Online education learning path optimization system based on big data and intelligent analysis

Abstract

The invention relates to the technical field of online education path optimization, and discloses an online education learning path optimization system based on big data and intelligent analysis, and the system comprises a data collection module which obtains basic data of a user; the user portrait generation module is used for generating a user portrait; the path generation module is used for generating an initial path candidate set according to the user portrait and screening the initial path candidate set to obtain a first learning path; the learning effect analysis module is used for obtaining a learning effect analysis result according to the historical learning performance data, the first learning data and a preset learning obstacle evaluation model; and the optimization module judges whether an obstacle exists or not, if yes, a plurality of nodes are added / reduced in the first learning path to obtain an optimized learning path, and otherwise, the first learning path is kept unchanged. The first learning path is adjusted according to the user portrait and the learning effect analysis result, personalized optimization is performed by integrating the learning effect and the learning habit, and improvement of the learning efficiency is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of online education path optimization, and particularly to an online education learning path optimization system based on big data and intelligent analysis. Background Art

[0002] In the context of the rapid development of modern educational informatization, traditional online education models have many pain points such as insufficient personalization, low learning efficiency, and a single learning path. Existing learning platforms are difficult to accurately match the knowledge level, learning style, and learning goals of users, resulting in serious constraints on the learning experience and learning effect. Therefore, there is an urgent need for a learning path optimization system that can be intelligent, dynamic, and personalized.

[0003] In summary, the traditional online education model has at least one technical problem among insufficient personalization, low learning efficiency, and a single learning path. Summary of the Invention

[0004] In view of this, the embodiments of the present invention provide an online education learning path optimization system based on big data and intelligent analysis to solve the technical problems of a single learning path and low learning efficiency in existing online education platforms.

[0005] To achieve the above object, an online education learning path optimization system based on big data and intelligent analysis is provided, which includes:

[0006] A data collection module for obtaining basic user data;

[0007] A user portrait generation module for generating a user portrait according to the basic user data;

[0008] A path generation module for generating an initial path candidate set according to the user portrait, screening the initial path candidate set to obtain a first learning path, and the first learning path includes several nodes;

[0009] A learning effect analysis module for obtaining first learning data and historical learning performance data when the user learns according to the first learning path, and obtaining an analysis result of the learning effect of the current node according to the historical learning performance data, the first learning data, and a preset learning obstacle evaluation model;

[0010] An optimization module for judging whether there is an obstacle in the current node according to the analysis result of the learning effect of the current node. If it is judged that there is an obstacle, several nodes are added to or removed from the first learning path according to the analysis result of the learning effect of the current node and the user portrait to obtain an optimized learning path. If it is judged that there is no obstacle, the first learning path remains unchanged.

[0011] The above technical solution has the following beneficial technical effects:

[0012] The present invention evaluates the learning effect through historical learning performance data and first learning data, determines whether there are obstacles in the learning path, and when there are obstacles, adjusts the first learning path according to the user profile and the analysis result of the learning effect, so as to optimize the learning path. By using the user profile to adjust the first learning path, not only the learning effect is considered during the adjustment process, but also the user's learning habits are combined, ensuring the personalization during path optimization and facilitating the improvement of learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The drawings are used to better understand the present invention and do not constitute an improper limitation to the present invention. Among them:

[0014] Figure 1 is a schematic structural diagram of an online education learning path optimization system based on big data and intelligent analysis of the present invention;

[0015] Figure 2 is a schematic structural diagram of a user profile generation module in an embodiment of the present invention;

[0016] Figure 3 is a schematic structural diagram of a user profile generation unit in an embodiment of the present invention;

[0017] Figure 4 is a schematic structural diagram of a path generation module in an embodiment of the present invention;

[0018] Figure 5 is a schematic structural diagram of a learning effect analysis module in an embodiment of the present invention;

[0019] Figure 6 is a schematic structural diagram of an optimization module in an embodiment of the present invention;

[0020] Figure 7 is a flowchart of an online education learning path optimization method based on big data and intelligent analysis in an embodiment of the present invention;

[0021] Figure 8 is a schematic structural diagram of a computer system in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following describes exemplary embodiments of the present invention with reference to the drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted below.

[0023] Embodiment 1

[0024] As Figure 1 shown, an online education learning path optimization system based on big data and intelligent analysis is provided in an embodiment of the present invention, which includes:

[0025] A data acquisition module, configured to acquire basic user data related to online education learning;

[0026] A user portrait generation module, configured to generate a user portrait according to the

[0027] basic user data;

[0028] A path generation module, configured to generate an initial path candidate set according to the user portrait, screen the initial path candidate set to obtain a first learning path, and the first learning path includes several nodes;

[0029] A learning effect analysis module, configured to acquire first learning data and historical learning performance data when the user learns according to the first learning path, and obtain an analysis result of the learning effect of the current node according to the historical learning performance data, the first learning data, and a preset learning obstacle evaluation model;

[0030] An optimization module, configured to determine whether there is an obstacle in the current node according to the analysis result of the learning effect of the current node. If it is determined that there is an obstacle, several nodes are added or reduced in the first learning path according to the analysis result of the learning effect of the current node and the user portrait to obtain an optimized learning path. If it is determined that there is no obstacle, the first learning path remains unchanged.

[0031] Specifically, the data acquisition module includes: a data acquisition unit, a protection unit, and a preprocessing unit.

[0032] The data acquisition unit is configured to acquire initial user data. The data acquisition unit includes several acquisition channels, and the several acquisition channels include but are not limited to learning platforms, mobile applications, online test systems, and learning terminals, etc. The learning behavior of the user is captured through the several acquisition channels to obtain the initial user data, and the initial user data includes but is not limited to the user's learning resource usage preferences, learning time, learning content, exercise correct rates when learning different contents, and learning efficiency, etc.

[0033] A protection unit is used to add noise to the user's initial data to obtain the first basic data. In the protection unit, the differential privacy algorithm is adopted to add noise to the user's initial data. Adding noise is beneficial to protecting individual privacy information. While ensuring user privacy, it still allows valuable statistical analysis of the data. The differential privacy algorithm adopts centralized differential privacy. When the user's initial data enters the server, the server adds noise to the user's initial data. Centralized differential privacy is applicable when the server is trusted. When adding noise, either the Gaussian method or the Laplace method can be used. Through the differential privacy algorithm, it is ensured that the personal privacy of learners is effectively protected during the data processing process. Valuable learning behavior characteristics can be extracted without disclosing personal sensitive information. This privacy protection method not only complies with data security regulations but also builds user trust in the learning platform.

[0034] A preprocessing unit is used to delete and repair outliers in the first basic data to obtain the user's basic data. Specifically, a deep learning algorithm is used to identify outliers. The deep learning algorithm includes:

[0035] An autoencoder, which compresses the input data into a low-dimensional representation through an encoder and then reconstructs the input data through a decoder. Outlier data is difficult to be reconstructed, and the reconstruction error of the outlier data is thus identified.

[0036] A generative adversarial network (GAN), which consists of a generator and a discriminator. The generator is responsible for generating samples of normal data, and the discriminator judges whether the input data is real data. When abnormal data is input, the generator cannot generate similar data, and the discriminator will identify it as abnormal data.

[0037] A variational autoencoder (VAE), which is a probabilistic generative model that reconstructs the input data through the latent distribution of the input data. The reconstruction error of outlier data is large, and the distribution of the latent variables is significantly different from that of normal data.

[0038] A deep learning model based on time series, which uses models such as long short-term memory networks or gated recurrent units to capture the time-dependent relationship of the input data, that is, the normal time series pattern of the input data. When models such as long short-term memory networks or gated recurrent units predict future data points, if the prediction error is large, it indicates the occurrence of an abnormal event, and the input data at this time is abnormal data.

[0039] Based on the Graph Neural Network (GNN), GNN can effectively capture the node relationships of the input data and learn the feature and structural patterns of the nodes. Nodes of abnormal data usually exhibit different patterns in terms of features or structures from normal nodes, thus identifying the abnormal data.

[0040] Specifically, when repairing outliers, methods such as deleting outliers, replacing outliers, data interpolation, or historical learning algorithm models can be used. The method of deleting outliers directly deletes the outliers. This method is simple and convenient, but there is a risk of data loss. The method of replacing outliers replaces the outliers by calculating the average, median, mode, or mean square filling method of the first basic data. The method of data interpolation uses interpolation methods to repair outliers. The interpolation methods include, but are not limited to, linear interpolation, spline interpolation, and moving average interpolation. For the model-based repair, the first basic data with outliers removed is input into the historical learning algorithm model. The historical learning algorithm model extracts valuable information from the first basic data with outliers removed and repairs and supplements the outliers. This data preprocessing method based on context and learning patterns improves the usability and accuracy of the data.

[0041] Specifically, the data acquisition unit further includes: a standardization subunit and an adaptive acquisition unit. The standardization subunit is used to standardize the user's initial data using feature engineering algorithms. The feature engineering algorithms not only perform simple numerical normalization but also consider multi-dimensional factors such as the learner's learning scenario, learning difficulty, and learning stage to construct more complex and accurate feature representations. This context-based feature engineering method can better capture the potential patterns and correlations in the data. The adaptive acquisition unit obtains the learner's learning behavior data in real time and determines whether the learner's learning habits have changed based on the learner's learning behavior data. When the learner's learning habits change, the type of user initial data obtained by the data acquisition unit also changes accordingly. For example, when the system detects a significant change in the learner's learning behavior, it can automatically adjust the depth and breadth of data acquisition to capture more valuable learning behavior features. This dynamic and intelligent data acquisition method enables the system to respond more agilely to the personalized learning needs of learners.

