Subject cognition teaching knowledge construction and intelligent navigation system based on knowledge graph

Abstract

The invention discloses a subject cognition teaching knowledge construction and intelligent navigation system based on a knowledge graph, and relates to the field of education technology and intelligent teaching. The knowledge graph construction module is used for extracting knowledge points and association relationships thereof from subject teaching resources and constructing a structured knowledge graph; the personalized cognitive state vector generation module analyzes the mastering degree of knowledge points by means of a cognitive diagnosis model, and generates a state vector reflecting the cognitive level of an individual; the navigation strategy generation module adopts a hierarchical reinforcement learning method, and dynamically generates a navigation strategy containing knowledge point priorities and an optimal learning path by taking a learning target as guidance; and after the learning effect adjustment module pushes the strategy to the learner, acquiring knowledge point mastering state deviation and learning strategy consistency deviation in real time, and synchronously updating the edge weight of the knowledge graph and cognitive diagnosis model parameters. According to the system, through dynamic evolution of the knowledge graph and self-adaptive updating of the cognitive model, accurate optimization of a teaching strategy is realized, and the personalized learning navigation efficiency is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of educational technology and intelligent teaching, and particularly relates to a subject cognitive teaching knowledge construction and intelligent navigation system based on a knowledge graph. BACKGROUND

[0002] Under the background of digital transformation of education, dynamic construction of subject knowledge and personalized learning navigation face many challenges. In traditional teaching systems, knowledge graphs are mostly constructed based on static textbook texts, which are difficult to integrate the development of subject frontiers and dynamic knowledge associations in real time, resulting in that the dependency relationship between knowledge points cannot accurately reflect the cognitive logic of the subject, for example, the association between "deep learning" and "neural network" in computer science often lags behind the development of technology, affecting the timeliness of the knowledge system.

[0003] In terms of cognitive diagnosis models, traditional methods (such as Bayesian knowledge tracking) are mostly based on fixed parameters to evaluate the mastery of knowledge points, which are difficult to adapt to the dynamic cognitive changes of learners. For example, the mastery of "data structure" by students in programming learning may fluctuate due to different practical scenarios, but traditional models cannot capture this difference in real time, resulting in deviation in the generation of cognitive state vectors.

[0004] In terms of navigation strategy generation, existing systems lack the coordinated optimization of learning goals, knowledge dependency and individual cognition, and the strategy recommendation often has the problem of "paying more attention to mastery degree improvement and less attention to knowledge continuity". For example, in college mathematics teaching, traditional strategies may recommend learning of high-level knowledge points while ignoring consolidation of prerequisite knowledge, resulting in learning path discontinuity. In addition, the learning effect feedback mechanism is lagging behind, and it is difficult to integrate multi-source data (such as answer records, programming assignments, and interactive behaviors) in real time. When students have knowledge gaps, the system cannot quickly update the knowledge graph and diagnosis model, affecting the accuracy of the teaching strategy.

[0005] Therefore, there is an urgent need for a teaching system that can dynamically construct a knowledge graph, adaptively update a cognitive diagnosis model, and realize intelligent optimization of a learning path, in order to overcome the limitations of traditional technology in terms of knowledge dynamics, cognitive adaptability and strategy accuracy. SUMMARY

[0006] The purpose of the present application is to provide a subject cognitive teaching knowledge construction and intelligent navigation system based on a knowledge graph, which overcomes the defects of traditional teaching systems such as static knowledge graph, fixed cognitive diagnosis model, lack of coordination of navigation strategy, and lagging feedback, through dynamic knowledge graph construction, adaptive cognitive diagnosis, hierarchical reinforcement learning strategy generation, and real-time feedback optimization mechanism.

