AI english teaching feedback optimization method and system based on learning behavior graph

Abstract

The application provides an AI English teaching feedback optimization method and system based on a learning behavior graph, and belongs to the technical field of intelligent education and artificial intelligence. The method comprises the following steps: multi-modal data acquisition; learning behavior graph construction, a method of fusing an improved adaptive gated cognitive diagnosis and a space-time graph neural network is adopted to realize dynamic modeling of the cognitive state of learners and the dependency relationship of knowledge points; English teaching feedback optimization, a graph perception state space and an action space are constructed based on the learning behavior graph, an improved dynamic reward function is constructed, and a personalized feedback strategy is generated in combination with a reinforcement learning model; adaptive learning content recommendation, a multi-constraint filtering and multi-factor weighted sorting method is adopted to obtain the optimal matching of recommended content and learning goals; teaching effect analysis; the application can realize integrated optimization of deep modeling and feedback recommendation of learning behavior, and improve the personalization, dynamics and continuity of English teaching feedback.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and education technology, specifically to an AI-based English teaching feedback optimization method and system based on learning behavior graphs. Background Technology

[0002] The AI-based English teaching feedback optimization method and system based on learning behavior mapping is a method that utilizes artificial intelligence technology to model and analyze learners' multimodal behavioral data (such as homework answers, learning time, click operations, and grade changes) during the English learning process. Its core is to construct a learning behavior mapping system that dynamically represents the relationship between learners and knowledge points, learning paths, and cognitive states. Based on this, feedback strategies are optimized and personalized content recommendations are provided. The method helps teachers and learning systems more accurately identify learners' weaknesses, generate teaching feedback and learning content tailored to individual needs in real time, thereby improving learning efficiency and teaching effectiveness, and achieving the teaching goals of individualized instruction and continuous optimization.

[0003] However, existing AI-based English teaching feedback optimization methods and systems based on learning behavior graphs suffer from technical problems such as limited data collection dimensions, isolated feedback optimization processes, and a lack of constrained correlation in recommendation mechanisms. Existing learning behavior graph construction methods also exhibit issues such as insufficient node feature updates, inadequate utilization of temporal information, and low accuracy of cognitive diagnostic models. Furthermore, existing English teaching feedback optimization methods and systems suffer from problems such as one-sided reward function design, difficulty in long-term optimization of feedback strategies, and insufficient adaptation to individual differences. Finally, existing adaptive learning content recommendation methods suffer from technical problems such as a disconnect between recommendation results and learning objectives, a simplistic ranking mechanism, and difficulty in balancing personalization with teaching principles. Summary of the Invention

[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an AI-based English teaching feedback optimization method and system based on learning behavior graphs. Addressing the technical problems of existing AI-based English teaching feedback optimization methods and systems based on learning behavior graphs, such as single data collection dimensions, isolated feedback optimization stages, and a lack of constrained correlation in the recommendation mechanism, this solution creatively adopts a comprehensive teaching feedback optimization process that combines improved learning behavior graph construction, feedback optimization reinforcement learning prediction, and adaptive learning content recommendation. This achieves a closed-loop modeling of the entire link between learner's individual behavior, cognitive state, and teaching resources, making feedback and recommendations more personalized and dynamically adaptable. Furthermore, addressing the technical problems of insufficient node feature updates, inadequate utilization of temporal information, and low accuracy of cognitive diagnostic models in existing learning behavior graph construction methods, this solution creatively adopts a learning behavior graph construction method that integrates improved adaptive gating cognitive diagnosis and spatiotemporal graph neural networks. This method dynamically updates learners' cognitive states and fully models the dependencies between knowledge points, achieving higher accuracy and stronger temporal correlation in the graph construction of learning behaviors, significantly improving the learning behavior graph's ability to... This solution addresses the technical issues in existing English teaching feedback optimization methods and systems, such as one-sided reward function design, difficulty in long-term optimization of feedback strategies, and insufficient adaptation to individual differences. It creatively employs an improved reward-enhanced feedback optimization method and system incorporating multimodal graph perception to optimize English teaching feedback. By introducing multi-dimensional factors such as learning outcome rewards, cognitive state rewards, behavioral habit rewards, and curriculum planning rewards, it achieves a balanced optimization of feedback strategies between immediacy and long-term sustainability, and between individualization and universality, effectively improving the accuracy and sustainability of English teaching feedback. Furthermore, it addresses the technical issues in existing adaptive learning content recommendation methods, such as a disconnect between recommendation results and learning objectives, a single ranking mechanism, and difficulty in balancing individualization and teaching principles. This solution creatively adopts a learning content recommendation method based on multi-constraint filtering for adaptive learning content recommendation. It introduces multiple constraints at the candidate content level, including knowledge point coverage, difficulty level, modality, and intensity, and achieves final recommendation through multi-factor weighted ranking. This achieves close coupling between recommended content and feedback strategies and learning objectives, improving the relevance, rationality, and diversity of the recommendation results.