[0042] The user profile generation module is used to generate a user profile based on the user's basic data. Specifically, as Figure 2 shown, in the user profile generation module, it specifically includes: a cognitive ability evaluation unit, a learning style classification unit, a knowledge graph construction unit, a learning motivation analysis unit, and a user profile generation unit.

[0043] The cognitive ability evaluation unit is used to obtain the cognitive ability evaluation score of the user according to the basic user data and a preset cognitive ability evaluation model.

[0044] Specifically, in the cognitive ability evaluation unit, it includes: a first feature extraction subunit, a cognitive ability analysis subunit, and a cognitive ability evaluation subunit. The first feature extraction subunit is used to extract features from the basic user data to obtain a number of first feature vectors. In the first feature extraction subunit, the principal component analysis method and the mutual information feature selection algorithm are used to screen out the most representative features from a large amount of basic user data as a number of first feature vectors. For example, for learning behavior data, key features such as learning duration, knowledge point completion rate, and exercise correct rate will be extracted. The goal of feature extraction is to convert complex raw data into structured and quantifiable feature vectors. The cognitive ability analysis subunit is used to use a multi-layer neural network and a long short-term memory network algorithm for a number of the first feature vectors to obtain the user's learning efficiency, knowledge mastery depth, and learning progress speed. The cognitive ability evaluation subunit is used to input the learning efficiency, the knowledge mastery depth, and the learning progress speed into the preset cognitive ability evaluation model to obtain the cognitive ability evaluation score of the user. The preset cognitive ability evaluation model has the following formula: Cognitive ability score = α * learning efficiency + β * knowledge mastery depth + γ * learning progress speed; where α is the first cognitive ability weight, β is the second cognitive ability weight, and γ is the third cognitive ability weight. The first cognitive ability weight, the second cognitive ability weight, and the third cognitive ability weight are weight coefficients that are dynamically adjusted through machine learning algorithms. For example, for a programming learner, the system will analyze the speed of solving programming problems, code complexity, algorithm design ability, etc. to comprehensively evaluate the user's cognitive ability in the programming direction.

[0045] Learning efficiency is a measure of the learning output effect of the user per unit time. It mainly models and analyzes the user behavior data through a multi-layer neural network (MLP) or a long short-term memory network (LSTM). MLP can use structured features such as learning duration, number of exercises, correct rate, and knowledge point mastery rate as inputs, optimize the hidden layer weights, identify the potential patterns of the user's learning behavior, and output the score of learning efficiency. For learning behavior data with time series characteristics (such as the daily learning duration changing over time), LSTM captures the changing rules of the user's learning behavior in the time dimension through memory and forgetting mechanisms and analyzes the dynamic changes of learning efficiency. Finally, the results of LSTM or MLP can quantify the learning achievements of the user per unit time as the evaluation value of learning efficiency.

[0046] The depth of knowledge mastery reflects the user's understanding and application level of knowledge points, and can be modeled and evaluated through the exercise data completed by the user, exam scores, as well as error and correction situations. The MLP inputs the user's test scores on specific knowledge points, knowledge point coverage rate, types of wrong questions and their correction rates, etc. into the network, and through hidden layer training, conducts aggregated analysis on the learning performance of each knowledge point, and outputs the depth of knowledge mastery score of the user. For time series characteristic exercise data (such as the change of multiple exercise scores on the same knowledge point over time), LSTM modeling can capture the learning progress trend of the user in the time dimension and reflect the dynamic change of their mastery of knowledge points. Finally, combining the analysis results of the neural network, the depth of knowledge mastery score can comprehensively reflect the comprehensive performance of the user on specific knowledge points.

[0047] The learning progress speed measures the growth rate of the user's learning achievements within a certain period of time, and mainly models the accumulation process of learning achievements through a time series model. For the sequence data of the daily learning progress (such as the number of knowledge points completed) and learning duration changing over time, LSTM can capture its dynamic rules, analyze the cumulative change trend of learning achievements, and output the score of the learning progress speed. If the learning progress data is structured (such as the ratio of the total learning duration to the total achievements), then the MLP can be used to analyze these features to quantify the change rate of the user's learning achievements within a certain period of time. Finally, the score of the learning progress speed can intuitively reflect the time efficiency and progress of the user's learning output.

[0048] The learning style classification unit is used to obtain the learning style classification result of the user by using the multi-class support vector machine algorithm and the confidence calculation method for the user basic data; specifically, in the learning style classification unit, it includes: a second feature extraction sub-unit, a classification sub-unit, a confidence calculation sub-unit, and a learning style output sub-unit.

[0049] The second feature extraction sub-unit is used to extract features from the user basic data to obtain a number of second feature vectors. The number of the second feature vectors includes but is not limited to the user's learning resource usage preferences, knowledge acquisition speed, exercise question answering patterns, learning time allocation, and depth of knowledge mastery, etc., and the learning resource usage preferences include but are not limited to video resources, text resources, interactive course resources, etc.

[0050] A classification subunit is used to obtain several classification results for several of the second feature vectors by using a multi-class support vector machine algorithm. The multi-class support vector machine algorithm uses a radial basis kernel function as the kernel function to effectively handle non-linear classification problems. By using the multi-class support vector machine algorithm to obtain classification results, users can be classified into preset learning style categories, and the preset learning style categories include but are not limited to visual, auditory, hands-on, logical, and social. The multi-class support vector machine algorithm realizes the accurate distinction of learning styles by finding the optimal classification hyperplane in the feature space. The algorithm not only focuses on the classification results but also can capture the subtle differences between learning styles.

[0051] A confidence calculation subunit is used to calculate the confidence of each classification result by using the softmax function according to several of the classification results. The formula for the confidence is as follows:

[0052]

[0053] In the formula, exp is the support vector machine score, and the support vector machine score is obtained according to the classification result. Specifically, by calculating the confidence of each classification result, not only the classification result of the learning style is given, but also the possibility of each learning style is quantified. Through confidence analysis, the system can identify the dominant learning style of the learner and at the same time reveal the potential characteristics of other learning styles.

[0054] A learning style output subunit is used to select the classification result with the highest confidence as the user's dominant learning style according to the confidence of each classification result, generate personalized recommendations according to the user's dominant learning style, and output the user's learning style classification result according to the dominant learning style, the personalized recommendation, and the confidence corresponding to the dominant learning style. Specifically, the user's learning style classification result includes the dominant learning style, the personalized recommendation, and the confidence corresponding to the dominant learning style. For example, for a user identified as visual, the corresponding personalized recommendations are charts, mind maps, and visual learning resources; for a hands-on user, the corresponding personalized recommendations are highly practical interactive courses and project-oriented learning content. The method adopted in the learning style classification unit is different from traditional questionnaire surveys or static classification methods. The learning style classification unit has stronger dynamic adaptability and can continuously learn and update the user's learning style characteristics. Through multi-dimensional feature synthesis, probabilistic classification, and high interpretability, the algorithm can accurately identify learning styles.

[0055] A knowledge graph construction unit for obtaining a user's mastery status knowledge graph by using a graph neural network algorithm based on the user's basic data. Specifically, in the knowledge graph construction unit, it includes: a basic node generation subunit, an association relationship generation subunit, an association strength calculation subunit, a structure construction subunit, an embedding subunit, a knowledge graph establishment subunit, and a knowledge graph optimization subunit.

[0056] The basic node generation subunit is used to extract a set of knowledge points from a preset learning field according to the user's basic data, and use the set of knowledge points as several basic nodes. The preset learning field includes but is not limited to mathematics, programming, and language learning, etc. The selection of the learning field is determined according to the user's basic data, and the user's basic data can reflect the user's current learning field. Each knowledge point in the set of knowledge points is a basic node, and each basic node contains detailed information about the knowledge point. The detailed information of the knowledge point includes knowledge name, difficulty level, associated resources, and skill tags. The associated resources include video resources, article resources, and exercise resources, etc. For example, for the knowledge point of "Python Basics", the node attributes include "Name: Python Basics", "Difficulty: Beginner", "Associated Resources: Tutorials, Exercises, Code Examples", etc.

[0057] The association relationship generation subunit is used to obtain the annotation information of the preset learning field and historical user learning data, generate the explicit relationships between each basic node and the basic weights of each basic node according to the annotation information, generate the implicit relationships between each basic node according to the historical user learning data, and generate the association relationships between each basic node according to the explicit relationships, the basic weights, and the implicit relationships. Specifically, the explicit relationship means that within the learning field, there is an internal logic and a clear teaching structure between the basic nodes with explicit relationships. The annotation information includes but is not limited to the syllabus of the learning field. The association relationships include but are not limited to prerequisite, inclusion, and similarity. Taking the prerequisite relationship as an example, "Variable Declaration" as the prerequisite knowledge of "Function Definition" can be represented by a directed edge through the explicit relationship, and the basic weight is assigned according to the annotation information. The basic weight is used to represent the importance of this explicit relationship. The implicit relationship means the potential connection between basic nodes. The implicit relationship is automatically mined through historical user learning data, and the historical user learning data is mainly the frequency of co-occurrence of basic nodes. For example, through historical user learning data, it can be known that after historical users learn the basic node - "Loop Statement", they will immediately learn the basic node - "Conditional Statement", which means there is an implicit relationship between the basic node - "Loop Statement" and the basic node - "Conditional Statement".

[0058] The association strength calculation subunit is used to obtain the learning data of the user in the learning field, extract features from the learning data in the learning field to obtain a number of strength feature vectors, and perform weighted calculation on the number of strength feature vectors to obtain the association strength. Specifically, the strength feature vectors include learning duration, completion rate, test score, and learning frequency, and the calculation formula of the association strength is as follows:

[0059] Association strength = α 1 · Learning duration + α 2 · Completion rate + α 3 · Test score + α 4 · Learning frequency;

[0060] In the formula, α 1 represents the first strength weight, α 2 represents the second strength weight, α 3 represents the third strength weight, α 4 represents the fourth strength weight. The first strength weight, the second strength weight, the third strength weight, and the fourth strength weight are all dynamically adjusted through machine learning algorithms. The association strength between each basic node and the user is reflected through the association strength, so as to provide a learning path more suitable for the user in path planning.

