. This may reflect a cyclical learning path, and the two-dimensional Betty number ( () may represent a complex structure of knowledge understanding.

[0095] The feature mapping unit 15 is connected to the persistent cohomology computation unit 14, and is used to fuse topological features with traditional learned features to establish a feature importance scoring mechanism and generate the final topological feature vector. The feature importance scoring uses an information gain-based method to calculate the contribution of each feature to the predicted learning performance. The final generated topological feature vector can be represented as:

[0096] ,

[0097] in, These represent the number of connected components, cycles, and holes, respectively, describing the topological features of the learning behavior; It represents the statistical characteristics of the persistent graph in each dimension, such as the distribution of lifetime length and persistence indicators, reflecting the stability and importance of topological features.

[0098] In an internationalized curriculum environment, these topological characteristics can help identify the unique learning patterns of learners from different cultural backgrounds. For example, learners from certain cultural backgrounds may tend to follow a linear learning path (lower threshold for learning). Learners from other cultural backgrounds may prefer exploratory learning (higher value), while learners from other cultural backgrounds may prefer exploratory learning (higher value). value).

[0099] like Figure 3 As shown, the spatiotemporal learning trajectory modeling module 2 includes a knowledge state evaluation unit 21, a knowledge manifold construction unit 22, a geodesic calculation unit 23, and a trajectory feature extraction unit 24.

[0100] The knowledge state assessment unit 21 is used to assess learners' mastery of knowledge points based on test, assignment, and interaction data, constructing a time-series knowledge state vector, represented as a point in the knowledge state space. The knowledge state vector can be represented as:

[0101] ,

[0102] in, Indicates the first The mastery level of each knowledge point is a continuous value in the range of [0,1], where 0 represents no mastery and 1 represents complete mastery. The total number of knowledge points in a domain depends on the complexity and level of detail of the course.

[0103] The assessment of knowledge mastery comprehensively considers multiple factors, including test scores, practice accuracy, time taken to answer questions, and review frequency. For different types of assessment data, a weighted fusion approach is used, with weights dynamically adjusted based on data reliability and importance. In international curricula, knowledge assessment also needs to consider the differences in evaluation standards across different education systems, ensuring fairness and comparability through standardized processing.

[0104] For example, in an international programming course, the system might assess a learner's understanding of loop structures as 0.75, indicating that the learner has a good grasp and application of the concept, but there is still room for improvement.

[0105] The knowledge manifold construction unit 22 is used to define Riemannian metrics based on the knowledge point association network, construct the geometric structure of the knowledge space, and establish the local coordinate system of the knowledge manifold. The knowledge point association network is established through the association matrix. It means that among them Representing knowledge points and The correlation strength is defined, with values ​​ranging from [0,1], where 0 indicates no correlation and 1 indicates a strong correlation. A Riemannian metric tensor is constructed based on the correlation matrix. , describes the local geometric properties of the knowledge space.

[0106] In practical applications, association matrices can be constructed by combining expert knowledge, curriculum structure analysis, and learning data mining. For internationalized curricula, it is necessary to consider the different understandings and organizational methods of knowledge structures in different national education systems, and to integrate multiple perspectives to build a more comprehensive knowledge association network.

[0107] The geodesic calculation unit 23 is connected to the knowledge manifold construction unit 22. It is used to define the initial and target knowledge states, apply the geodesic algorithm to calculate the optimal learning path, and generate a reference geodesic trajectory. On the knowledge manifold, a geodesic represents the shortest path between two points, reflecting the optimal learning path from the initial knowledge state to the target knowledge state. The geodesic satisfies the following equation:

[0108] ,

[0109] in, It represents the connection on a Riemannian manifold and is used to define the derivative of a vector field; This represents a parameterized curve, signifying the learning trajectory. The tangent vector of the curve represents the learning direction and speed.

[0110] In practical implementation, a discrete approximation algorithm is used to solve the geodesic equations. First, the continuous problem is discretized into a shortest path problem on a graph, and then Dijkstra's algorithm or A* algorithm is applied to solve it. To improve computational efficiency, a hierarchical strategy can be adopted: first, the approximate path is calculated on a coarse-grained knowledge graph, and then optimization is performed at a fine-grained level.