[0005] The technical solution adopted by this invention is as follows: The AI ​​English teaching feedback optimization method and system based on learning behavior graph provided by this invention includes the following steps:

[0006] Step S1: Multimodal data acquisition;

[0007] Step S2: Construction of the learning behavior map;

[0008] Step S3: Optimize English teaching feedback;

[0009] Step S4: Adaptive learning content recommendation;

[0010] Step S5: Analysis of teaching effectiveness.

[0011] Furthermore, in step S1, the multimodal data acquisition is used to obtain multi-source multimodal data of learners in the process of learning English, specifically by obtaining the original dataset of learning behavior through multi-type data acquisition and data preprocessing;

[0012] The original dataset of learning behaviors includes text data, interaction log data, and performance data.

[0013] Further, in step S2, the construction of the learning behavior graph is used to transform the original data into a structured learning behavior graph. Specifically, based on the original learning behavior dataset, a learning behavior graph construction method that integrates improved adaptive gating cognitive diagnosis and spatiotemporal graph neural networks is used to construct the learning behavior graph, resulting in the learning behavior graph. This includes the following steps:

[0014] Step S21: Initialize the heterogeneous learning graph, specifically by constructing the node set, edge set, node attributes, and time series respectively to obtain the initial heterogeneous learning graph;

[0015] Step S22: Construct an improved adaptive gating unit, specifically by constructing the cognitive state vector corresponding to the student node based on the initial attributes of the nodes in the initial heterogeneous learning graph, and constructing an improved adaptive gating unit through an adaptive cognitive influence gating mechanism, updating the cognitive state vector, and obtaining updated cognitive influence vector data.

[0016] Step S23: Dynamic cognitive diagnosis, specifically, involves concatenating the cognitive state vector from the previous time, the features of the current interaction log data, and the updated cognitive impact vector data, and then calculating the dynamic cognitive diagnosis update data by constructing a dynamic cognitive diagnosis function.

[0017] Step S24: Construct an edge-aware improved spatiotemporal graph network. Specifically, the initial heterogeneous learning graph is updated based on the dynamic cognitive diagnosis update data. The edge weight coefficients are calculated by jointly modeling the updated node and edge features. Based on the edge weight coefficients, the edge-aware improved spatiotemporal graph network is constructed.

[0018] Step S25: Learning behavior embedding, specifically, based on the dynamic cognitive diagnosis update data and the spatiotemporal graph convolutional features output by the edge-aware improved spatiotemporal graph network, and through feature fusion, a learning behavior embedding vector is obtained;

[0019] Step S26: Construction of the learning behavior graph. Specifically, based on the learning behavior embedding vector, the learning behavior embedding vector is used as a new node representation. At the same time, the weights of the interaction edges and prior relation edges are adjusted to update the heterogeneous graph, thereby obtaining the learning behavior graph, which is then stored in a graph database.

[0020] Further, in step S3, the English teaching feedback optimization is used to generate personalized feedback strategies. Specifically, based on the learning behavior map, an improved reward reinforcement feedback optimization method and system combining multimodal map perception is used to optimize English teaching feedback and obtain a teaching feedback optimization strategy, including the following steps:

[0021] Step S31: Construction of the graph perception state space, specifically, based on the learned behavior graph, the state space is constructed to obtain the state space parameters;

[0022] Step S32: Motion space construction, specifically, by constructing composite motion parameters, the motion space is constructed to obtain motion space parameters;

[0023] Step S33: Improve the construction of the dynamic reward function. Specifically, by integrating immediate learning outcome rewards, cognitive state rewards, learning behavior rewards, and course planning feedback rewards, the dynamic reward function is improved and the improved reward function is obtained.

[0024] Step S34: Reinforcement learning training, specifically, through the construction of the graph perception state space, the construction of the action space and the construction of the improved dynamic reward function, reinforcement learning training is carried out through standard reinforcement learning proximal policy optimization training to obtain an English teaching feedback optimization model;

[0025] Step S35: English teaching feedback optimization, specifically, using the English teaching feedback optimization model to optimize English teaching feedback and obtain teaching feedback optimization strategies.

[0026] Further, in step S4, the adaptive learning content recommendation is used to intelligently recommend suitable learning content. Specifically, based on the learning behavior graph, an adaptive learning content recommendation method based on multi-constraint filtering is used to obtain learning content recommendation data, including the following steps:

[0027] Step S41: Strategy instruction organization, specifically, transforming the teaching feedback optimization strategy into recommendation constraint instructions to obtain a recommendation instruction vector;

[0028] Step S42: Multi-constraint filtering of learning content. Specifically, multi-constraint filtering of learning content is performed based on the recommendation instruction vector, and the initial filtered learning content data is obtained by setting constraints.