[0061] The structure construction subunit is used to construct a structural knowledge graph according to the association strength, a number of the basic nodes, and each association relationship between the basic nodes.

[0062] Specifically, the structure construction subunit is specifically used to: obtain the user's real-time learning data, use a number of the basic nodes as nodes, use the association relationships between the basic nodes as the edges between the nodes, adjust the basic weights of each basic node according to the user's real-time learning data, delete the basic nodes whose basic weights are less than the preset basic threshold, and also delete the edges involved in the deleted basic nodes, and assign a recommendation degree to each node according to the association strength to obtain a structural knowledge graph.

[0063] An embedding subunit, which is used to map the structure knowledge graph to a low-dimensional vector space by using a graph embedding algorithm to obtain a number of embedding vectors. Specifically, the embedding subunit is used to: adopt the Node2Vec or DeepWalk algorithm to learn the semantic expressions of each node through random walks, so as to map the structure knowledge graph to a high-dimensional vector space; secondly, adopt the multi-head attention mechanism (Multi-head Attention) to represent each node and edge in the structure knowledge graph as a number of initial embedding vectors; finally, aggregate the information between adjacent nodes through a graph neural network algorithm, and optimize a number of the initial embedding vectors according to the information between the adjacent nodes to obtain a number of embedding vectors. The graph neural network algorithm can adopt the Graph Convolutional Networks (GCN) and the Graph Attention Network (GAT). GCN updates the initial embedding vector of a node by aggregating the information of adjacent nodes. The initial embedding vector of each node is a weighted sum of the initial embedding vectors of its adjacent nodes, and the weights are determined by the edges between the nodes. Through the GCN, the local correlation information of the nodes can be captured, and through multiple convolutional operations, the initial embedding vector of the node gradually contains more extensive adjacent information, so as to reflect the position of the node in the local structure. GAT uses the attention mechanism to calculate the influence weight of adjacent nodes on the current node. The initial embedding vector of each node is a weighted sum of the embedding vectors of its adjacent nodes, and the weights are dynamically calculated by the attention mechanism. Through the multi-head attention mechanism and the multi-layer network structure, the GAT can capture both local correlation information and global graph structure relationships at the same time. The attention mechanisms of different heads can focus on different feature subspaces, enabling the initial embedding vector to reflect the global position of the node from multiple perspectives. Through GCN and GAT, the local correlation information and global graph structure relationships of the nodes are fused into the initial embedding vector to obtain the embedding vector. The local information enables the node embedding to reflect the specific position of the node in the knowledge graph and its relationship with adjacent nodes, and the global information enables the node embedding to capture the role and importance of the node in the entire structure knowledge graph. After multiple layers of processing by the graph neural network algorithm, the initial embedding vector of the node is continuously optimized, and finally a low-dimensional vector representation that can capture both local correlation information and reflect global graph structure relationships is obtained, that is, the embedding vector. These embedding vectors provide strong support for subsequent tasks such as learning path recommendation and knowledge association analysis.

[0064] A knowledge graph construction subunit, which is used to generate a mastery status knowledge graph by using a force-directed algorithm according to the structure knowledge graph and a number of the embedding vectors.

[0065] Specifically, the knowledge graph construction subunit is specifically used for: initializing the positions of all nodes, that is, randomly or according to a certain preset rule to initialize the positions of all nodes. Calculating the repulsive force between each pair of nodes, where the magnitude of the repulsive force is inversely proportional to the square of the distance between each pair of nodes. According to the structural knowledge graph, calculating the attractive force between the two nodes connected by each edge based on the optimal length of the edge and the distance between the two nodes connected by the edge. Updating the positions of all nodes according to the attractive force and the repulsive force, and the moving direction and speed of the nodes are determined according to the attractive force and the repulsive force received by the nodes. Repeating the above steps to update the positions of all nodes until the maximum number of iterations is reached or a preset termination condition is satisfied, obtaining a basic knowledge graph, where the preset termination condition is that the edges do not overlap. The optimal length of the edge is determined according to the size of the nodes and the density of the structural knowledge graph. In some cases, the layout can be further optimized by adjusting the curvature of the edges. For example, for the edges connecting distant nodes, the curvature can be appropriately increased to avoid overlapping with other edges or nodes. According to the basic knowledge graph and a number of the embedding vectors, displaying the embedding vectors at the nodes of the basic knowledge graph to obtain a mastery status knowledge graph. Each node on the mastery status knowledge graph is a knowledge point required to achieve the user's learning goal. Through the mastery status knowledge graph, the knowledge points that still need to be learned and the sequence of knowledge points can be reflected. By using the force-directed algorithm to optimize the layout of nodes and edges, the graph structure is clear and intuitive. The user can click on the nodes to view the detailed information of the knowledge points, including associated resources, learning progress, and learning effects. In addition, the system supports generating personalized knowledge graphs, and uses colors to mark the knowledge mastery status (for example, green indicates good mastery, red indicates weakness, and gray indicates not learned), making it clear at a glance for the user.

[0066] The knowledge graph optimization subunit is used to perform consistency verification on the mastery status knowledge graph. If the verification result is unqualified, the mastery status knowledge graph is corrected according to the verification result. If the verification result is qualified, the mastery status knowledge graph is output. Performing consistency verification on the mastery status knowledge graph is used to judge the rationality of the mastery status knowledge graph. If the judgment is that it is not reasonable, the edges in the mastery status knowledge graph are removed according to the verification result. If the judgment is that it is reasonable, a test learning path is generated according to the reasonable mastery status knowledge graph, and the mastery status knowledge graph is optimized according to the learning feedback data corresponding to the test learning path. The learning feedback data includes the user's test scores, learning duration, etc. The optimization content includes the weights of the edges and the representations of the nodes. The consistency verification is used to check whether the edges between the nodes are reasonable.

[0067] The learning motivation analysis unit is used to input the user's basic data into a preset learning motivation analysis model to obtain the user's learning motivation evaluation result. Specifically, in the learning motivation analysis unit, it includes: a motivation feature extraction subunit, a learning motivation score calculation subunit, and a learning motivation evaluation subunit.

[0068] A motivation feature extraction subunit, configured to extract features from the user's basic data to obtain a number of motivation feature vectors. The user's basic data includes learning time distribution, usage frequency of learning resources, types of learning resources used, completion rate of learning plans, consecutive learning days, setting of learning goals, learning goal achievement rate, and number of learning tasks completed. The motivation feature vectors include the proportion of learning input time, resource switching frequency, and goal achievement rate, etc.

[0069] Specifically, the motivation feature extraction subunit is specifically configured to: first perform normalization processing on the user's basic data to obtain normalized data, and then use an autoencoder to extract features from the normalized data to obtain a number of motivation feature vectors. In the process of performing normalization processing on the user's basic data to obtain normalized data, the user's basic data is uniformly converted into normalized data in the range of [0, 1].

[0070] Specifically, the autoencoder includes an encoder and a decoder. In the step of using the autoencoder to extract features from the normalized data to obtain a number of motivation feature vectors, specifically, the normalized data is input into the encoder to obtain a low-dimensional embedding vector. The decoder retains the key features in the low-dimensional embedding vector through a reconstruction loss function. The encoder maps the "resource switching frequency", "goal completion rate", and "learning plan completion time" of the learner into a potential three-dimensional embedding space and outputs a vector that can represent the learning motivation of the learner. The optimization goal of the autoencoder is to minimize the reconstruction error (such as mean square error) to ensure the effective extraction of learning motivation features.

[0071] A motivation clustering subunit, configured to use a clustering algorithm for a number of the motivation feature vectors to obtain a number of motivation clustering results, and generate a number of learning motivation types according to the number of motivation clustering results.

[0072] Specifically, in the motivation clustering subunit, the clustering algorithm selects K-means or DBSCAN. Each clustering result represents a typical learning motivation type, such as "high internal drive type", "task-oriented type", "interest-driven type", etc. The system automatically determines the optimal number of clusters based on the distribution characteristics of the learners in the embedding space. For example, the elbow method or silhouette score is used to evaluate the rationality of the clustering results. For each clustering result, the system further analyzes the eigenvalue of its center point, such as "high learning input", "high goal achievement rate", etc., to provide an explanatory motivation label. In this way, the system can identify the main learning motivation types of the learners.

[0073] A learning motivation score calculation subunit, configured to input a plurality of the motivation feature vectors into a preset learning motivation analysis model to obtain the learning motivation score of a user.

[0074] Specifically, the formula of the learning motivation analysis model is as follows:

[0075] Learning motivation score = f(learning engagement, goal attainment rate, learning coherence);

[0076] In the formula, the learning engagement, goal attainment rate, and learning coherence are all motivation feature vectors. The formula of the learning motivation analysis model involves all motivation feature vectors, not limited to the above three. The learning engagement is quantified and determined by the learning duration and the frequency of learning resource usage. The goal attainment rate is determined by the completion of learning tasks and the success rate of goal achievement. The learning coherence is determined by the completion of the learning plan and the number of consecutive learning days. The mapping function f is a Multilayer Perceptron (MLP), which dynamically adjusts the weights through training to balance the importance of the three dimensions. Finally, the learning motivation score, as a comprehensive indicator, reflects the magnitude of the learner's learning driving force.

[0077] A learning motivation evaluation subunit, configured to obtain a learning motivation evaluation result according to the learning motivation score and a plurality of the learning motivation types.