[0111] For internationalized curricula, the optimal learning path may vary depending on cultural background and educational tradition. For example, some educational systems may emphasize mastering theoretical foundations before applying them in practice, while others may prefer a learning approach that balances theory and practice. By analyzing the success paths of learners from different cultural backgrounds, the system can provide more suitable learning path suggestions for different groups.

[0112] The trajectory feature extraction unit 24 is connected to the knowledge state evaluation unit 21 and the geodesic calculation unit 23, respectively. It is used to calculate the geometric characteristics of the actual learning trajectory, measure the deviation between the actual trajectory and the geodesic, and extract the trajectory feature vector. The trajectory feature vector can be represented as:

[0113] ,

[0114] in, The trajectory curvature represents the degree of curvature of the learning path; a larger value indicates a more drastic change in the path. For torsion, it reflects the degree of distortion of the learning path and represents the frequency of changes in the learning direction; These represent velocity, acceleration, and , respectively, indicating changes in learning progress. Indicates the speed of knowledge acquisition. Represents the rate of change of velocity. Indicates the rate of change of acceleration; The average and maximum geodesic deviations represent the learning efficiency; the smaller the value, the closer it is to the optimal path.

[0115] In internationalized curricula, these trajectory characteristics can reveal differences in learning strategies among learners from different cultural backgrounds. For example, some learners may exhibit a high curvature (frequently jumping between different knowledge points), while others may exhibit a low curvature but high speed (a linear but efficient learning style).

[0116] like Figure 4 As shown, the topological invariant prediction module 3 includes a feature fusion unit 31, a topological invariant processing unit 32, a temporal dependency modeling unit 33, a multi-level prediction generation unit 34, and a prediction quality evaluation unit 35.

[0117] Feature fusion unit 31 receives topological feature vectors and trajectory feature vectors, applies a nonlinear transformation to map the features to the latent space, and generates a comprehensive feature representation. Feature fusion employs a nonlinear mapping function:

[0118] ,

[0119] in, It is a topological feature vector, containing the topological structural features of the learned behavior; The trajectory feature vector contains the geometric properties of the learned path; The latent space is a high-dimensional feature space used to represent the fused features; the mapping function... This is typically achieved using a multilayer perceptron, which unifies features from different sources into a common feature space.

[0120] In an internationalized curriculum environment, feature integration needs to consider the relative importance of features across different cultural backgrounds. For example, for educational backgrounds that emphasize independent learning, trajectory features may be more predictive; while for educational backgrounds that emphasize collaborative learning, connectivity patterns in topological features may be more important. By analyzing learning data, the system can dynamically adjust feature weights for different groups.

[0121] The topological invariant processing unit 32 is connected to the feature fusion unit 31 and is used to extract topological invariants from the features, construct an invariant weight allocation mechanism, and generate a weighted feature representation. Topological invariants refer to properties that remain unchanged under topological transformations (such as continuous deformation), and can provide a robust representation of the essential features of the learning behavior.

[0122] In internationalized curricula, topological invariants are particularly valuable because they can transcend cultural and linguistic differences to capture common characteristics of learning behaviors. For example, cyclical structures in the learning process (representing cyclical review patterns) may be an effective cross-cultural learning strategy, and the system can identify such patterns and emphasize them in recommendations.

[0123] The temporal dependency modeling unit 33 is connected to the topological invariant processing unit 32 to process the temporal dependencies of the learning behavior sequence, capture long and short-term learning patterns, and generate context-enhanced feature representations. The temporal dependency modeling adopts a gated recurrent unit (GRU) or long short-term memory network (LSTM) structure, which can effectively process time series data.

[0124] For internationalized courses, temporal pattern analysis can reveal differences in time management and learning planning among learners from different cultural backgrounds. For example, some learners may prefer intensive learning (short periods of high intensity), while others may prefer distributed learning (long periods of low intensity). By identifying these patterns, the system can provide learners with learning plan suggestions that are more in line with their habits.

[0125] The multi-level prediction generation unit 34 is connected to the temporal dependency modeling unit 33, and is used to perform multi-level prediction from the behavioral level to the knowledge level and then to the ability level to generate learning effect prediction results. The multi-level prediction adopts a hierarchical structure, from low-level behavioral prediction (such as completion time and participation) to mid-level knowledge prediction (such as knowledge point mastery and concept understanding depth) and then to high-level ability prediction (such as knowledge application ability and transfer ability).