[0007] To achieve the above object, the application provides a subject cognitive teaching knowledge construction and intelligent navigation system based on a knowledge graph, comprising a knowledge graph construction module, a personalized cognitive state vector generation module, a navigation strategy generation module and a learning effect adjustment module. The knowledge graph construction module collects subject teaching resource data through a natural language processing technology, extracts knowledge points and associated relationships, and constructs a knowledge graph. The personalized cognitive state vector generation module analyzes the knowledge point mastery degree based on the learner's answering data using a cognitive diagnosis model to generate a personalized cognitive state vector. The navigation strategy generation module dynamically generates a navigation strategy based on the personalized cognitive state vector and the knowledge graph using a hierarchical reinforcement learning method combined with a learning goal. The learning effect adjustment module pushes the navigation strategy to the learner and collects learning effect data deviation in real time, and when the learning effect data deviation exceeds a preset threshold, updates the edge weight of the knowledge graph and the parameters of the cognitive diagnosis model.

[0008] Further, the natural language processing technology includes text preprocessing, domain term recognition and knowledge point classification. The text preprocessing includes word segmentation, part-of-speech tagging and named entity recognition. The domain term recognition uses FuzzyWuzzy for fuzzy matching by constructing a term dictionary. The knowledge point classification uses a BERT model to classify knowledge points.

[0009] Further, the knowledge graph includes node definition, edge definition and graph structure mathematical representation. The node definition is a knowledge point node . The total number of knowledge point nodes is . The edge definition is a knowledge point dependency edge , representing the precedence relationship of knowledge points, , . The graph structure mathematical representation is as follows: Knowledge graph , wherein: is a set of knowledge point nodes; is a set of knowledge point dependency edges, and the edge weight , , .

[0010] Further, the learner's answering data includes but is not limited to programming assignments. The cognitive diagnosis model uses a Bayesian knowledge tracing model; The calculation process of the Bayesian knowledge tracing model includes a state transition matrix and a correct answer probability; The state transition matrix: From not mastered to mastered: ; From mastered to maintained: ; From mastered to not mastered: ; From not mastered to not mastered: ; The correct answer probability: Mastered and correct: ; Mastered and wrong: ; Not mastered and correct: ; Not mastered and wrong: ; Wherein, represents the mastery state of the knowledge point at time , 1 for mastered, 0 for not mastered; represents the answer situation at time , 1 for correct, 0 for error; represents the retention probability of the mastered knowledge point; represents the learning probability of the not mastered knowledge point; represents the probability of guessing correctly; represents the probability of error of the mastered knowledge point.

[0011] Further, the recursive process of the Bayesian knowledge tracing model includes state prediction, likelihood calculation and state update; Let the initial time , the probability of not mastering the knowledge point is , and the probability of mastering is ; The Bayesian knowledge tracing model is updated according to the following steps: State prediction: based on the state of the previous moment, the prior probability of the state at the current moment is calculated: Wherein, represents the mastery state of the knowledge point at time , 1 for mastered, 0 for not mastered; represents the retention probability of the mastered knowledge point; represents the learning probability of the not mastered knowledge point; Likelihood calculation: calculate the likelihood probability according to the current state prior probability and the answer result wherein, represents the mastery state of the knowledge point at time , 1 for mastery, 0 for non-mastery; represents the answer situation at time , 1 for correct, 0 for error; represents the probability of guessing correctly; represents the probability of failure of the mastered knowledge point; State update: update the posterior probability of the current state according to the answer result using the Bayes formula: wherein, represents the mastery state of the knowledge point at time , 1 for mastery, 0 for non-mastery; represents the answer situation at time , 1 for correct, 0 for error; represents the probability of guessing correctly; represents the probability of failure of the mastered knowledge point; The personalized cognitive state vector is represented as wherein: represents the mastery probability of the knowledge point at time t, .

[0012] Further, the hierarchical reinforcement learning method comprises: State space: ; wherein, is the personalized cognitive state vector at time t; is the knowledge graph structure of the non-mastered knowledge point, containing knowledge point dependency edges; represents the knowledge point gap matrix of the current programming homework of the learner, i.e., the knowledge point mastery probability does not reach the preset threshold; represents the expected knowledge point mastery state of the learner at time t; Action space: ; , represents whether to select to learn the knowledge point ; Reward function: wherein, represents the theoretical mastery degree improvement amount; ​This represents the knowledge point dependency conflict degree, which is the sum of edge weights that do not satisfy the prerequisite knowledge points. The preset threshold; Indicates the distance between the current state and the target state; Navigation strategy update formula: in, For navigation strategy network parameters; :state Take action below The probability of; The preset threshold; For the dominant function, , As a discount factor, It is a value function. ; , .