[0078] Specifically, the learning motivation evaluation result includes the learning motivation score, a plurality of the learning motivation types, and the psychological state. The psychological state analyzes the user's psychological state by using a psychological feature extraction algorithm based on the learning motivation score and a plurality of the learning motivation types. For example, sentiment analysis technology is used to identify the user's learning emotion (positive, negative, or neutral), and the psychological engagement is extracted by analyzing the interaction behaviors (such as the number of evaluations and questions) in the learning log or learning record. In addition, the system infers the user's goal orientation type (such as short-term goal or long-term goal) and frustration resistance ability through the correlation analysis of the clustering results and learning behavior characteristics. The output of the psychological features includes multiple dimensions, such as learning interest, concentration, and learning confidence.

[0079] As a preferred implementation manner of this embodiment, the learning motivation evaluation result of a certain user includes the following information: the learning motivation score is 0.85 (high), the motivation type is interest-driven, the psychological state is relatively high concentration, and the frustration resistance ability is medium. In addition, a visualization tool (radar chart or bar chart) is used to display the learning motivation evaluation result, which is convenient for users and educators to intuitively understand the learning motivation characteristics.

[0080] Specifically, the evaluation results of learning motivation can be dynamically updated through a closed-loop feedback mechanism. Each learning behavior generates new data inputs, and the system retrains the autoencoder and clustering model regularly to ensure the dynamic adaptability of motivation features. For example, when the user's learning input time increases or the learning goals are changed frequently, the system automatically adjusts the learning motivation score and motivation type. In addition, the system feeds back the analysis results to the user and educator, providing personalized learning suggestions such as "appropriately increase challenging tasks" or "guide the formulation of long-term learning goals", thereby further optimizing the learning experience.

[0081] The user portrait generation unit is used to perform weighted calculation based on the cognitive ability evaluation score, the learning style classification result, the mastery status knowledge graph, and the learning motivation evaluation result, and generate a user portrait according to the weighted calculation result.

[0082] Specifically, as Figure 3 shown, in the user portrait generation unit, it includes: a data integration subunit, a portrait embedding vector generation subunit, a label construction subunit, a comprehensive scoring subunit, an interactive radar generation subunit, and a portrait generation subunit.

[0083] The data integration subunit is used to adopt an ensemble learning algorithm to perform feature weighted fusion on the cognitive ability evaluation score, the learning style classification result, the mastery status knowledge graph, and the learning motivation evaluation result, and obtain a number of comprehensive feature vectors.

[0084] Specifically, the data integration subunit is specifically used for: first, integrating the cognitive ability evaluation score, the learning style classification result, the mastery status knowledge graph, and the learning motivation evaluation result to obtain a number of comprehensive feature vectors. The features of each dimension have been standardized into numerical representations and stored in a unified data structure (such as a database or a distributed file system). The integration process adopts an ensemble learning algorithm. For example, the cognitive ability evaluation score is represented as cognitive ability features (such as the depth of knowledge mastery) and learning styles (such as visual, auditory) after passing through the ensemble learning algorithm, and comprehensive weights are assigned according to the importance of a number of comprehensive feature vectors in learning path optimization. The specific values of the comprehensive weights are automatically adjusted through model training of historical learning effects. In addition, to avoid redundancy or conflict between a number of comprehensive feature vectors, the system adopts a feature selection algorithm (such as recursive feature elimination RFE), and only retains the features related to learning effect prediction.

[0085] The portrait embedding vector generation subunit is used to adopt a deep embedding learning algorithm for a number of comprehensive feature vectors to obtain a number of portrait embedding vectors;

[0086] Specifically, the portrait embedding vector generation subunit is specifically used for: mapping a number of comprehensive feature vectors into high-dimensional embedding vectors by using a deep embedding learning algorithm. The high-dimensional embedding vectors not only retain the semantic information of the comprehensive feature vectors but also capture the potential correlations between a number of comprehensive feature vectors. The mapping process uses a deep embedding learning algorithm (such as the multi-task learning network MTL), and then the high-dimensional embedding vectors are transformed into portrait embedding vectors by jointly optimizing the loss function. For example, learning style, cognitive ability, mastery degree knowledge graph, and motivation characteristics are respectively input as subtasks into a shared hidden layer, so as to generate portrait embedding vectors containing all-dimensional feature information.

[0087] The label construction subunit is used to classify a number of the portrait embedding vectors by using a supervised learning algorithm to obtain a number of label classification results, calculate the importance scores of each label classification result, and generate a number of labels according to the number of the label classification results and the importance scores corresponding to each label classification result.

[0088] Specifically, in the label construction subunit, the supervised learning algorithm includes the random forest algorithm or the distributed gradient boosting library algorithm. A number of label classification results are obtained through the supervised learning algorithm, such as "high cognitive potential type", "task-driven type", or "interest-oriented type", etc. At the same time, for each label classification result, the importance scores of each label classification result are also calculated, and labelized portrait summary information is generated based on these scores. For example, a certain user may be classified as "interest-oriented type", and the label summary information is: "The main features are high learning interest, rapid progress in knowledge mastery, and the motivation type is internal drive interest type."

[0089] The comprehensive scoring subunit is used to calculate the comprehensive score of the user based on a number of the label classification results.

[0090] Comprehensive score = w1·Knowledge mastery degree + w2·Cognitive ability + w3·Motivation intensity + w4·Behavior consistency comprehensive score;

[0091] In the formula, w1 is the first comprehensive scoring weight, w2 is the second comprehensive scoring weight, w3 is the third comprehensive scoring weight, w4 is the fourth comprehensive scoring weight. w1, w2, w3, and w4 are all dynamically adjusted through historical learning data. The comprehensive score ∈ [0, 100]. The comprehensive score is used as a part of the user portrait to intuitively reflect the comprehensive ability and potential of the user.

[0092] The interactive radar generation subunit generates an interactive radar chart according to a number of the label classification results.

[0093] Specifically, each axis of the interactive radar chart represents a label classification result, such as "knowledge mastery", "learning style preference", "motivation intensity", "learning coherence", etc. The eigenvalue of each user is represented by a broken line in the radar chart, intuitively showing their performance in different dimensions. For example, the radar chart of a user may show a relatively high "knowledge mastery", a medium "motivation intensity", and a relatively low "learning coherence", thus providing targeted guidance for educators. The radar chart supports dynamic interaction, and users can choose to zoom in on a certain dimension to view the detailed eigenvalue or historical change trend.

[0094] The portrait generation subunit is configured to generate a user portrait according to the mastery status knowledge graph, several of the label classification results, the comprehensive score, and the interactive radar chart.

[0095] Specifically, the user portrait includes several display areas, which are respectively used to display the mastery status knowledge graph, several of the label classification results, the comprehensive score, and the interactive radar chart. The mastery status knowledge graph is presented in the form of nodes and edges. The nodes represent knowledge points, and the edges represent the association relationships between knowledge points. The knowledge mastery status of the user is intuitively presented through the color and size of the nodes. For example, green indicates the mastered knowledge points, red indicates insufficient mastery, and gray indicates unlearned. The size of the node is dynamically adjusted according to the learning time of the user on this knowledge point. In addition, a dynamic interaction function is also provided. Users can click on the node to view the detailed information of this knowledge point (such as learning resources, associated test scores, learning time, etc.) or recommended learning paths for this knowledge point. Learning suggestions can be analyzed and generated according to the user portrait, such as "strengthen the review of specific knowledge points" or "try high-difficulty learning tasks". The label classification results include learning progress display, etc., and the learning progress display refers to presenting the phased achievements and progress of the user through a time axis.

[0096] Specifically, the user portrait is dynamically updated through a closed-loop mechanism. Each learning behavior of the user (such as completing a task, change in test results) will trigger the recalculation of the user portrait and update the user portrait display in real time. For example, when the user masters new knowledge points or improves a certain dimension feature, the interactive radar chart and the mastery status knowledge graph will be automatically refreshed to reflect the latest learning status. The system supports version control, and users and educators can recall the historical user portraits at any time to compare the performance of the user at different stages.

[0097] Specifically, it also includes an interactive feedback function. In the user profile interface, users can evaluate the accuracy of the user profile or provide feedback on learning suggestions, while educators can put forward revision opinions on the content of the user profile or supplement the learning plan. These feedback messages will be used as data inputs for model improvement to further enhance the accuracy and applicability of user profile generation. For example, if most users feedback that the description of a certain dimensional feature is inaccurate, the feature weighting model will be automatically adjusted or the visualization presentation method will be optimized.

[0098] Through the above steps, the system constructs a complete integrated and visualized process for user intelligent profiling, which not only realizes the integration and display of multi-dimensional features, but also enhances the practicality of the profile through interactive functions, providing comprehensive support for personalized learning.

[0099] A path generation module, configured to input the user profile into a preset path evaluation model to obtain a first learning path;

[0100] Specifically, as Figure 4 shown, in the path generation module, it includes: an initial path candidate set construction unit, a candidate path feature extraction unit, a screening unit, and an initial path optimization unit.

[0101] The initial path candidate set construction unit is configured to generate an initial path candidate set according to the user profile. The initial path candidate set construction unit specifically includes: a profile parsing subunit, a knowledge point screening subunit, and an initial path candidate set construction subunit.

[0102] The profile parsing subunit is configured to extract a number of feature dimensions and user learning objectives according to the user profile; specifically, the feature dimensions include dimensions such as cognitive ability, learning style, mastered status knowledge graph, and learning motivation, and analyze the user's current knowledge mastery status, learning preferences, and potential learning obstacles according to the profile data. For example, the cognitive ability feature can help the system judge the user's knowledge absorption efficiency, and the knowledge graph can provide a clear structure of the user's mastered and unmastered knowledge points. In addition, the system extracts target knowledge points and expected time ranges through the user input learning objectives (for example, the short-term goal is to complete the learning of a certain knowledge module, and the long-term goal is to master a certain skill), providing clear target guidance for learning path generation.