[0126] In an internationalized curriculum environment, multi-level forecasting needs to adapt to different educational goals and assessment standards. For example, some education systems may place greater emphasis on assessing knowledge mastery, while others may emphasize the cultivation of application and innovation skills. The system adjusts the focus and weight of forecasts by analyzing curriculum standards and learning objectives.

[0127] The prediction quality assessment unit 35 is connected to the multi-level prediction generation unit 34, and is used to calculate prediction error and confidence interval, evaluate topological stability indicators, and generate a prediction quality report. The prediction quality assessment uses multiple indicators, including accuracy, precision, recall, and F1 score. Simultaneously, a topological stability indicator is introduced. The robustness of the quantitative prediction model to topological perturbations.

[0128] For internationalized courses, predictive quality assessment also needs to consider cross-cultural adaptability, evaluating the consistency of the model's performance across learner data from different cultural backgrounds. This can help identify potential biases in the model and ensure the fairness and universality of the predictions.

[0129] like Figure 5 As shown, the course adaptation and recommendation module 4 includes a learner profile unit 41, a course feature analysis unit 42, a matching degree calculation unit 43, and a recommendation path generation unit 44.

[0130] The learner profiling unit 41 is used to construct a multi-dimensional feature model of learners based on learning behavior data and knowledge status data, and to identify learning styles, knowledge base, and learning goals. The learner profile includes multiple dimensions, such as cognitive characteristics (e.g., knowledge structure, learning ability), behavioral characteristics (e.g., learning habits, interaction patterns), and emotional characteristics (e.g., learning motivation, preferences).

[0131] In an internationalized curriculum environment, learner profiling also needs to consider the impact of cultural background and educational experience on learning styles. For example, learners from systems that emphasize critical thinking may exhibit more questioning and exploratory behaviors, while learners from systems that emphasize memorization and repetition may exhibit more systematic learning paths. By analyzing these characteristics, the system can better understand learners' needs and preferences.

[0132] The course feature analysis unit 42 is used to extract course content features, structural features, and difficulty features to construct a course feature vector. The course feature analysis employs multimodal content analysis technology to process course resources in various formats, including text, images, and videos. For text content, natural language processing techniques are used to extract features such as theme, difficulty, and structure; for multimedia content, computer vision and audio analysis techniques are used to extract content features.

[0133] For internationalized courses, content analysis also needs to consider language and cultural factors, such as language complexity, cultural adaptability, and the degree of localization. For example, the system can identify culturally specific references or assumptions in course materials and assess their comprehensibility for learners from different backgrounds.

[0134] The matching degree calculation unit 43 is connected to the learner profiling unit 41 and the course feature analysis unit 42. It is used to calculate the degree of matching between learner features and course features and generate a matching degree score. The matching degree calculation adopts the weighted cosine similarity method, taking into account the importance of different feature dimensions. The matching degree score ranges from [0,1], with a larger value indicating a higher degree of matching.

[0135] In recommending international courses, matching accuracy needs to balance universality and personalization. On the one hand, the system should recommend courses that adapt to learners' cultural backgrounds and learning styles; on the other hand, it also needs to consider the general requirements and standards of knowledge acquisition. The matching algorithm seeks to balance these two aspects through dynamic weight adjustments.

[0136] The recommended path generation unit 44 is connected to the matching degree calculation unit 43 and the topology invariant prediction module 3. It is used to generate personalized learning paths and course recommendations based on the matching degree score and learning effect prediction results. The recommended path generation employs a multi-objective optimization method, simultaneously considering factors such as matching degree, learning effect prediction, and learning time constraints. The recommended results include course combinations, learning order, expected completion time, and predicted learning effects.

[0137] For international courses, learning path recommendations also need to consider the differences in curriculum structure and certification requirements of different education systems. For example, if a learner's goal is to obtain a degree or certification from a specific country, the system will prioritize recommending course combinations and learning paths that meet that certification requirement.

[0138] like Figure 6 As shown, the quality assessment module 5 includes a multi-dimensional indicator collection unit 51, a correlation analysis unit 52, a quality scoring unit 53, and an improvement suggestion unit 54.

[0139] The multi-dimensional indicator collection unit 51 is used to collect multi-dimensional evaluation indicators such as course content quality, instructional design quality, interactive experience quality, and learning support quality. Data sources for these quality indicators include expert evaluations, learner feedback, and platform monitoring data.