[0013] Furthermore, the learning outcome data deviation includes deviations in knowledge point mastery status and deviations in the consistency of learning strategies; The knowledge point mastery deviation is the difference between the learner's actual knowledge point mastery and the expected knowledge point mastery, expressed by the following formula: in, Deviation in the mastery of knowledge points; This represents the total number of knowledge points. Indicates the first The actual probability of mastering each knowledge point; For the first The expected mastery status of each knowledge point; when When this happens, the weights of the knowledge graph edges, i.e., the weights of the edges that depend on the knowledge points, are updated. The preset threshold; The formula for updating the weight of edges that depend on knowledge points is as follows: in, This indicates the updated knowledge points. and Knowledge points depend on edge weights. The weights of the knowledge points' dependencies before the update; The learning rate of the knowledge graph; Representing knowledge points The actual set of prerequisite knowledge; For indicator functions, if yes The actual prior knowledge is 1 if it is 1, otherwise it is 0; The learning strategy consistency deviation is the difference between the actual learning path of the learner and the recommended path of the navigation strategy, reflects the deviation of the strategy implementation, and is expressed by the following formula: wherein, is the learning strategy consistency deviation, represents the actual learning path, i.e., the actual learning knowledge point sequence; : the navigation strategy recommended path, i.e., the navigation strategy recommended knowledge point sequence; represents the number of common knowledge points in the actual and recommended paths; When , the cognitive diagnosis model parameter update is triggered, the preset threshold; The cognitive diagnosis model parameter update formula is expressed as follows: wherein, is the learning rate of the cognitive model; is the learning strategy consistency deviation; represents the learning probability of the knowledge point not mastered before the update; represents the probability of the knowledge point mastered before the update; represents the probability of guessing correctly before the update; represents the learning probability of the knowledge point not mastered after the update; represents the probability of the knowledge point mastered after the update; represents the probability of guessing correctly after the update. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0015] Figure 1 is the system module diagram of the present application; Figure 2 is the structure diagram of the personalized cognitive state vector generation module of the present application; Figure 3 is the structure diagram of the navigation strategy generation module of the present application. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present application.

[0017] Example 1: Please refer to Figure 1 As shown, this embodiment is a subject-based cognitive teaching knowledge construction and intelligent navigation system based on knowledge graphs, using the data structures and algorithms course for university computer science majors as the application object, and includes the following modules: The knowledge graph construction module uses natural language processing technology to collect subject teaching resource data, extract knowledge points and relationships, and construct a knowledge graph. In this embodiment, the natural language processing technology includes text preprocessing, domain terminology recognition, and knowledge point classification; The text preprocessing includes word segmentation, part-of-speech tagging, and named entity recognition; The domain terminology recognition is achieved by constructing a terminology dictionary and using FuzzyWuzzy for fuzzy matching, as shown below: from fuzzywuzzy import fuzz term_dict = {"hash table", "dynamic programming", ...} def match_term(text): for term in term_dict: if fuzz.partial_ratio(text, term)>80: return term return None The knowledge point classification uses the BERT model to classify knowledge points, as shown below: from transformers import BertForSequenceClassification, BertTokenizer model = BertForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=2) tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') inputs = tokenizer("Time Complexity Analysis", return_tensors="pt") outputs = model(**inputs) classification = outputs.logits.argmax().item() The subject teaching resource data are derived from textbooks and literature; The textbooks and literature mentioned include textbooks on classic data structures and papers on classic data structure algorithms; The knowledge graph includes node definitions, edge definitions, and mathematical representations of the graph structure. The node is defined as a knowledge point node. ; This represents the total number of knowledge point nodes. Dimensionless; The edge is defined as a knowledge point dependent edge. This indicates the prerequisite relationships between knowledge points. , ， Dimensionless; The mathematical representation of the graph structure is as follows: knowledge graph ,in: A collection of knowledge point nodes, where each knowledge point node is dimensionless, such as an "array" or "linked list"; The set of edges that depend on knowledge points, and the edge weights. , , Dependency edges (dimensionless, representing prerequisite relations); Post-operation dimensions: The graph structure is a set relation, dimensionless; The knowledge graph follows the basic definition of graph theory, with nodes representing knowledge points and edges representing dependencies. Weights quantify the strength of dependencies (e.g., the strength of a "hash table" depending on an "array" is 0.5). Through mathematical representation, the association of knowledge points can be quantified, providing a foundation for subsequent path planning and conforming to the logic of subject knowledge (e.g., a "linked list" requires prior knowledge of an "array").