[0029] Step S43: Multi-factor weighted ranking, specifically by constructing a weighted ranking function for candidate learning content, calculating the comprehensive score of each candidate learning content, and obtaining the comprehensive score data for learning content recommendation;

[0030] Step S44: Adaptive learning content recommendation, specifically, based on the comprehensive score data of the learning content recommendation, the learning content is scored and ranked to obtain the learning content recommendation data.

[0031] Furthermore, in step S5, the teaching effectiveness analysis is used to evaluate the feedback and recommendation results. Specifically, based on the teaching feedback optimization strategy and the learning content recommendation data, reference data for teaching and learning evaluation analysis is obtained through data visualization analysis and manual evaluation.

[0032] The AI ​​English teaching feedback optimization system based on learning behavior graph provided by this invention includes a data acquisition and perception module, a learning behavior graph module, a feedback optimization module, a content recommendation module, and an evaluation and analysis module.

[0033] The acquisition and perception module is used for multimodal data acquisition. Through multimodal data acquisition, the original dataset of learning behavior is obtained, and the original dataset of learning behavior is sent to the learning behavior graph module.

[0034] The learning behavior graph module is used to construct a learning behavior graph. Through the construction of the learning behavior graph, a learning behavior graph is obtained, and the learning behavior graph is sent to the feedback optimization module and the content recommendation module.

[0035] The feedback optimization module is used for optimizing English teaching feedback. Through the optimization of English teaching feedback, a teaching feedback optimization strategy is obtained, and the teaching feedback optimization strategy is sent to the evaluation and analysis module.

[0036] The content recommendation module is used for adaptive learning content recommendation. Through adaptive learning content recommendation, learning content recommendation data is obtained and the learning content recommendation data is sent to the evaluation and analysis module.

[0037] The evaluation and analysis module is used for teaching effectiveness analysis, and through this analysis, reference data for teaching and learning evaluation and analysis is obtained.

[0038] The beneficial effects achieved by adopting the above solution are as follows:

[0039] (1) In view of the technical problems in the existing AI English teaching feedback optimization methods and systems based on learning behavior graphs, such as single data collection dimension, isolated feedback optimization links, and lack of constraint association in the recommendation mechanism, this solution creatively adopts a comprehensive teaching feedback optimization process that combines learning behavior graph improvement construction, feedback optimization reinforcement learning prediction, and adaptive learning content recommendation. This realizes the full-link closed-loop modeling between learner individual behavior, cognitive state and teaching resources, making feedback and recommendation more personalized and dynamically adapted.

[0040] (2) In view of the technical problems in the existing learning behavior graph construction methods, such as insufficient node feature updates, insufficient utilization of temporal information, and low accuracy of cognitive diagnosis models, this solution creatively adopts a learning behavior graph construction method that integrates improved adaptive gating cognitive diagnosis and spatiotemporal graph neural network to construct the learning behavior graph. This method can dynamically update the learner's cognitive state and fully model the dependencies between knowledge points, thereby achieving higher accuracy and stronger temporal correlation in the graph construction of learning behavior. This significantly improves the ability of the learning behavior graph to represent individual differences and knowledge structure.

[0041] (3) In view of the technical problems in existing English teaching feedback optimization methods and systems, such as one-sided reward function design, difficulty in long-term optimization of feedback strategies, and insufficient adaptation to individual differences, this solution creatively adopts an improved reward reinforcement feedback optimization method and system that combines multimodal graph perception to optimize English teaching feedback. By introducing multi-dimensional factors such as learning outcome rewards, cognitive state rewards, behavioral habit rewards and curriculum planning rewards, the solution achieves a balanced optimization of feedback strategies between immediacy and long-term, individualization and universality, effectively improving the accuracy and sustainability of English teaching feedback.

[0042] (4) In view of the technical problems in existing adaptive learning content recommendation methods, such as the disconnect between recommendation results and learning objectives, the single ranking mechanism, and the difficulty in taking into account personalization and teaching rules, this solution creatively adopts a learning content recommendation method based on multi-constraint filtering to recommend adaptive learning content. At the candidate content level, multiple constraints such as knowledge point coverage, difficulty level, modality and intensity are introduced, and the final recommendation is achieved through multi-factor weighted ranking. This realizes the close coupling between recommended content and feedback strategy and learning objectives, and improves the relevance, rationality and diversity of recommendation results. Attached Figure Description

[0043] Figure 1 A flowchart illustrating the AI-based English teaching feedback optimization method based on learning behavior graphs provided by this invention;

[0044] Figure 2 This is a flowchart illustrating the process of multimodal data acquisition in step S1.

[0045] Figure 3 A flowchart illustrating the process of constructing the behavior map for step S2;

[0046] Figure 4 A flowchart illustrating the optimization of English teaching feedback in step S3;

[0047] Figure 5 This is a flowchart illustrating the adaptive learning content recommendation process in step S4.