[0140] In an internationalized curriculum environment, quality assessment also needs to consider cross-cultural adaptability indicators, such as the cultural neutrality of the content, the quality of multilingual support, and the proportion of international case studies. These indicators help assess the applicability and inclusivity of the curriculum globally.

[0141] The correlation analysis unit 52 is connected to the multi-dimensional indicator acquisition unit 51 and the topological invariant prediction module 3 to analyze the correlation between quality indicators and learning effect prediction results, and to identify key influencing factors. The correlation analysis employs various statistical methods and machine learning techniques, such as correlation analysis, regression analysis, and feature importance assessment.

[0142] For internationalized curricula, correlation analysis can reveal the relative importance of different quality factors across different cultural contexts. For example, in some cultural contexts, the systematic and comprehensive nature of the curriculum may be a key factor, while in others, interactivity and engagement may be more important. By identifying these differences, the system can provide more targeted recommendations for curriculum improvement.

[0143] The quality scoring unit 53 is connected to the correlation analysis unit 52, and is used to quantitatively score the quality of various aspects of the course based on the correlation analysis results, generating a comprehensive quality score. The quality scoring adopts the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation method to generate multi-dimensional quality scores and an overall score.

[0144] In the assessment of internationalized curricula, quality scoring needs to consider different evaluation standards and cultural expectations. The system ensures the rationality and comparability of scoring results across different cultural contexts through adaptive weight adjustments. For example, for education systems that emphasize theory, higher weight may be given to the depth and systematic nature of the content; while for education systems that emphasize practice, more emphasis may be placed on applicability and practicality.

[0145] The improvement suggestion unit 54 is connected to the quality scoring unit 53, and is used to generate targeted improvement suggestions based on the quality score and correlation analysis results, forming a quality assessment report. The improvement suggestions adopt a rule-based and case-based approach, combining the quality score and key influencing factors to provide specific and feasible improvement measures.

[0146] For internationalized curricula, improvement suggestions need to consider a balance between globalization and localization. On the one hand, maintain the universality and standardization of the curriculum; on the other hand, provide sufficient flexibility to adapt to different cultures and educational environments. The system can generate regional adaptation suggestions based on feedback and performance from learners in different regions, supporting the global promotion and localization adjustments of the curriculum.

[0147] like Figure 7 As shown, the adaptive control module 6 includes an effect comparison unit 61, a parameter adjustment unit 62, and a strategy optimization unit 63.

[0148] The performance comparison unit 61 is used to compare the predicted learning performance with the actual learning performance, calculate the prediction error, and identify model deficiencies. The performance comparison uses multiple evaluation metrics, such as mean squared error (MSE), mean absolute error (MAE), and R-squared value, to comprehensively evaluate the prediction performance.

[0149] In internationalized curricula, effectiveness comparisons also require group-specific analysis to evaluate the model's performance on data from learners with different cultural backgrounds. This helps identify potential cultural biases or adaptation issues in the model, providing a basis for subsequent parameter adjustments.

[0150] The parameter adjustment unit 62 is connected to the performance comparison unit 61 and is used to dynamically adjust the parameters of the topology feature extraction module 1, the spatiotemporal learning trajectory modeling module 2, and the topology invariant prediction module 3 based on the prediction error, thereby optimizing model performance. The parameter adjustment adopts an adaptive learning rate method, adjusting the learning rate according to the magnitude and trend of the prediction error.

[0151] In an internationalized curriculum environment, parameter tuning requires balancing global and regional optimization. The system can maintain a global model and multiple regionally specialized models, dynamically selecting the most suitable model parameters based on data distribution and the prediction task. For example, for learner groups with specific cultural backgrounds, it may be necessary to adjust feature weights or prediction thresholds to improve prediction accuracy.

[0152] The strategy optimization unit 63 is connected to the parameter adjustment unit 62 and is used to optimize the feature selection strategy, trajectory modeling strategy, and prediction algorithm strategy to improve the overall system performance. The strategy optimization adopts a reinforcement learning method, using the improvement in system performance as a reward signal, and finding the optimal strategy through trial and error.