[0018] The personalized cognitive state vector generation module, based on learners' answer data, uses a cognitive diagnostic model to analyze the degree of knowledge mastery and generates personalized cognitive state vectors, such as... Figure 2 As shown; In this embodiment, the learner answer data includes but is not limited to programming homework; The cognitive diagnosis model uses a Bayesian knowledge tracing model; The Bayesian knowledge tracing model has four core parameters, including: Learning probability : the probability of a knowledge point changing from never mastered to mastered; Retention probability : the probability of a knowledge point remaining mastered; Guessing probability : the probability of a learner guessing correctly when the knowledge point is not mastered; Error probability : the probability of a learner answering incorrectly when the knowledge point is mastered; The Bayesian knowledge tracing model calculation process includes a state transition matrix and a correct answer probability; The state transition matrix: From never mastered to mastered: : dimensionless; From mastered to retained: : dimensionless; From mastered to never mastered: : dimensionless; From never mastered to never mastered: : dimensionless; The correct answer probability: Mastered and correct: : dimensionless; Mastered and incorrect: : dimensionless; Never mastered and correct: : dimensionless; Never mastered and incorrect: : dimensionless; Wherein, represents the mastery state of a knowledge point at time , 1 for mastered and 0 for never mastered; represents the answer situation at time , 1 for correct and 0 for incorrect; represents the retention probability of a mastered knowledge point; represents the learning probability of a never mastered knowledge point; represents the probability of guessing correctly; represents the error probability of a mastered knowledge point; The recursive process of the Bayesian knowledge tracing model includes state prediction, likelihood calculation, and state update; Let the initial time be the probability of a knowledge point being never mastered., the probability of mastering The Bayesian knowledge tracking model updates as follows: State prediction: based on the state of the previous time , the prior probability of the state at the current time is calculated, dimensionless: wherein the above formula is the total probability formula, represents the mastery state of the knowledge point at time , 1 for mastering, 0 for not mastering; represents the retention probability of the mastered knowledge point; represents the learning probability of the unmastered knowledge point; Likelihood calculation: according to the prior probability of the current state and the answer result , the likelihood probability is calculated: wherein, represents the mastery state of the knowledge point at time , 1 for mastering, 0 for not mastering; represents the answer situation at time , 1 for correct, 0 for error; represents the probability of guessing correctly; represents the probability of error of the mastered knowledge point; State update: using the Bayesian formula, the posterior probability of the current state is updated according to the answer result, dimensionless: wherein, represents the mastery state of the knowledge point at time , 1 for mastering, 0 for not mastering; represents the answer situation at time , 1 for correct, 0 for error; represents the probability of guessing correctly; represents the probability of error of the mastered knowledge point; : all are probabilities (dimensionless, taking values [0, 1]); : probability (dimensionless, taking values [0, 1]); The dimension of the operation is: the result is a probability (dimensionless), and both addition and multiplication are dimensionless operations, which can be directly executed; The Bayesian knowledge tracking model dynamically updates the mastery probability through the answer data, conforms to the law of human learning "correcting cognition from errors", and has been widely verified in the field of education (such as knowledge point mastery evaluation in programming homework).