[0103] A knowledge point screening subunit, configured to generate target knowledge points according to the user's learning objectives, and use a recursive algorithm based on the target knowledge points and several of the feature dimensions to obtain several prerequisite knowledge points; several of the feature dimensions include a mastery status knowledge graph, and the mastery status knowledge graph includes several nodes, and each node also has knowledge points. By using a recursive algorithm on the target knowledge points, the prerequisite knowledge points of the target knowledge points in the mastery status knowledge graph are found. The prerequisite knowledge points and the target value knowledge points are both part of the learning path, and the recursive algorithm can adopt a loop structure or function call.

[0104] An initial path candidate set construction subunit, configured to generate several initial paths according to several of the prerequisite knowledge points and the target knowledge points, and establish an initial path candidate set according to the several initial paths. The heuristic algorithm can adopt an A* search algorithm or a shortest path algorithm to search for all possible paths in the mastery status knowledge graph as the initial paths. The length (number of learning steps) and coverage (number of knowledge points) of the initial paths are both constrained by the learning objectives and learning time.

[0105] A candidate path feature extraction unit, configured to perform feature extraction on the initial path candidate set to obtain several candidate feature vectors;

[0106] Specifically, the several candidate feature vectors include knowledge point difficulty scores, initial path lengths, path difficulty gradients, path coverage, path relevance, and learning time prediction. The path difficulty gradient is used to reflect whether the difficulty distribution of the knowledge points in the initial path conforms to the user's ability and whether there is a progression from basic to advanced. The path difficulty gradient is calculated according to the difficulty scores of each knowledge point in the initial path and the total length of the initial path. The path coverage is the ratio of the number of several of the prerequisite knowledge points to the number of knowledge points passed by the initial path. The path relevance is a logical type based on the association strength to ensure that the initial path conforms to the semantic association between nodes. The learning time prediction combines the learning efficiency features in the user profile to predict the time required to complete the initial path and compares it with the user's target learning time to ensure the feasibility of the initial path.

[0107] Specifically, the formula for calculating the path difficulty gradient is as follows:

[0108] Path difficulty gradient = (highest knowledge point difficulty score - lowest knowledge point difficulty score) / total length of the initial path;

[0109] A screening unit for obtaining historical user data, and screening the initial path candidate set according to a number of the candidate feature vectors, the user profile, and the historical user data by using a collaborative filtering algorithm to obtain a first initial path set; the collaborative filtering algorithm is an intelligent analysis method applied to recommendation systems. It mines potential interest associations through the similarity between users or between items to provide personalized recommendations for users. In learning path optimization, the collaborative filtering algorithm can analyze other users with similar learning behaviors or characteristics to the target user based on the user profile, candidate feature vectors, and historical user data, so as to screen out the key parts of the successful learning paths of these users and recommend them to the target user. Collaborative filtering is usually divided into user-based collaborative filtering and item-based collaborative filtering: User-based collaborative filtering recommends the paths selected by them by identifying user groups with learning behaviors similar to the target user; Item-based collaborative filtering recommends optimized paths related to the paths already learned by the target user by analyzing the similarity between learning paths. This algorithm does not need to rely on the feature description of specific path content, but dynamically adjusts the recommended content based on big data analysis of user historical behaviors, providing precise support for learning path optimization.

[0110] Specifically, in the screening unit, it includes: a first screening subunit, a historical data acquisition subunit, a similarity calculation subunit, and a second screening subunit.

[0111] The first screening subunit is used to screen the initial path candidate set according to a number of candidate feature vectors corresponding to each initial path and a preset screening threshold to obtain a first screened path set; specifically, the categories of a number of feature vectors and the screening threshold are all path difficulty gradient, path coverage, path relevance, and learning time prediction. For each initial path in the initial path candidate set, if there is a feature vector that does not meet the screening threshold, the corresponding initial path is deleted, and all initial paths that meet the screening threshold form the first screened path set.

[0112] The historical data acquisition subunit is used to obtain historical user data; specifically, the historical user data includes historical user profiles.

[0113] The similarity calculation subunit is used to calculate the similarity of each learning path in the first screened path set according to the historical user data and the user profile;

[0114] Specifically, the similarity calculation formula is as follows:

[0115]

[0116] In the formula, u i represents the i-th eigenvalue of the user profile, and v i represents the i-th eigenvalue of the historical user profile.

[0117] A second screening subunit, configured to screen out a number of first screening paths with a similarity greater than a first threshold from the first screening path set as a first initial path set.

[0118] Specifically, the first threshold is a preset value, and the first threshold should not be too large. When the first threshold is too large, it will lead to insufficient samples in the first initial path set.

[0119] An initial path optimization unit, configured to extract a number of knowledge point sequences from the first initial path set, input the number of knowledge point sequences into a pre-trained recommendation model to obtain a number of recommendation scores, and select the first initial path corresponding to the highest recommendation score as the first learning path.

[0120] Specifically, in the initial path optimization unit, it includes: a knowledge point sequence extraction subunit, a recommendation score subunit, and an initial path optimization subunit. The knowledge point sequence extraction subunit is configured to extract a number of knowledge point sequences from the first initial path set. Specifically, the knowledge point sequence is each knowledge point in the first initial path and can be directly obtained from the first initial path. The recommendation score subunit,

[0121] is configured to input the number of knowledge point sequences into a pre-trained recommendation model to obtain a number of recommendation scores. The initial path optimization subunit is configured to select the one with the highest recommendation score in the first initial path set as the first learning path.

[0122] Specifically, the pre-trained recommendation model is based on a sequence recommendation algorithm (such as a transformer model or a long short-term memory network). When training, the recommendation model uses the learning paths of historical users, the learning effects corresponding to the historical user learning paths, and the historical user portraits as the training set. The learning effects corresponding to the historical user learning paths include, but are not limited to, learning completion rate and knowledge point mastery, etc. The historical user portrait includes the learning motivation and learning style of historical users. The recommendation model optimizes the ranking of each knowledge point in the first initial path by maximizing the recommendation score. For example, the system may find that the arrangement order of a certain knowledge point in the first initial path set is more in line with the user's learning pattern, so it preferentially recommends the first initial path as the first learning path.

[0123] Specifically, the first learning path not only covers the learning objective knowledge points, but also meets the personalized needs of users in terms of difficulty gradient, time feasibility, logical relevance, etc. For example, for a beginner in programming, the path generated by the system may start from the basic knowledge point "variable declaration" and gradually progress to "loop structure", "function call", "recursive algorithm", etc. For users with a programming foundation, the path may directly start from "recursive algorithm" and expand to more complex "dynamic programming". In addition, the system attaches learning resources (such as recommended videos, practice questions) and time plans to each knowledge point to help users complete learning tasks efficiently. It also includes a method for real-time monitoring and dynamic updating of the first learning path. Through the user's learning behavior data (such as learning completion rate, test scores) and learning feedback, the system optimizes the path. For example, if the user spends much more time than expected on a certain knowledge point and has a poor test score, the system will automatically insert relevant supplementary knowledge points or reduce the difficulty of subsequent knowledge points; conversely, if the user performs excellently, the system may skip some redundant knowledge points or recommend more challenging content. The dynamic adjustment is achieved through a reinforcement learning model (such as a path optimization algorithm based on Q-learning), and the reward function aims to improve the learning effect.

[0124] Specifically, the first learning path is presented to users and educators through an interactive interface. The first learning path is displayed in the form of a flowchart or a knowledge graph, with each knowledge point represented by a node, and the connection lines between nodes represent the learning order. The color and size of the nodes can reflect the difficulty and learning priority of the knowledge points. Users can click on the nodes to view detailed information (such as knowledge point content, associated resources) and provide feedback on the path (such as whether the path is reasonable, whether adjustment is needed). The feedback data is collected by the system and used for subsequent algorithm optimization. In addition, the user's learning progress is updated in real-time to the path diagram. For example, the completed knowledge points are marked in green, and the uncompleted ones are marked in gray.

[0125] The learning effect analysis module is used to obtain the first learning data and historical learning performance data when the user learns according to the first learning path, and obtain the learning effect analysis result of the current node according to the historical learning performance data, the first learning data, and a preset learning obstacle evaluation model.

[0126] Specifically, as Figure 5 shown, in the learning effect analysis module, it includes: a first learning data acquisition unit, a historical learning performance acquisition unit, and a learning effect analysis unit.

[0127] The first learning data acquisition unit is used to obtain the first learning data when the user learns according to the first learning path;

[0128] Specifically, when obtaining the first learning data, it is tracked and collected through the logs of the learning platform, and the learning behavior data of users is collected in real time, including knowledge point access records, learning duration, test scores, and usage of learning resources. These data are uploaded in real time through the API interfaces of the learning management system or the learning analysis platform and stored in the data storage module. The system adopts an event-driven method. Whenever a user completes an operation (such as completing a test question, submitting an assignment, or switching resources), the relevant data will be automatically updated to the database. For example, after a user completes an online test on "loop structure", the user's test score, answering time, and error rate will be recorded in real time.

[0129] A historical learning performance acquisition unit for acquiring historical learning performance data.

[0130] Specifically, the historical users corresponding to the historical learning performance data are similar in learning style to the current user, and the historical learning performance data includes multiple historical users.

[0131] A learning effect analysis unit for obtaining the learning effect analysis result of the current learning node according to the historical learning performance data, the first learning data, and a preset learning obstacle evaluation model.

[0132] Specifically, in the learning effect analysis unit, it includes: a performance analysis subunit, a mastery quantification subunit, a behavior pattern analysis subunit, and a learning effect analysis subunit.

[0133] The performance analysis subunit is used to calculate several performance deviations of the user according to the historical learning performance data and the first learning data.