[0153] In the application of internationalized courses, strategy optimization needs to consider a balance between diversity and adaptability. The system pursues prediction accuracy, computational efficiency, and adaptive flexibility simultaneously through multi-objective optimization. For example, for regions with limited resources, a simplified strategy with higher computational efficiency may be preferred; while for regions with abundant data, a more complex but accurate prediction strategy may be adopted.

[0154] like Figure 8As shown, the persistent homology computation unit 14 includes a boundary operator construction module 141, a matrix reduction module 142, a persistent barcode generation module 143, and a statistical feature extraction module 144.

[0155] The boundary operator construction module 141 is used to construct the boundary operator matrix of the simplex, recording the inclusion relationship between simplexes of different dimensions. The boundary operator matrix $\partial_n$ defines the relationship between an n-dimensional simplex and an (n-1)-dimensional simplex, where the matrix element $\partial_n(i,j)$ indicates whether the j-th n-dimensional simplex contains the i-th (n-1)-dimensional simplex. If it does, the value is 1 or -1 (depending on the orientation of the simplex), otherwise it is 0.

[0156] In the analysis of learning behavior in internationalized courses, boundary operator matrices can represent the inclusion and transformation relationships between learning activities. For example, completing a course module (high-dimensional simplex) involves multiple smaller learning activities (low-dimensional simplex), and the boundary operator matrix records this hierarchical relationship.

[0157] The matrix reduction module 142 is connected to the boundary operator construction module 141 and is used to perform row reduction operations on the boundary operator matrix to identify the generators of the homology group. The matrix reduction uses the Smith canonical form algorithm to transform the boundary operator matrix into a diagonal matrix form, thereby identifying the generators and structure of the homology group.

[0158] In learning data analysis, matrix reduction helps identify basic patterns and structural characteristics of learning behaviors. This is particularly valuable for analyzing behavioral differences among learners from different cultural backgrounds in internationalized courses, revealing the fundamental components of different learning styles.

[0159] The persistent barcode generation module 143 is connected to the matrix reduction module 142 and is used to record the birth and death scales of topological features to generate persistent barcode representations. The persistent barcode is the core output of persistent homology analysis, recording the scale parameter values ​​for the appearance and disappearance of each topological feature (such as connected components, cycles, and holes). The barcode is typically represented as a set of intervals, where each interval [birth, death] corresponds to the lifecycle of a topological feature.

[0160] In international curriculum learning analysis, persistent patterns can reflect the stability and importance of learning patterns. Long-lasting features usually represent stable patterns of learning behavior and may have a stronger correlation with learning outcomes; while short-lived features may represent noise or temporary behavior.

[0161] The statistical feature extraction module 144 is connected to the persistent barcode generation module 143 and is used to extract statistical features from the persistent barcodes, including lifecycle distribution, persistence metrics, and topological complexity measures. Statistical features include the number of barcodes, average lifecycle length, longest lifecycle, lifecycle distribution entropy, etc., which reflect the topological complexity and structural characteristics of the dataset.

[0162] In an internationalized curriculum environment, these statistical characteristics can be used to compare the complexity of learning behaviors among learners from different cultural backgrounds. For example, some learner groups may exhibit higher topological complexity (more cycles and holes), while other groups may exhibit simpler topological structures (fewer cycles and more connected components). These characteristics can help educators understand different learning styles and optimize curriculum design to accommodate diverse learning needs.

[0163] like Figure 9 As shown, the trajectory feature extraction unit 24 includes a curvature calculation module 241, a velocity field analysis module 242, a geodesy deviation quantification module 243, and a feature fusion module 244.

[0164] The curvature calculation module 241 is used to calculate the curvature of the learning trajectory on the knowledge manifold, quantifying the degree of curvature of the learning path. The curvature calculation employs a discrete curve curvature estimation method, estimating the local curvature by determining the radius of a circle using three adjacent points. A larger curvature indicates a more drastic change in the learning path, potentially reflecting difficulties encountered by the learner at that stage or a shift in learning strategies.

[0165] In internationalized curvature courses, curvature analysis can identify turning points and challenging areas in the knowledge acquisition process for different learners. For example, when learners encounter concepts or cases with significant cultural differences, they may exhibit higher curvature, indicating a significant shift in their learning path. Based on this information, the system can provide learners with targeted support and explanations to help them overcome cultural barriers.