[0019] The personalized cognitive state vector is represented as , wherein: The probability of mastering a knowledge point (dimensionless, value [0,1]). Key points at time t The probability of mastering ; The navigation strategy generation module, based on the personalized cognitive state vector and the knowledge graph, dynamically generates a navigation strategy using a hierarchical reinforcement learning method combined with learning objectives, such as... Figure 3 As shown; In this embodiment, the hierarchical reinforcement learning method includes: State space: ; in, Let be the personalized cognitive state vector at time t; The knowledge graph structure for knowledge points that have not been mastered includes knowledge point dependency edges. This represents the knowledge gap matrix of the learner's current programming assignment, indicating that the probability of mastering a knowledge point has not reached a preset threshold. This indicates the learner's expected mastery of the knowledge points at time t; Action space: ; , indicates whether to select a knowledge point to learn. ; Reward function: in, This represents the increase in theoretical mastery, and is dimensionless. This represents the knowledge point dependency conflict degree, which is the sum of edge weights that do not satisfy the prerequisite knowledge points. The preset threshold; This represents the distance between the current state and the target state; Dimensions after calculation: All terms are dimensionless, can be directly added or subtracted, and the result is the reward value (dimensionless). Navigation strategy update formula: in, These are the network parameters for the navigation strategy, and are dimensionless. :state Take action below The probability is dimensionless; The threshold value is dimensionless. The dominant function is dimensionless. , The discount factor is dimensionless. It is a value function. ; , ; Dimensions after calculation: All terms are dimensionless and can be directly added together. The result is the updated parameters (dimensionless). The training process of the hierarchical reinforcement learning is as follows: Initialization: Setting learning objectives Get the initial state of the students ; High-level decision-making: generating a sequence of sub-objectives; Low-level execution: For each sub-goal, execute the following in a loop: state awareness → action selection → reward acquisition → strategy update; Feedback optimization: If the reward is below the threshold, readjust the sub-goal or action selection navigation strategy; The navigation strategy is an optimal knowledge point learning path generated by a hierarchical reinforcement learning framework.

[0020] The learning effect adjustment module pushes navigation strategies to learners and collects learning effect data deviations in real time. When the learning effect data deviation exceeds a preset threshold, it updates the edge weights of the knowledge graph and the parameters of the cognitive diagnosis model. The learning outcome data deviations include deviations in the mastery of knowledge points and deviations in the consistency of learning strategies. The knowledge point mastery deviation is the difference between the learner's actual knowledge point mastery and the expected knowledge point mastery, expressed by the following formula: in, To assess the deviation in the mastery of knowledge points, the mean absolute difference is used to intuitively measure the gap between the actual and expected mastery. This represents the total number of knowledge points. Indicates the first The actual probability of mastering each knowledge point; For the first The expected mastery status of each knowledge point; when When this happens, the weights of the knowledge graph edges, i.e., the weights of the edges that depend on the knowledge points, are updated. The preset threshold; The formula for updating the weight of edges that depend on knowledge points is as follows: in, This indicates the updated knowledge points. and The knowledge points depend on the edge weights, which are dimensionless. The weights of the knowledge points' dependencies before the update are dimensionless. The learning rate of the knowledge graph is dimensionless. Representing knowledge points The actual set of prior knowledge is dimensionless; For indicator functions, dimensionless, if yes The actual prerequisite knowledge is 1, otherwise 0; each item is dimensionless, and the multiplication is still a weight (dimensionless, ∈ [0, 1]); The learning strategy consistency deviation is the difference between the actual learning path of the learner and the recommended path of the navigation strategy, reflecting the deviation of the strategy implementation, and the formula is as follows: wherein, is the learning strategy consistency deviation, based on the cosine similarity variant (similarity ), which measures the degree of path deviation, represents the actual learning path, i.e., the actual learning knowledge point sequence, which is dimensionless; : navigation strategy recommended path, i.e., the navigation strategy recommended knowledge point sequence, which is dimensionless; represents the number of common knowledge points in the actual and recommended paths, which is dimensionless; the operation is dimensionless: deviation value (dimensionless, value [0, 1]), which can be directly operated; When , the cognitive diagnosis model parameter update is triggered, the preset threshold; The cognitive diagnosis model parameter update formula is as follows: wherein, is the learning rate of the cognitive model, which is dimensionless; is the learning strategy consistency deviation, which is dimensionless; represents the learning probability of the knowledge point not mastered before updating, which is dimensionless; represents the probability of failure of the knowledge point mastered before updating, which is dimensionless; represents the probability of guessing correctly before updating, which is dimensionless; represents the learning probability of the knowledge point not mastered after updating, which is dimensionless; represents the probability of failure of the knowledge point mastered after updating, which is dimensionless; represents the probability of guessing correctly after updating, which is dimensionless; the operation is dimensionless: probability after updating (dimensionless, ∈ [0, 1]), which can be directly added; the parameter update adopts “deviation x learning rate”, which conforms to the parameter optimization logic (such as SGD) of machine learning; Through the deviation, the weights of the knowledge graph (such as increasing the weight if a certain prerequisite knowledge actually has a greater impact) and the parameters of the cognitive model (such as increasing the guessing probability of the learner) are dynamically adjusted to realize the adaptive optimization of the system.