[0134] Specifically, the average performance data is obtained by averaging the historical learning performance data. The performance data corresponding to the user is extracted from the first learning data, and the performance deviation of the user is calculated according to the average performance data and the user performance data. For example, if the test correct rate of most users for the "recursive algorithm" knowledge point is 80%, while the current user is only 40%, it is considered that the user's performance is poor and there are learning obstacles. Performance deviations include, but are not limited to, test correct rate, learning time deviation, and task completion situation deviation.

[0135] The mastery quantification subunit is used to calculate the knowledge mastery degree of the user according to the first learning data and a preset mastery quantification model.

[0136] Specifically, the formula of the mastery quantification model is as follows:

[0137]

[0138] In the formula, score i represents the score of the user on question i, and the question weighti The contribution weight of question i to the target knowledge point. The questions are designed in units of knowledge points, and each question is marked with the relevant knowledge point weight. The questions include, but are not limited to, multiple-choice questions, open-ended questions, project practices, and simulation scenario tests. The knowledge mastery degree is updated in real time to the knowledge graph of the mastery state, and the mastery degree of the knowledge point is represented by the color or size of the node. For example, green indicates high mastery and red indicates low mastery.

[0139] The behavior pattern analysis subunit is used to process the first learning data by using a clustering algorithm to obtain several behavior clustering results. According to the several behavior clustering results, it is judged whether the user behavior is abnormal. If it is judged that the user behavior is abnormal, the learning effect analysis result that there is an obstacle in the current learning node of the user in the first learning path is output, and the learning effect analysis subunit is skipped. If it is judged to be normal, the learning effect analysis subunit is entered.

[0140] Specifically, the clustering algorithm in the behavior pattern analysis subunit can adopt (k-means clustering algorithm or density-based clustering algorithm). The behavior clustering results include, but are not limited to, learning time, test correct rate, learning resource switching frequency, and / or task completion rate, etc. When judging whether each behavior clustering result is abnormal, the average value of the historical data of the behavior clustering result is calculated, and the average value of the historical data is used as the threshold. For example, if the learning time is too long but the correct rate is low, or the resources are frequently switched but the task is not completed, etc., it is judged as abnormal.

[0141] The learning effect analysis subunit is used to input several of the performance deviations into a preset learning obstacle evaluation model to obtain a learning obstacle score, and generate a learning effect analysis result of the current learning node according to the learning obstacle score and the knowledge mastery degree.

[0142] Specifically, the formula of the preset learning obstacle evaluation model is as follows:

[0143] Learning obstacle score = β 1 * Test correct rate deviation + β 2 * Learning time deviation + β 3 * Task completion situation deviation;

[0144] In the formula, β 1 represents the first effect evaluation coefficient, β 2 represents the second effect evaluation coefficient, β 3 represents the third effect evaluation coefficient. The sum of the β 1 , β 2 and β 3 is 1. The learning effect analysis result includes the learning obstacle score and the knowledge mastery degree.

[0145] It also includes a dynamic feedback unit, which is used to generate learning feedback based on the learning effect analysis result. The learning feedback includes, but is not limited to, the current performance, the knowledge points that need to be improved, and the recommended learning resources. For example, "Your mastery level of the recursive algorithm knowledge point is 40%. It is recommended to review the following resources: Recursive Visualization Tutorial and Recursive Problem Case Analysis". In addition, the system will propose specific learning path adjustment suggestions based on the learning disability score, such as reducing the difficulty of subsequent knowledge points or increasing supplementary resources.

[0146] Specifically, the learning effect analysis result can be displayed in real time through a visualization tool to help users and educators understand the learning status. For example: Knowledge graph view: The mastery level of knowledge points is displayed in the knowledge graph encoded by color and size. Green indicates high mastery, red indicates low mastery, and gray indicates not learned. Learning progress chart: Learning time, task completion status, and test scores are dynamically displayed through bar charts, line charts, or pie charts. Learning disability alert: When a knowledge point is identified as a learning disability, the system highlights it in red and attaches learning suggestions to help users adjust their learning strategies in a timely manner. These visualization tools provide dynamic interaction functions, such as clicking on a knowledge point to view detailed information, or selecting a time range to view historical learning progress.

[0147] Specifically, there is a multi-layer verification mechanism in the learning effect analysis module to verify the accuracy of the knowledge mastery quantification model through historical learning performance data, that is, the correlation between knowledge mastery and subsequent learning performance. Secondly, the effectiveness of the learning disability identification algorithm is verified through user feedback and actual learning effects. For example, if the user's performance improves after the system recommends reviewing a certain knowledge point, it proves that the algorithm has high practical value. The evaluation accuracy verification results are regularly used to optimize the system parameters and algorithm logic.

[0148] An optimization module is used to judge whether there are obstacles in the current node according to the learning effect analysis result of the current node. If it is judged that there are obstacles, several nodes are added or reduced in the first learning path according to the learning effect analysis result of the current node and the user profile to obtain an optimized learning path. If it is judged that there are no obstacles, the first learning path remains unchanged.

[0149] Specifically, as Figure 6 shown, in the optimization module, it includes: a node obstacle judgment unit, a learning effect judgment unit, a node addition and subtraction unit, and an optimized path generation unit.

[0150] The node obstacle judgment unit is used to judge whether there are obstacles in the current node according to the learning effect analysis result of the current node. If it is judged that there are no obstacles, the first learning path remains unchanged. If it is judged that there are obstacles, it enters the learning effect judgment unit;

[0151] Specifically, the learning disability score and knowledge mastery degree in the learning effect analysis result of the current node are compared with preset judgment thresholds. When there is a learning disability score or knowledge mastery degree less than the judgment threshold, the judgment result is that there is an obstacle; otherwise, there is no obstacle. For example, if the user gets low scores multiple times in the test and the learning time exceeds the recommended time, the system identifies it as a learning disability and initiates adjustment; for users with rapid learning progress and excellent performance, the system evaluates their abilities and prepares high-difficulty content.

[0152] A learning effect judgment unit, configured to judge the addition or deletion of nodes according to the learning effect analysis result of the current node;

[0153] Specifically, in the learning effect judgment unit, the learning effect analysis result of the current node includes the learning disability score and the knowledge mastery degree. When the learning disability score is less than the preset first performance threshold and the knowledge mastery degree is greater than the preset second performance threshold, several nodes are added; when the learning disability score is greater than or equal to the preset first performance threshold and the knowledge mastery degree is less than or equal to the preset second performance threshold, several nodes are deleted. The first performance threshold and the second performance threshold are set according to historical user learning data. If the user has difficulties with the knowledge points corresponding to a certain node, the system analyzes whether the prerequisite knowledge points are mastered sufficiently and inserts prerequisite knowledge points or reduces the learning difficulty when necessary; for example, in the case of difficulties in learning "recursive algorithms", the system can recommend learning the basic content of "function calls" and "loop structures". For users with excellent performance, the system skips redundant knowledge points and recommends higher-difficulty knowledge points, such as directly moving from "basic sorting algorithms" to "advanced sorting algorithms". The system adjusts the path priority through the knowledge graph and updates it in real time.

[0154] A node addition and deletion unit, configured to extract the mastery status knowledge graph from the user portrait. When the learning effect judgment unit judges to add nodes, the nodes with the edge attribute of prerequisite relationship with the current node are used as the nodes to be added; when it is judged to delete nodes, the nodes with the edge attribute of preceding relationship with the current node are used as the subsequent nodes, and the nodes to be deleted are obtained according to the subsequent nodes.

[0155] Specifically, the pre - relationship means that the current node needs to be learned first before learning the subsequent node. When obtaining the nodes to be subtracted based on the subsequent node, the following steps are included: Screen out several nodes in the mastered state knowledge graph whose edge attribute with the subsequent node is the prerequisite relationship as process nodes, calculate the association strength between each process node and the subsequent node respectively, compare each association strength with a preset strength threshold, and use the process nodes with association strength greater than the strength threshold as the nodes to be subtracted. For knowledge points with learning disabilities, the system increases their priority and inserts additional supplementary tasks to ensure timely resolution; for learners with excellent performance, skip the mastered knowledge points and optimize the path into a core knowledge point sequence. The system dynamically adjusts the path length, which not only avoids wasting time on long learning paths but also helps weak learners gradually break through difficulties.

[0156] The optimized path generation unit is used to, when it is determined to add a node, add the node to be added to the first learning path to obtain an optimized learning path; when it is determined to subtract a node, delete the node to be subtracted from the first learning path to obtain an optimized learning path.

[0157] Specifically, there are several nodes to be added and nodes to be subtracted. When a certain node to be added already exists on the first learning path, there is no need to add duplicate nodes.

[0158] Specifically, personalized learning content recommendations can also be made according to parameters such as the user's learning style in the user profile, including video tutorials, interactive courses, case analyses, and test exercises. The system selects the best resource type by analyzing the learning style (such as visual or hands - on) and dynamically optimizes the recommendation order using a deep - learning model. For example, visual learners are preferentially recommended visualization tutorials, while hands - on learners are recommended project practice tasks; if the current resource is ineffective or has poor results, the system switches to other content to improve learning efficiency. The system continuously optimizes and adjusts the algorithm through real - time learner data and feedback. The effects of each path adjustment (such as improved learning efficiency and better mastery of knowledge points) are stored in the database for training the reinforcement learning model to optimize the adjustment logic. For example, if a certain knowledge point often causes obstacles, the system introduces a more refined difficulty grading and task design to improve the adjustment accuracy and enhance the recommendation effect.