[0166] The velocity field analysis module 242 is used to calculate the speed of progress of the learning trajectory at different knowledge points and identify points of change in learning speed. Velocity field analysis obtains the learning velocity vector by calculating the rate of change of the knowledge state vector over time. Velocity field visualization can intuitively display acceleration and deceleration regions in the learning process, helping to identify learning bottlenecks and efficient learning stages.

[0167] In international education, velocity field analysis can reveal the characteristics of knowledge acquisition among learners with different educational backgrounds. For example, learners with relevant prerequisite knowledge may exhibit a higher learning speed in a specific area, while learners lacking background knowledge may need more time to master basic concepts. By analyzing these patterns, the system can adjust the pace and depth of learning content, providing a more personalized learning experience.

[0168] The geodesic deviation quantification module 243 is used to calculate the geodesic distance between the actual learning trajectory and the theoretical optimal path, thus quantifying the learning efficiency. The geodesic deviation is obtained by integrating the point-to-point distance between the actual trajectory and the geodesic line. The larger the geodesic deviation, the further the learning path deviates from the optimal path, and the lower the learning efficiency may be.

[0169] In an internationalized curriculum environment, geodesic bias analysis can help understand the efficiency differences among different learning methods. For example, learners from certain cultural backgrounds may prefer exploratory learning (high geodesic bias but potentially better knowledge transfer ability), while other learners may prefer direct learning (low geodesic bias but potentially lacking flexible application ability). Based on this information, the system can provide teaching strategy suggestions that balance different learning styles.

[0170] The feature fusion module 244 is connected to the curvature calculation module 241, the velocity field analysis module 242, and the geodesic deviation quantification module 243, respectively, and is used to integrate various geometric features to generate a comprehensive trajectory feature vector. The feature fusion adopts a weighted combination method, and the weights are dynamically adjusted based on the contribution of each feature to the prediction of the learning effect.

[0171] In internationalized curricula, feature integration needs to consider the relative importance of features across different educational backgrounds. By analyzing learning data from different groups, the system can adjust feature weights for learners from diverse cultural backgrounds, improving the accuracy and adaptability of predictions.

[0172] like Figure 10 As shown, the topological invariant processing unit 32 includes a feature space transformation module 321, an invariant identification module 322, a stability evaluation module 323, and a weight optimization module 324.

[0173] The feature space transformation module 321 is used to map topological features and trajectory features to a latent space with topology-preserving properties, thus retaining the topological structure of the data. The feature space transformation employs various dimensionality reduction and embedding techniques, such as t-SNE and UMAP, which can preserve the local and global structure of the data in low-dimensional representations.

[0174] In the analysis of internationalized curricula, feature space transformation helps to visualize and compare the behavioral patterns of learners from different cultural backgrounds. By mapping high-dimensional learning behavior data to two- or three-dimensional space, the distribution and clustering of learner groups can be observed intuitively, revealing potential culture-related patterns.

[0175] The invariant identification module 322 is connected to the feature space transformation module 321 and is used to identify topological invariants in the feature representation, including connectivity, ring structures, and high-dimensional holes. Invariant identification is based on persistent homology analysis results, extracting topological features that remain stable under different deformations. These invariants are of great significance for understanding the essential structure of learning behavior.

[0176] In an internationalized educational environment, topological invariants can reveal common learning patterns across cultures. For example, regardless of cultural background, successful learners may exhibit specific patterns of knowledge connection or learning cycle structures. Identifying these invariants helps in designing more universally applicable international curricula and teaching strategies.

[0177] The stability assessment module 323 is connected to the invariant identification module 322 and is used to assess the stability of topological features to disturbances and calculate the topological stability index. Stability assessment is achieved by adding controlled disturbances to the original data and observing the degree of change in topological features. Stability index Defined as the retention rate of topological features under perturbation, with a value range of [0,1]. A larger value indicates higher stability.

[0178] In internationalized curricula, stability assessment is particularly important because it helps identify learning patterns that remain stable across cultural contexts. High stability characteristics may represent more general learning patterns, while low stability characteristics may be more influenced by cultural or individual factors.

[0179] The weight optimization module 324 is connected to the invariant identification module 322 and the stability evaluation module 323, respectively. It is used to assign dynamic weights to topological features based on the importance and stability of invariants, generating weighted feature representations. The weight optimization employs a gradient-based optimization method, and the objective function comprehensively considers both prediction accuracy and model stability.