[0048] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0049] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0050] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0051] Example 1, see Figure 1 The present invention provides an AI-based English teaching feedback optimization method and system based on learning behavior graphs. The method includes the following steps:

[0052] Step S1: Multimodal data acquisition;

[0053] Step S2: Construction of the learning behavior map;

[0054] Step S3: Optimize English teaching feedback;

[0055] Step S4: Adaptive learning content recommendation;

[0056] Step S5: Analysis of teaching effectiveness.

[0057] By performing the above operations, this solution addresses the technical problems in existing AI English teaching feedback optimization methods and systems based on learning behavior graphs, such as single data collection dimensions, isolated feedback optimization links, and a lack of constraint correlation in the recommendation mechanism. This solution creatively adopts a comprehensive teaching feedback optimization process that combines improved construction of learning behavior graphs, feedback optimization to enhance learning prediction, and adaptive learning content recommendation. This achieves a closed-loop modeling of the entire link between learner's individual behavior, cognitive state, and teaching resources, making feedback and recommendation more personalized and dynamically adaptable.

[0058] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the multimodal data acquisition is used to obtain multi-source multimodal data of learners in the process of learning English. Specifically, it is to obtain the original dataset of learning behavior through multi-type data acquisition and data preprocessing.

[0059] The original dataset of learning behavior includes text data, interaction log data, and performance data;

[0060] The multi-type data acquisition and data preprocessing specifically include the following steps:

[0061] Step S11: Multi-type data collection, used to acquire multi-source data of learners in the English learning process, specifically including: text data collection, used to acquire learners' input homework answers, wrong answer records and correction comments; interaction log data collection, used to acquire interaction event data such as learning time, click behavior and answering speed; performance data collection, used to acquire accuracy rate, score and historical performance trend; multi-source timestamp unification, used to convert time information from different data sources into a unified timestamp format;

[0062] Step S12: Data cleaning, used to eliminate abnormal and incomplete data, specifically including: missing value handling, used to impute missing score records or mark missing log items as null; outlier removal, used to remove unreasonable answer times and duplicate log records; format standardization, used to unify text data into standard encoding, unify scores into a percentage system, and unify time information into a standard time zone.

[0063] Step S13: Data standardization and feature generation, used to enhance the computability of data, specifically including: numerical data standardization, used to normalize continuous data such as learning time and answering speed; categorical data encoding, used to classify and encode question types, error types, etc.; and text data preprocessing, used to remove noisy characters and perform word segmentation, part-of-speech tagging, and error feature extraction.

[0064] Step S14: Data fusion and preliminary storage, used to form a unified original dataset of learning behaviors, specifically including: multimodal data fusion, used to align text data, log data and performance data based on learner identifiers and timestamps; preliminary storage, used to store the fused data in a database or big data platform to generate reusable intermediate data tables;

[0065] Preferably, Table 1 is a specific field example table of the original dataset of the learning behavior. As shown in the table, the text data specifically includes the original data of input homework answers and the data of incorrect questions; the interaction log data specifically includes the data of learning duration, click behavior and answering speed; the performance data specifically includes the accuracy rate data and the score trend data.

[0066] Table 1. Examples of specific fields in the original dataset of learning behaviors

[0067]

[0068] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the learning behavior graph construction is used to transform the original data into a structured learning behavior graph. Specifically, based on the original learning behavior dataset, a learning behavior graph construction method that integrates improved adaptive gating cognitive diagnosis and spatiotemporal graph neural networks is used to construct the learning behavior graph, resulting in a learning behavior graph. This includes the following steps:

[0069] Step S21: Initialize the heterogeneous learning graph, specifically by constructing the node set, edge set, node attributes, and time series respectively to obtain the initial heterogeneous learning graph;

[0070] Preferably, the calculation formula for the initial heterogeneous learning graph is:

[0071] G0=(V,E,A,T);

[0072] In the formula, G0 is the initial heterogeneous learning graph, V is the node set, including student nodes and knowledge component nodes, E is the edge set, including interaction edges and prior relation edges, A is the node attribute, including student node attributes and knowledge component node attributes, and T is the time series, used to represent the time state of the graph.

[0073] In the initial heterogeneous learning graph, the student node is used to represent an individual learner, and the knowledge component node is used to represent a micro-knowledge point.

[0074] The interaction edge is used to connect learners and knowledge points, and carries an interaction feature vector. The interaction feature vector is specifically obtained by aggregating the data of each student node in the original dataset of the learning behavior. The prior relation edge is used to connect different knowledge component nodes, and the edge weight is automatically generated through knowledge graph embedding learning to reflect the dependency relationship between knowledge points.

[0075] The student node attributes are initialized based on the learner's historical average performance, and the knowledge component node attributes are fixed as the difficulty and importance meta-features of the knowledge points.