[0159] Specifically, it also includes a closed - loop feedback mechanism. The closed - loop feedback mechanism includes: continuously collecting the user's behavior data during the learning process, including learning time, knowledge point completion rate, test scores, resource usage, and specific records of each learning path adjustment. These data are uploaded and stored in the data warehouse in real time through the learning management system. The data is stored classified by knowledge points, learning stages, and user characteristics, providing structured input for subsequent analysis. For example, for the knowledge point of "recursive algorithm", the stored data may include a learning time of 2 hours, a test correct rate of 70%, and a resource usage frequency of 3 times, etc.

[0160] After each round of learning, a comprehensive evaluation of the learning effect is conducted based on the collected data. The evaluation content includes learning efficiency, knowledge point mastery, test improvement, and learning path execution. For example, calculate the overall improvement rate of the user after optimizing the learning path. If the test score of "recursive algorithm" increases from 60% to 85%, it is considered that the adjustment is effective. The system uses these evaluation results as feedback information, generates a detailed learning effect report and returns it to the user and the educator.

[0161] After the evaluation is completed, the system feeds back the learning effect, learning path adjustment records, and feature data to the learning path generation algorithm and the personalized recommendation model. This process is achieved through a closed-loop feedback mechanism:

[0162] The system optimizes the path generation logic according to the actual execution effect of the learning path. For example, if a certain type of user performs poorly after skipping basic knowledge points, the algorithm will reduce the priority of similar paths. The recommendation model feedback: The system optimizes the recommendation model by analyzing the usage of resource recommendations and the learning effect. For example, if the visualization tutorial of "recursive algorithm" obtains a high click-through rate and the learning effect improves, the model will be more inclined to recommend similar resources.

[0163] Specifically, the trained and optimized model is updated to the production environment through the system and applied in real time to new learning path generation and content recommendation tasks. For example, the optimized path generation algorithm gives priority to effective adjustment patterns in historical data when generating paths, while the recommendation model dynamically adjusts the sorting and content types of recommended resources. The system ensures the smoothness of model updates and avoids interfering with ongoing learning paths. Continuously monitor the performance of the optimized model in the closed-loop feedback, and evaluate the model effect through A / B testing or cross-validation. For example, allocate a part of users to the optimized path generation algorithm and keep the original algorithm for the other part, and compare the improvement of learning efficiency of the two groups of users. The verification metrics include the completion rate of the learning path, the mastery of knowledge points, and user satisfaction. If the optimization effect is achieved, the system solidifies the model optimization strategy into the next learning cycle. Through continuous optimization to achieve dynamic adaptability, it can provide accurate personalized learning paths and content according to the characteristics of different users. For example, for some users who need more basic knowledge points, the system will generate a more detailed learning path; while for users with fast learning progress, the path adjustment will be more simplified and highlight high-difficulty knowledge points. The system reflects the results of the optimized model into the new learning path generation and recommendation process to continuously improve the personalized learning experience. The system visually displays the results of closed-loop feedback and continuous optimization to users and educators through visualization tools. For example, generate a dynamic learning effect report, including the execution effect of the learning path, the usage of recommended resources, and the changing trend of knowledge point mastery. Users can view the specific improvement points before and after the adjustment, and educators can understand the improvement of the overall teaching quality by model optimization, providing a reference for further teaching strategy design.

[0164] Embodiment 2

[0165] As Figure 7 shown, the embodiment of the present invention also provides an online education learning path optimization method based on big data and intelligent analysis, including the following steps:

[0166] S10: Obtain basic user data;

[0167] S20: Generate a user profile according to the basic user data;

[0168] S30: Generate an initial path candidate set according to the user profile, screen the initial path candidate set, and obtain a first learning path, where the first learning path includes several nodes;

[0169] S40: Obtain the first learning data and historical learning performance data when the user learns according to the first learning path, and obtain the learning effect analysis result of the current node according to the historical learning performance data, the first learning data, and a preset learning obstacle evaluation model;

[0170] S50: Based on the analysis result of the learning effect of the current node, determine whether there are obstacles in the current node. If it is determined that there are obstacles, then add or subtract a number of nodes in the first learning path according to the analysis result of the learning effect of the current node and the user profile to obtain an optimized learning path. If it is determined that there are no obstacles, then keep the first learning path unchanged.

[0171] Specifically, in step S10, the following steps are included: S11: Obtain user initial data; S12: Add noise to the user initial data to obtain first basic data; S13: Delete and repair outliers in the first basic data to obtain user basic data.

[0172] Specifically, in step S20, the following steps are included: S21: Obtain the cognitive ability evaluation score of the user according to the user basic data and a preset cognitive ability evaluation model; S22: Use the multi-class support vector machine algorithm and the confidence calculation method for the user basic data to obtain the learning style classification result of the user; S23: Use the graph neural network algorithm for the user basic data to obtain the knowledge mastery status knowledge graph of the user; S24: Input the user basic data into a preset learning motivation analysis model to obtain the learning motivation evaluation result of the user; S25: Perform weighted calculation according to the cognitive ability evaluation score, the learning style classification result, the knowledge mastery status knowledge graph, and the learning motivation evaluation result, and generate a user profile according to the weighted calculation result.

[0173] Specifically, in step S21, the following steps are included: S211: Extract features from the user basic data to obtain a number of first feature vectors; S212: Use the multi-layer neural network and the long short-term memory network algorithm for the number of first feature vectors to obtain the learning efficiency, the depth of knowledge mastery, and the learning progress speed of the user; S213: Input the learning efficiency, the depth of knowledge mastery, and the learning progress speed into the preset cognitive ability evaluation model to obtain the cognitive ability evaluation score of the user.

[0174] Specifically, in step S22, the following steps are included: S221: Extract features from the user basic data to obtain a number of second feature vectors; S222: Use the multi-class support vector machine algorithm for the number of second feature vectors to obtain a number of classification results; S223: Calculate the confidence of each classification result using the softmax function according to the number of classification results; S224: According to the confidence of each category, select the classification result with the highest confidence as the dominant learning style of the user, generate personalized recommendations according to the dominant learning style of the user, and output the learning style classification result of the user according to the dominant learning style, the personalized recommendation, and the confidence corresponding to the dominant learning style.

[0175] Specifically, in step S23, the following steps are included: S231: Extract a set of knowledge points from the preset learning field according to the basic user data, and use the set of knowledge points as a number of basic nodes; S232: Obtain the annotation information of the preset learning field and the historical user learning data, generate explicit relationships between the basic nodes and basic weights of the basic nodes according to the annotation information, generate implicit relationships between the basic nodes according to the historical user learning data, and generate association relationships between the basic nodes according to the explicit relationships, the basic weights, and the implicit relationships; S233: Obtain the learning data of the user in the learning field, perform feature extraction on the learning data in the learning field to obtain a number of intensity feature vectors, and perform weighted calculation on the number of intensity feature vectors to obtain an association intensity; S234: Construct a structural knowledge graph according to the association intensity, a number of the basic nodes, and the association relationships between the basic nodes; S235: Use a graph embedding algorithm to map the structural knowledge graph to a low-dimensional vector space to obtain a number of embedding vectors; S236: Generate a mastery status knowledge graph according to the structural knowledge graph and a number of the embedding vectors using a force-directed algorithm.

[0176] Specifically, in step S24, the following steps are included: S241: Perform feature extraction on the basic user data to obtain a number of motivation feature vectors; S242: Use a clustering algorithm on the number of motivation feature vectors to obtain a number of motivation clustering results, and generate a number of learning motivation types according to the number of motivation clustering results; S243: Input the number of motivation feature vectors into a preset learning motivation analysis model to obtain the user's learning motivation score; S244: Obtain a learning motivation evaluation result according to the learning motivation score and the number of learning motivation types.

[0177] Specifically, in step S25, the following steps are included: S251: Use an ensemble learning algorithm to perform feature weighted fusion on the cognitive ability evaluation score, the learning style classification result, the mastery status knowledge graph, and the learning motivation evaluation result to obtain a number of comprehensive feature vectors; S252: Use a deep embedding learning algorithm on the number of comprehensive feature vectors to obtain a number of portrait embedding vectors; S253: Use a supervised learning algorithm to classify the number of portrait embedding vectors to obtain a number of label classification results, calculate the importance scores of each label classification result, and generate a number of labels according to the number of label classification results and the importance scores corresponding to each label classification result; S254: Generate an interactive radar chart from the number of label classification results; S255: Generate a user portrait according to the mastery status knowledge graph, the number of label classification results, the comprehensive score, and the interactive radar chart.

[0178] Specifically, in step S30, the following steps are included: S31: Generate an initial path candidate set according to the user profile; S32: Extract features from the initial path candidate set to obtain a number of candidate feature vectors corresponding to each initial path; S33: Obtain historical user data, and use a collaborative filtering algorithm to screen the initial path candidate set according to the number of candidate feature vectors, the user profile, and the historical user data to obtain a first initial path set; S34: Extract a number of knowledge point sequences from the first initial path set, and input the number of knowledge point sequences into a pre-trained recommendation model to obtain a number of recommendation scores, and select the first initial path corresponding to the highest recommendation score as the first learning path.

[0179] Specifically, in step S31, the following steps are included: S311: Extract a number of feature dimensions and user learning objectives according to the user profile; S312: Generate target knowledge points according to the user learning objectives, and use a recursive algorithm according to the target knowledge points and the number of feature dimensions to obtain a number of prerequisite knowledge points; S313: Generate a number of initial paths according to the number of prerequisite knowledge points and the target knowledge points, and establish an initial path candidate set according to the number of initial paths.

[0180] Specifically, in step S33, the following steps are included: S331: Screen the initial path candidate set according to a number of candidate feature vectors corresponding to each initial path and a preset screening threshold to obtain a first screened path set; S332: Obtain historical user data; S333: Calculate the similarity of each learning path in the first screened path set according to the historical user data and the user profile; S334: Screen a number of first screened paths with a similarity greater than the first threshold from the first screened path set as the first initial path set.