[0180] In an internationalized curriculum environment, weight optimization needs to balance generality and specificity. The system can maintain a set of core feature weights applicable to all learners, while simultaneously adding weight adjustments for specific groups based on cultural background and learning styles. This balancing strategy can improve prediction accuracy while maintaining the model's cross-cultural adaptability.

[0181] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made under the concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A neural network-driven intelligent adaptation and quality assessment system for internationalized courses, characterized in that: include: The topology feature extraction module is used to collect multidimensional data of learning behavior, construct a learning behavior feature space, extract the topological structure features of learning behavior through topological data analysis technology, and generate topological feature vectors. The spatiotemporal learning trajectory modeling module is connected to the topological feature extraction module. It is used to receive the topological feature vector, construct a knowledge state manifold, represent the learning process as a trajectory on the knowledge manifold, calculate the deviation between the actual learning trajectory and the theoretical optimal path, and extract the geometric characteristics of the learning trajectory. The topological invariant prediction module is connected to the spatiotemporal learning trajectory modeling module. It is used to receive the geometric characteristics of the learning trajectory, construct a prediction model based on topological invariants, and generate learning effect prediction results. The course adaptation and recommendation module is connected to the topological invariant prediction module. It is used to receive the learning effect prediction results and generate personalized course recommendations based on learner characteristics and prediction results. The quality assessment module, connected to the topological invariant prediction module, is used to receive the learning effect prediction results, assess the correlation between course quality and learning effect, and generate a quality assessment report. An adaptive control module is connected to the topology feature extraction module, the spatiotemporal learning trajectory modeling module, and the topology invariant prediction module, respectively, and is used to dynamically adjust model parameters and optimize prediction performance based on the deviation between the prediction results and the actual results.

2. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 1, characterized in that, The topology feature extraction module includes: The data acquisition unit is used to collect learning behavior data, including timestamps, learning activity types, activity durations, interaction intensity, learning content locations, and feedback results, to construct a six-dimensional learning behavior feature space. A point cloud processing unit, connected to the data acquisition unit, is used to standardize and reduce the dimensionality of the original learning behavior data to generate learning behavior point cloud data. The complex construction unit, connected to the point cloud processing unit, is used to construct a Vietoris-Rips complex based on a distance matrix, set multi-level distance thresholds, and construct a simple complex sequence. The persistent homology computation unit, connected to the complex construction unit, is used to compute the persistent graph of the Betti number of each dimension as a function of a threshold, extract the statistical features of the persistent graph, and generate a topological summary vector. The feature mapping unit, connected to the persistent cohomology computing unit, is used to fuse topological features with traditional learned features, establish a feature importance scoring mechanism, and generate the final topological feature vector.

3. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 1, characterized in that, The spatiotemporal learning trajectory modeling module includes: The knowledge state assessment unit is used to assess learners’ mastery of knowledge points based on test, assignment and interaction data, and to construct a time series knowledge state vector, which is represented as a point in the knowledge state space. Knowledge manifold construction unit, used to define Riemannian metrics based on knowledge point association networks, construct the geometric structure of the knowledge space, and establish the local coordinate system of the knowledge manifold; The geodesic calculation unit, connected to the knowledge manifold construction unit, is used to define the initial and target knowledge states, apply the geodesic algorithm to calculate the optimal learning path, and generate a reference geodesic trajectory. The trajectory feature extraction unit is connected to the knowledge state evaluation unit and the geodesic calculation unit, respectively, and is used to calculate the geometric characteristics of the actual learning trajectory, measure the deviation between the actual trajectory and the geodesic, and extract the trajectory feature vector.

4. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 1, characterized in that, The topological invariant prediction module includes: The feature fusion unit receives topological feature vectors and trajectory feature vectors, applies nonlinear transformations to map the features to the latent space, and generates a comprehensive feature representation. A topological invariant processing unit, connected to the feature fusion unit, is used to extract topological invariants from features, construct an invariant weight allocation mechanism, and generate a weighted feature representation. The temporal dependency modeling unit, connected to the topological invariant processing unit, is used to process the temporal dependency of learning behavior sequences, capture long and short-term learning patterns, and generate context-enhanced feature representations. A multi-level prediction generation unit, connected to the temporal dependency modeling unit, is used to perform multi-level predictions from the behavior level to the knowledge level and then to the ability level, and generate learning effect prediction results. The prediction quality assessment unit, connected to the multi-level prediction generation unit, is used to calculate prediction errors and confidence intervals, evaluate topological stability indicators, and generate prediction quality reports.

5. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 1, characterized in that, The course adaptation and recommendation module includes: The learner profiling unit is used to build a multi-dimensional feature model of learners based on learning behavior data and knowledge status data, and to identify learning styles, knowledge base and learning goals. The course feature analysis unit is used to extract course content features, structural features, and difficulty features, and to construct a course feature vector. The matching degree calculation unit, connected to the learner profiling unit and the course feature analysis unit, is used to calculate the degree of matching between learner features and course features and generate a matching degree score. The recommended path generation unit, connected to the matching degree calculation unit and the topology invariant prediction module, is used to generate personalized learning paths and course recommendations based on the matching degree score and learning effect prediction results.

6. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 1, characterized in that, The quality assessment module includes: The multi-dimensional indicator collection unit is used to collect multi-dimensional evaluation indicators such as course content quality, instructional design quality, interactive experience quality, and learning support quality. The correlation analysis unit, connected to the multi-dimensional indicator acquisition unit and the topological invariant prediction module, is used to analyze the correlation between quality indicators and learning effect prediction results, and to identify key influencing factors. The quality scoring unit, connected to the correlation analysis unit, is used to quantitatively score the quality of various aspects of the course based on the correlation analysis results and generate a comprehensive quality score. An improvement suggestion unit, connected to the quality scoring unit, is used to generate targeted improvement suggestions based on the quality score and correlation analysis results, forming a quality assessment report.

7. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 1, characterized in that, The adaptive control module includes: The performance comparison unit is used to compare the predicted learning performance with the actual learning performance, calculate the prediction error, and identify model deficiencies. The parameter adjustment unit, connected to the effect comparison unit, is used to dynamically adjust the parameters of the topology feature extraction module, the spatiotemporal learning trajectory modeling module, and the topology invariant prediction module based on the prediction error, thereby optimizing the model performance. The strategy optimization unit, connected to the parameter adjustment unit, is used to optimize the feature selection strategy, trajectory modeling strategy, and prediction algorithm strategy to improve the overall system performance.

8. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 2, characterized in that, The persistent coherence calculation unit includes: The boundary operator construction module is used to construct the boundary operator matrix of the simple complex, which records the inclusion relationship between simplexes of different dimensions; The matrix reduction module, connected to the boundary operator construction module, is used to perform row reduction operations on the boundary operator matrix and identify the generators of the homology group. A persistent barcode generation module, connected to the matrix reduction module, is used to record the birth and death scales of topological features and generate persistent barcode representations. The statistical feature extraction module, connected to the persistent barcode generation module, is used to extract statistical features from the persistent barcode, including lifecycle distribution, persistence index, and topological complexity measure.

9. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 3, characterized in that, The trajectory feature extraction unit includes: The curvature calculation module is used to calculate the curvature of the learning trajectory on the knowledge manifold, quantifying the degree of bending of the learning path; The velocity field analysis module is used to calculate the speed at which the learning trajectory progresses across different knowledge points and to identify points where the learning speed changes. The geodesic deviation quantification module is used to calculate the geodesic distance between the actual learning trajectory and the theoretical optimal path, thereby quantifying the learning efficiency. The feature fusion module is connected to the curvature calculation module, the velocity field analysis module, and the geodesy deviation quantification module, respectively, and is used to integrate various geometric features to generate a comprehensive trajectory feature vector.

10. The neural network-driven intelligent adaptation and quality assessment system for internationalized courses according to claim 4, characterized in that, The topological invariant processing unit includes: The feature space transformation module is used to map topological features and trajectory features to a latent space with topology preservation properties, thus preserving the topological structure of the data. An invariant identification module, connected to the feature space transformation module, is used to identify topological invariants in the feature representation, including connectivity, ring structures, and high-dimensional holes. The stability assessment module, connected to the invariant identification module, is used to assess the stability of topological features to disturbances and calculate the topological stability index. The weight optimization module is connected to the invariant identification module and the stability evaluation module, respectively, and is used to assign dynamic weights to topological features based on the importance and stability index of invariants to generate weighted feature representations.

Images