[0076] Step S22: Construct an improved adaptive gating unit, specifically by constructing the cognitive state vector corresponding to the student node based on the initial attributes of the nodes in the initial heterogeneous learning graph, and constructing an improved adaptive gating unit through an adaptive cognitive influence gating mechanism, updating the cognitive state vector, and obtaining updated cognitive influence vector data.

[0077] The initial calculation formula for the cognitive state vector is:

[0078] ;

[0079] In the formula, It is the cognitive state vector corresponding to the initial time. It is a sigmoid activation function, where W0 is the initial cognitive state weight. This is the initial attribute of the student node, a k It is a node attribute of the knowledge component. It is a vector concatenation operator;

[0080] The calculation formula for the adaptive cognitive influence gating mechanism is as follows:

[0081] ;

[0082] In the formula, It updates the cognitive influence vector data, W g It's about updating the weights. f is the cognitive state vector at time t-1. e These are the feature data corresponding to the interaction log data in the original dataset of learning behavior, where tanh is the hyperbolic tangent activation function, and W... d It is the weight of historical data. It is the element-wise multiplication operator;

[0083] Step S23: Dynamic cognitive diagnosis, specifically, involves concatenating the cognitive state vector from the previous time, the features of the current interaction log data, and the updated cognitive impact vector data, and then calculating the dynamic cognitive diagnosis update data by constructing a dynamic cognitive diagnosis function.

[0084] The formula for calculating the dynamic cognitive diagnostic update data is as follows:

[0085] ;

[0086] In the formula, z represents the dynamic cognitive diagnosis update data, and W... z It is a dynamic cognitive diagnostic weight;

[0087] Step S24: Construct an edge-aware improved spatiotemporal graph network. Specifically, the initial heterogeneous learning graph is updated based on the dynamic cognitive diagnosis update data. The edge weight coefficients are calculated by jointly modeling the updated node and edge features. Based on the edge weight coefficients, the edge-aware improved spatiotemporal graph network is constructed.

[0088] The formula for calculating the edge weight coefficient is as follows:

[0089] ;

[0090] In the formula, It is the edge weight coefficient. It is the transpose of the attention vector, LeakyReLU is a non-linear activation function, and W is the attention weight. It is the representation vector of node i, used to represent the dynamic cognitive diagnosis update data. It is the representation vector of neighboring node j. It is the feature vector of the edge between node i and its neighboring node j. L is the normalized set of neighbor nodes, and L is the normalized neighbor node index. It is the normalized representation vector of neighboring node L. It is the edge feature vector between node i and its normalized neighbor node L;

[0091] Step S25: Learning behavior embedding, specifically, is to obtain the learning behavior embedding vector by using the dynamic cognitive diagnosis update data and the spatiotemporal graph convolutional features output by the edge-aware improved spatiotemporal graph network and through feature fusion.

[0092] Step S26: Construction of the learning behavior graph. Specifically, based on the learning behavior embedding vector, the learning behavior embedding vector is used as a new node representation. At the same time, the weights of the interaction edges and prior relation edges are adjusted to update the heterogeneous graph, thereby obtaining the learning behavior graph, which is then stored in a graph database.

[0093] By performing the above operations, this solution addresses the technical problems of insufficient node feature updates, inadequate utilization of temporal information, and low accuracy of cognitive diagnostic models in existing learning behavior graph construction methods. It creatively adopts a learning behavior graph construction method that integrates improved adaptive gated cognitive diagnosis and spatiotemporal graph neural networks. This method dynamically updates learners' cognitive states and fully models the dependencies between knowledge points, achieving higher accuracy and stronger temporal correlation in the graph construction of learning behaviors. This significantly improves the learning behavior graph's ability to represent individual differences and knowledge structures.

[0094] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the English teaching feedback optimization is used to generate personalized feedback strategies. Specifically, based on the learning behavior map, an improved reward reinforcement feedback optimization method and system combining multimodal map perception is used to optimize English teaching feedback and obtain a teaching feedback optimization strategy, including the following steps:

[0095] Step S31: Construction of the graph perception state space, specifically, based on the learned behavior graph, the state space is constructed to obtain the state space parameters;

[0096] The formula for calculating the state space parameters is as follows:

[0097] ;

[0098] In the formula, s t These are state-space parameters. It is the student node embedding vector. It is a cognitive state vector. It is a learning behavior map G t The global feature vector obtained by performing a read operation;

[0099] Step S32: Motion space construction, specifically, by constructing composite motion parameters, the motion space is constructed to obtain motion space parameters;

[0100] The formula for calculating the motion space parameters is as follows:

[0101] ;

[0102] In the formula, a t These are motion space parameters. These are content actions, used to represent specific strategies for providing instructional feedback, namely, selecting one or more knowledge components. These are modal actions, used to represent the specific presentation format of instructional feedback. It refers to intensity actions, used to indicate the intensity of teaching feedback;

[0103] Step S33: Improve the construction of the dynamic reward function. Specifically, by integrating immediate learning outcome rewards, cognitive state rewards, learning behavior rewards, and course planning feedback rewards, the dynamic reward function is improved and the improved reward function is obtained.