[0181] Specifically, in step S40, the following steps are included: S41: Obtain first learning data when the user learns according to the first learning path; S42: Obtain historical learning performance data; S43: Obtain an analysis result of the learning effect of the current learning node according to the historical learning performance data, the first learning data, and a preset learning obstacle evaluation model.

[0182] Specifically, in step S43, the following steps are included: S431: Calculate several performance deviations of the user according to the historical learning performance data and the first learning data; S432: Calculate the user's knowledge mastery degree according to the first learning data and the pre-trained mastery degree quantization model; S433: Process the first learning data by using a clustering algorithm to obtain several behavior clustering results. According to the several behavior clustering results, determine whether the user's behavior is abnormal. If it is determined that the user's behavior is abnormal, generate an analysis result of the learning effect of the current node according to the knowledge mastery degree, and skip the learning effect analysis subunit. If it is determined to be normal, proceed to step S434; S434: Input the several performance deviations into a preset learning disability evaluation model to obtain a learning disability score, and generate an analysis result of the learning effect of the current learning node according to the learning disability score and the knowledge mastery degree.

[0183] Specifically, in step S50, the following steps are included: S51: Determine whether there is an obstacle at the current node according to the analysis result of the learning effect of the current node. If it is determined that there is no obstacle, keep the first learning path unchanged. If it is determined that there is an obstacle, proceed to S52; S52: Determine the increase or decrease of the node according to the analysis result of the learning effect of the current node; S53: Extract the mastery status knowledge graph from the user portrait. When it is determined in S52 that a node is to be added, use the node whose edge attribute with the current node is a prerequisite relationship as the node to be added; when it is determined that a node is to be reduced, use the node whose edge attribute with the current node is a preposition relationship as the post-node, and obtain the node to be reduced according to the post-node; S54: When it is determined that a node is to be added, add the node to be added to the first learning path to obtain an optimized learning path; when it is determined that a node is to be reduced, delete the node to be reduced from the first learning path to obtain an optimized learning path.

[0184] Embodiment III

[0185] The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the online education learning path optimization method based on big data and intelligent analysis as described above.

[0186] Embodiment IV

[0187] The present invention also provides an electronic device. The electronic device according to the embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the online education learning path optimization method provided by the present invention. The following refers to Figure 8 , which shows a schematic structural diagram of a computer system 800 of an electronic device suitable for implementing the embodiment of the present invention. AsFigure 8 As shown, computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage section 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0188] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read from it is installed into the storage section 808 as needed.

[0189] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An online education learning path optimization system based on big data and intelligent analysis, characterized by: include: Data collection module, used to obtain basic user data; A user portrait generation module, used to generate a user portrait based on the user basic data; A path generation module, used to generate an initial path candidate set according to the user portrait, and screen the initial path candidate set to obtain a first learning path, wherein the first learning path includes a plurality of nodes; A learning effect analysis module, used to obtain first learning data and historical learning performance data of the user when learning according to the first learning path, and obtain a learning effect analysis result of the current node according to the historical learning performance data, the first learning data and a preset learning obstacle evaluation model; The optimization module is used to determine whether there is an obstacle at the current node based on the learning effect analysis result of the current node. If it is determined that there is an obstacle, a number of nodes are added or reduced in the first learning path according to the learning effect analysis result of the current node and the user portrait to obtain an optimized learning path. If it is determined that there is no obstacle, the first learning path is kept unchanged.

2. The online education learning path optimization system based on big data and intelligent analysis according to claim 1 is characterized in that: The data acquisition module specifically includes: A data collection unit, used to obtain initial user data from multiple collection channels; A protection unit, configured to add noise to the user's initial data by using a differential privacy algorithm to obtain first basic data; A preprocessing unit is used to use a deep learning algorithm to identify abnormal values ​​in the first basic data, and perform repair processing on the identified abnormal values ​​to obtain user basic data.

3. The online education learning path optimization system based on big data and intelligent analysis according to claim 1 is characterized in that: The user portrait generation module specifically includes: A cognitive ability evaluation unit, used to obtain a cognitive ability evaluation score of the user according to the user basic data and a preset cognitive ability evaluation model; A learning style classification unit, used to apply a multi-classification support vector machine algorithm and a confidence calculation method to the basic data of the user to obtain a learning style classification result of the user; A knowledge graph construction unit, used to obtain a user's mastery status knowledge graph using a graph neural network algorithm according to the user's basic data; A learning motivation analysis unit, used for inputting the basic data of the user into a learning motivation analysis model based on a multi-layer perceptron to obtain a learning motivation evaluation result of the user; A user portrait generating unit is used to perform weighted calculation according to the cognitive ability evaluation score, the learning style classification result, the mastery status knowledge graph and the learning motivation evaluation result, and generate a user portrait according to the weighted calculation result.

4. The online education learning path optimization system based on big data and intelligent analysis according to claim 1 is characterized in that: The path generation module specifically includes: An initial path candidate set construction unit, used to generate an initial path candidate set according to the user portrait; A candidate path feature extraction unit is used to extract features from the initial path candidate set to obtain a number of candidate feature vectors corresponding to each initial path; A screening unit, configured to obtain historical user data, and screen the initial path candidate set using a collaborative filtering algorithm according to the candidate feature vectors, the user portraits and the historical user data, to obtain a first initial path set; The initial path optimization unit is used to extract several knowledge point sequences from the first initial path set, and input several of the knowledge point sequences into a pre-trained recommendation model to obtain several recommendation scores, and select the first initial path corresponding to the highest recommendation score as the first learning path.

5. The online education learning path optimization system based on big data and intelligent analysis according to claim 1 is characterized in that: The learning effect analysis module specifically includes: A first learning data acquisition unit, used to acquire first learning data when the user learns according to the first learning path; A historical learning performance acquisition unit, used to acquire historical learning performance data; A learning effect analysis unit is used to obtain a learning effect analysis result of a current learning node based on the historical learning performance data, the first learning data and a preset learning obstacle evaluation model.

6. The online education learning path optimization system based on big data and intelligent analysis according to claim 1 is characterized in that: The optimization module specifically includes: a node obstacle judgment unit, configured to judge whether there is an obstacle at the current node according to the learning effect analysis result of the current node, and if it is judged that there is no obstacle, keep the first learning path unchanged; if it is judged that there is an obstacle, enter the learning effect judgment unit; A learning effect judgment unit, used to judge whether to increase or decrease a node according to the learning effect analysis result of the current node; A node adding and reducing unit is used to extract the mastering state knowledge graph from the user portrait, and when the learning effect judging unit judges to add a node, a node with a pre-repair relationship with the edge attribute of the current node is used as a node to be added; when the learning effect judging unit judges to reduce a node, a node with a pre-repair relationship with the edge attribute of the current node is used as a post-node, and a node to be reduced is obtained according to the post-node; The optimized path generation unit is used to add the node to be added to the first learning path to obtain the optimized learning path when it is determined that a node is to be added; and to delete the node to be removed from the first learning path to obtain the optimized learning path when it is determined that a node is to be reduced.

7. The online education learning path optimization system based on big data and intelligent analysis according to claim 3 is characterized in that: The cognitive ability evaluation unit specifically includes: A first feature extraction subunit, configured to extract features from the user basic data to obtain a plurality of first feature vectors; A cognitive ability analysis subunit, used to apply a multi-layer neural network and a long short-term memory network algorithm to the first feature vectors to obtain the user's learning efficiency, knowledge mastery depth and learning progress speed; The cognitive ability evaluation subunit is used to input the learning efficiency, the knowledge mastering depth and the learning progress speed into the preset cognitive ability evaluation model to obtain the user's cognitive ability evaluation score.

8. The online education learning path optimization system based on big data and intelligent analysis according to claim 3 is characterized in that: The learning style classification unit specifically includes: A second feature extraction subunit, configured to extract features from the user basic data to obtain a plurality of second feature vectors; A classification subunit, used for applying a multi-classification support vector machine algorithm to a plurality of said second feature vectors to obtain a plurality of classification results; A confidence calculation subunit, used to calculate the confidence of each classification result using a normalized exponential function according to the classification results; A learning style output subunit is used to select the classification result with the highest confidence as the user's dominant learning style according to the confidence of each classification result, generate personalized recommendations according to the user's dominant learning style, and output the user's learning style classification result according to the dominant learning style, the personalized recommendations and the confidence corresponding to the dominant learning style.

9. The online education learning path optimization system based on big data and intelligent analysis according to claim 5 is characterized in that: The optimization path generation unit is specifically used for: The learning effect analysis unit specifically includes: a performance analysis subunit, configured to calculate a number of performance deviations of the user based on the historical learning performance data and the first learning data; A mastery quantification subunit, configured to calculate the user's knowledge mastery according to the first learning data and a preset mastery quantification model; A behavior pattern analysis subunit is used to process the first learning data using a clustering algorithm to obtain a number of behavior clustering results, and judge whether the user behavior is abnormal based on the behavior clustering results. If the user behavior is judged to be abnormal, a learning effect analysis result of the current node is generated based on the knowledge mastery, and the learning effect analysis subunit is skipped. If it is judged to be normal, the learning effect analysis subunit is entered; The learning effect analysis subunit is used to input the performance deviations into a preset learning disability evaluation model to obtain a learning disability score, and generate a learning effect analysis result of the current learning node according to the learning disability score and the knowledge mastery.

10. The online education learning path optimization system based on big data and intelligent analysis according to claim 6 is characterized in that: The optimization path generation unit is specifically used for: In the mastering state knowledge graph, several nodes having an edge attribute of a prior relationship with a subsequent node are selected as process nodes; The association strength between each process node and the post-positioned node is calculated, each association strength is compared with a preset strength threshold, and the process nodes whose association strength is greater than the strength threshold are used as nodes to be reduced.

Images