[0021] Actual case example (take 3 knowledge points of data structure course as an example): Set knowledge points: "Array", "Linked list" (dependent on array, ), "Hash table" (dependent on array and linked list, ; ).

[0022] 1. Knowledge graph construction , ; , weight ; Mathematical representation: , consistent with the definition of graph theory.

[0023] 2. Bayesian knowledge tracking model (take "array" as an example) Initial state: (initial mastery probability 20%); Parameters: (unmastered → mastered probability), (mastery → retention probability), (unmastered guessing probability), (mastery error probability); First correct answer ( ): State prediction: ; Likelihood calculation: ; State update: , i.e. (array mastery probability 75%).

[0024] 3. Personalized cognitive state vector Assume , "linked list" mastery probability , "hash table" , then .

[0025] 4. Navigation strategy generation Learning goal: (array 90%, linked list 70%, hash table 80%); State space: ( for , ); Reward function calculation ): (array is higher than initial promotion); (hash table depends on the chain table, the chain table is not enough, take ); ; (reward is negative, need to adjust strategy: priority learning chain table).

[0026] 5. Learning effect adjustment Master state deviation: actual mastery after learning , target : ; If (threshold), trigger edge weight update: prerequisite knowledge of hash table (3) , , , , : (increase the dependence weight of the chain table on the hash table).

[0027] 6. Strategy consistency deviation Recommended path , actual path : (array, hash table); ; If (threshold), no trigger cognitive model parameter update.

[0028] Conclusion: The formulas in the embodiments are based on mature theory derivation, and the parameters are all dimensionless quantities, and the operation is reasonable; The operability of the formula is verified through actual cases, and the symbols are not repeated and have clear explanations.

[0029] The above formulas are all dimensionless forms, only numerical values are used for calculation. These formulas are based on a large amount of data and obtained through software simulation, aiming to be as close to the actual situation as possible. The preset parameters in the formula can be adjusted by those skilled in the art according to specific needs.

[0030] In the description of the specification, reference to "one embodiment", "an example", "a specific example" or the like means that a particular feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. The appearances of the phrases "in one embodiment", "an example", "a specific example" or the like in various places in the specification are not necessarily referring to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0031] The preferred embodiments of the application disclosed above are only to help explain the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments. Obviously, many modifications and variations can be made according to the content of the specification. The specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. A subject-based cognitive teaching knowledge construction and intelligent navigation system based on knowledge graphs, characterized in that, Includes the following modules: The knowledge graph construction module uses natural language processing technology to collect subject teaching resource data, extract knowledge points and relationships, and construct a knowledge graph. The personalized cognitive state vector generation module analyzes the degree of mastery of knowledge points using a cognitive diagnostic model based on learners' answer data and generates personalized cognitive state vectors. The navigation strategy generation module dynamically generates a navigation strategy based on the personalized cognitive state vector and the knowledge graph, using a hierarchical reinforcement learning method combined with learning objectives. The learning effect adjustment module pushes navigation strategies to learners and collects learning effect data deviations in real time. When the learning effect data deviation exceeds a preset threshold, it updates the edge weights of the knowledge graph and the parameters of the cognitive diagnostic model.