[0104] The formula for calculating the improved reward function is as follows:

[0105] ;

[0106] In the formula, It is an improvement on the reward function, R performance It integrates instant learning achievement rewards. It is the cognitive state reward weight, R cognitive It is a cognitive state reward. It is the reward weight for learning behavior, R behavior It is a reward for learning behavior. It is the weighting of course planning feedback rewards, R curriculum It is a reward for feedback on course planning;

[0107] Preferably, the calculation formula for the integrated instant learning achievement reward is as follows:

[0108] ;

[0109] In the formula, Score t+1 It is the student's specific score at time t+1. t It represents the student's score at time t;

[0110] The formula for calculating the cognitive state reward is as follows:

[0111] ;

[0112] In the formula, It is the set of knowledge points targeted by the feedback action goal, where k is the knowledge point index. It's the weight of knowledge points. It is the state cognition vector corresponding to the k-th knowledge point at time t+1. It is the state cognition vector corresponding to the k-th knowledge point at time t;

[0113] The formula for calculating the reward for the learning behavior is as follows:

[0114] ;

[0115] In the formula, f is the learning behavior quantification function, which is specifically determined based on the learning duration parameter. And answering speed parameters Calculations are performed to determine the degree of change in learning behavior and to reward that behavior.

[0116] The course planning feedback reward is used to reward course navigation decisions that conform to teaching principles;

[0117] Step S34: Reinforcement learning training, specifically, through the construction of the graph perception state space, the construction of the action space and the construction of the improved dynamic reward function, reinforcement learning training is carried out through standard reinforcement learning proximal policy optimization training to obtain an English teaching feedback optimization model;

[0118] Step S35: English teaching feedback optimization, specifically, using the English teaching feedback optimization model to optimize English teaching feedback and obtain teaching feedback optimization strategies.

[0119] By performing the above operations, this solution addresses the technical problems in existing English teaching feedback optimization methods and systems, such as one-sided reward function design, difficulty in long-term optimization of feedback strategies, and insufficient adaptation to individual differences. It creatively adopts an improved reward-reinforcement feedback optimization method and system that combines multimodal graph perception to optimize English teaching feedback. By introducing multi-dimensional factors such as learning outcome rewards, cognitive state rewards, behavioral habit rewards, and curriculum planning rewards, it achieves a balanced optimization of feedback strategies between immediacy and long-term sustainability, and between individualization and universality, effectively improving the accuracy and sustainability of English teaching feedback.

[0120] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the adaptive learning content recommendation is used to intelligently recommend suitable learning content. Specifically, based on the learning behavior graph, an adaptive learning content recommendation method based on multi-constraint filtering is used to perform adaptive learning content recommendation to obtain learning content recommendation data, including the following steps:

[0121] Step S41: Strategy instruction organization, specifically, transforming the teaching feedback optimization strategy into recommendation constraint instructions to obtain a recommendation instruction vector;

[0122] The formula for calculating the recommended instruction vector is:

[0123] ;

[0124] In the formula, A t It is a recommended instruction vector. It is the set of knowledge points targeted by the feedback action goal, M type It is the modality of recommended content, D level It refers to the difficulty level of the learning content, I intensity This is the recommended strength;

[0125] Step S42: Multi-constraint filtering of learning content. Specifically, multi-constraint filtering of learning content is performed based on the recommendation instruction vector, and the initial filtered learning content data is obtained by setting constraints.

[0126] The constraints specifically include constraints on covering the set of target knowledge points, constraints on the difficulty level being consistent with or within the allowable range, and constraints on the consistency of the presentation modality and intensity of the recommended content.

[0127] Step S43: Multi-factor weighted ranking, specifically by constructing a weighted ranking function for candidate learning content, calculating the comprehensive score of each candidate learning content, and obtaining the comprehensive score data for learning content recommendation;

[0128] The formula for calculating the weighted ranking function of the candidate learning content is:

[0129] ;

[0130] In the formula, Score(r) i ) represents the comprehensive score data for learning content recommendations, w1 represents the knowledge relevance weight, and S rel (r i ) is the candidate learning content r i The relevance score to the target knowledge point, w2 is the content quality weight, and S qual (r i ) is the content quality score, w3 is the preference matching weight, and S pref (r i ) is the preference matching score, w4 is the content novelty weight, S novel (r i W5 is the score for content novelty, W5 is the weighting of course sequence, and S... seq (r i (This refers to the course sequence grading;)

[0131] Step S44: Adaptive learning content recommendation, specifically, based on the comprehensive score data of the learning content recommendation, the learning content is scored and ranked to obtain the learning content recommendation data.

[0132] By performing the above operations, this solution addresses the technical problems of existing adaptive learning content recommendation methods, such as the disconnect between recommendation results and learning objectives, a single ranking mechanism, and difficulty in balancing personalization and teaching principles. It creatively adopts a learning content recommendation method based on multi-constraint filtering for adaptive learning content recommendation. This method introduces multiple constraints at the candidate content level, including knowledge point coverage, difficulty level, modality, and intensity, and achieves the final recommendation through multi-factor weighted ranking. This results in a close coupling between recommended content, feedback strategies, and learning objectives, improving the relevance, rationality, and diversity of the recommendation results.

[0133] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the teaching effect analysis is used to evaluate the feedback and recommendation results. Specifically, based on the teaching feedback optimization strategy and the learning content recommendation data, teaching and learning evaluation analysis reference data is obtained through data visualization analysis and manual evaluation.

[0134] Example 7, see Figure 1 This embodiment is based on the above embodiments. The AI ​​English teaching feedback optimization system based on learning behavior graph provided by the present invention includes a data acquisition and perception module, a learning behavior graph module, a feedback optimization module, a content recommendation module, and an evaluation and analysis module.

[0135] The acquisition and perception module is used for multimodal data acquisition. Through multimodal data acquisition, the original dataset of learning behavior is obtained, and the original dataset of learning behavior is sent to the learning behavior graph module.

[0136] The learning behavior graph module is used to construct a learning behavior graph. Through the construction of the learning behavior graph, a learning behavior graph is obtained, and the learning behavior graph is sent to the feedback optimization module and the content recommendation module.

[0137] The feedback optimization module is used for optimizing English teaching feedback. Through the optimization of English teaching feedback, a teaching feedback optimization strategy is obtained, and the teaching feedback optimization strategy is sent to the evaluation and analysis module.

[0138] The content recommendation module is used for adaptive learning content recommendation. Through adaptive learning content recommendation, learning content recommendation data is obtained and the learning content recommendation data is sent to the evaluation and analysis module.

[0139] The evaluation and analysis module is used for teaching effectiveness analysis, and through this analysis, reference data for teaching and learning evaluation and analysis is obtained.

[0140] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0141] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0142] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An AI-based English teaching feedback optimization method based on learning behavior graphs, characterized by: The method includes the following steps: Step S1: Multimodal data acquisition. Through multi-type data acquisition and data preprocessing, the original dataset of learning behavior is obtained. Step S2: Learning behavior graph construction. Based on the original dataset of learning behavior, a learning behavior graph construction method that integrates improved adaptive gating cognitive diagnosis and spatiotemporal graph neural network is used to construct the learning behavior graph, which includes the following steps: heterogeneous learning graph initialization; construction of improved adaptive gating units; dynamic cognitive diagnosis; construction of edge-aware improved spatiotemporal graph network; learning behavior embedding; and learning behavior graph construction. Step S3: English teaching feedback optimization. Based on the learning behavior map, an improved reward reinforcement feedback optimization method combining multimodal map perception is adopted to optimize English teaching feedback, resulting in a teaching feedback optimization strategy, including the following steps: map perception state space construction; action space construction; improved dynamic reward function construction; reinforcement learning training; English teaching feedback optimization. Step S4: Adaptive learning content recommendation. Based on the learning behavior graph, an adaptive learning content recommendation method based on multi-constraint filtering is used to recommend learning content and obtain learning content recommendation data. Step S5: Teaching effectiveness analysis. Based on the teaching feedback optimization strategy and the learning content recommendation data, reference data for teaching and learning evaluation analysis is obtained through data visualization analysis and manual evaluation.

2. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 1, characterized in that: In step S1, the original dataset of learning behavior includes text data, interaction log data, and performance data.

3. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 2, characterized in that: In step S2, the learning behavior graph construction is used to transform the original data into a structured learning behavior graph. Specifically, based on the original learning behavior dataset, a learning behavior graph construction method that integrates improved adaptive gating cognitive diagnosis and spatiotemporal graph neural networks is used to construct the learning behavior graph, resulting in the learning behavior graph. This includes the following steps: Step S21: Initialize the heterogeneous learning graph, specifically by constructing the node set, edge set, node attributes, and time series respectively to obtain the initial heterogeneous learning graph; Step S22: Construct an improved adaptive gating unit, specifically by constructing the cognitive state vector corresponding to the student node based on the initial attributes of the nodes in the initial heterogeneous learning graph, and constructing an improved adaptive gating unit through an adaptive cognitive influence gating mechanism, updating the cognitive state vector, and obtaining updated cognitive influence vector data. Step S23: Dynamic cognitive diagnosis, specifically, involves concatenating the cognitive state vector from the previous time, the features of the current interaction log data, and the updated cognitive impact vector data, and then calculating the dynamic cognitive diagnosis update data by constructing a dynamic cognitive diagnosis function. Step S24: Construct an edge-aware improved spatiotemporal graph network. Specifically, the initial heterogeneous learning graph is updated based on the dynamic cognitive diagnosis update data. The edge weight coefficients are calculated by jointly modeling the updated node and edge features. Based on the edge weight coefficients, the edge-aware improved spatiotemporal graph network is constructed. Step S25: Learning behavior embedding, specifically, based on the dynamic cognitive diagnosis update data and the spatiotemporal graph convolutional features output by the edge-aware improved spatiotemporal graph network, and through feature fusion, a learning behavior embedding vector is obtained; Step S26: Construction of the learning behavior graph. Specifically, based on the learning behavior embedding vector, the learning behavior embedding vector is used as a new node representation. At the same time, the weights of the interaction edges and prior relation edges are adjusted to update the heterogeneous graph, thereby obtaining the learning behavior graph, which is then stored in a graph database.

4. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 3, characterized in that: In step S3, the English teaching feedback optimization is used to generate personalized feedback strategies. Specifically, based on the learning behavior map, an improved reward reinforcement feedback optimization method combining multimodal map perception is used to optimize English teaching feedback and obtain a teaching feedback optimization strategy, including the following steps: Step S31: Construction of the graph perception state space, specifically, based on the learned behavior graph, the state space is constructed to obtain the state space parameters; Step S32: Motion space construction, specifically, by constructing composite motion parameters, the motion space is constructed to obtain motion space parameters; Step S33: Improve the construction of the dynamic reward function. Specifically, by integrating immediate learning outcome rewards, cognitive state rewards, learning behavior rewards, and course planning feedback rewards, the dynamic reward function is improved and the improved reward function is obtained. Step S34: Reinforcement learning training, specifically, through the construction of the graph perception state space, the construction of the action space and the construction of the improved dynamic reward function, reinforcement learning training is carried out through standard reinforcement learning proximal policy optimization training to obtain an English teaching feedback optimization model; Step S35: English teaching feedback optimization, specifically, using the English teaching feedback optimization model to optimize English teaching feedback and obtain teaching feedback optimization strategies.

5. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 4, characterized in that: In step S4, the adaptive learning content recommendation is used to intelligently recommend suitable learning content. Specifically, based on the learning behavior graph, an adaptive learning content recommendation method based on multi-constraint filtering is used to obtain learning content recommendation data, including the following steps: Step S41: Strategy instruction organization; Step S42: Multi-constraint filtering of learning content; Step S43: Multi-factor weighted ranking; Step S44: Adaptive learning content recommendation.

6. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 5, characterized in that: In step S41, the strategy instruction organization specifically involves converting the teaching feedback optimization strategy into recommendation constraint instructions to obtain a recommendation instruction vector; in step S42, the learning content multi-constraint filtering specifically involves performing multi-constraint filtering of learning content based on the recommendation instruction vector, and performing multi-constraint filtering of learning content by setting constraints to obtain initial filtered learning content data.

7. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 6, characterized in that: In step S43, the multi-factor weighted ranking specifically involves constructing a weighted ranking function for candidate learning content, calculating the comprehensive score of each candidate learning content, and obtaining comprehensive score data for recommended learning content. In step S44, the adaptive learning content recommendation specifically involves ranking the learning content based on the comprehensive score data for recommended learning content, and obtaining recommended learning content data.

8. The AI ​​English teaching feedback optimization method based on learning behavior graphs according to claim 7, characterized in that: In step S5, the teaching effectiveness analysis is used to evaluate the feedback and recommendation results. Specifically, based on the teaching feedback optimization strategy and the learning content recommendation data, reference data for teaching and learning evaluation analysis is obtained through data visualization analysis and manual evaluation.

9. An AI-based English teaching feedback optimization system based on learning behavior graphs, used to implement the intelligent agricultural machinery performance improvement method as described in any one of claims 1-7, characterized in that: It includes a data acquisition and perception module, a learning behavior mapping module, a feedback optimization module, a content recommendation module, and an evaluation and analysis module; The acquisition and perception module is used for multimodal data acquisition. Through multimodal data acquisition, the original dataset of learning behavior is obtained, and the original dataset of learning behavior is sent to the learning behavior graph module. The learning behavior graph module is used to construct a learning behavior graph. Through the construction of the learning behavior graph, a learning behavior graph is obtained, and the learning behavior graph is sent to the feedback optimization module and the content recommendation module. The feedback optimization module is used for optimizing English teaching feedback. Through English teaching feedback optimization, a teaching feedback optimization strategy is obtained, and the teaching feedback optimization strategy is sent to the evaluation and analysis module. The content recommendation module is used for adaptive learning content recommendation. Through adaptive learning content recommendation, learning content recommendation data is obtained and the learning content recommendation data is sent to the evaluation and analysis module. The evaluation and analysis module is used for teaching effectiveness analysis, and through this analysis, reference data for teaching and learning evaluation and analysis is obtained.

Images