2. The system according to claim 1, characterized in that, The natural language processing technology includes text preprocessing, domain terminology recognition, and knowledge point classification. The text preprocessing includes word segmentation, part-of-speech tagging, and named entity recognition; The domain terminology recognition is achieved by constructing a terminology dictionary and using FuzzyWuzzy for fuzzy matching; The knowledge point classification uses the BERT model to classify knowledge points.

3. The system according to claim 2, characterized in that, The knowledge graph includes node definitions, edge definitions, and mathematical representations of the graph structure. The node is defined as a knowledge point node. ; This represents the total number of knowledge point nodes. ; The edge is defined as a knowledge point dependent edge. This indicates the prerequisite relationships between knowledge points. , ; The mathematical representation of the graph structure is as follows: knowledge graph ,in: A collection of knowledge point nodes; The set of edges that depend on knowledge points, and the edge weights. , , .

4. The system according to claim 1, characterized in that, The learner's answer data includes, but is not limited to, programming assignments; The cognitive diagnostic model uses a Bayesian knowledge tracing model; The calculation process of the Bayesian knowledge tracing model includes a state transition matrix and the probability of a correct answer; The state transition matrix: From never having mastered it to having mastered it: ; From mastery to retention: ; From mastery to lack of mastery: ; From never having mastered it to not having mastered it: ; The probability of a correct answer: Answer correctly when you have mastered the material: ; Answering incorrectly when mastering the knowledge: ; Answer correctly when you haven't mastered the knowledge: ; Answering incorrectly when not having mastered the knowledge: ; in, Indicates the knowledge points in time The mastery status is indicated by 1 for mastery and 0 for non-mastery. Indicates time The answer status is 1 for correct and 0 for incorrect; This represents the probability of retaining previously mastered knowledge points. This indicates the probability of learning a knowledge point that has not yet been mastered. This indicates the probability that the guess is correct. This indicates the probability of making a mistake even when you have mastered the knowledge.

5. The system according to claim 1, characterized in that, The recursive process of the Bayesian knowledge tracing model includes state prediction, likelihood calculation, and state update. Let the initial time be At that time, the probability of not mastering the knowledge point is: The probability of mastering it is The Bayesian knowledge tracing model is updated according to the following steps: State prediction: based on the state at the previous time step Calculate the prior probability of the current state: in, Indicates the knowledge points in time The mastery status is indicated by 1 for mastery and 0 for non-mastery. This represents the probability of retaining previously mastered knowledge points. This indicates the probability of learning a knowledge point that has not yet been mastered. Likelihood calculation: based on the prior probability of the current state and the answer result Calculate the likelihood probability: in, Indicates the knowledge points in time The mastery status is indicated by 1 for mastery and 0 for non-mastery. Indicates time The answer status is 1 for correct and 0 for incorrect; This indicates the probability that the guess is correct. This indicates the probability of making a mistake when you have already mastered the knowledge points; State Update: Using Bayes' theorem, update the posterior probability of the current state based on the answer result. in, Indicates the knowledge points in time The mastery status is indicated by 1 for mastery and 0 for non-mastery. Indicates time The answer status is 1 for correct and 0 for incorrect; This indicates the probability that the guess is correct. This indicates the probability of making a mistake when you have already mastered the knowledge points; The personalized cognitive state vector is represented as: ,in: Key points at time t The probability of mastering .

6. The system according to claim 1, characterized in that, The hierarchical reinforcement learning method includes: State space: ; in, Let be the personalized cognitive state vector at time t; The knowledge graph structure for knowledge points that have not been mastered includes knowledge point dependency edges. This represents the knowledge gap matrix of the learner's current programming assignment, indicating that the probability of mastering a knowledge point has not reached a preset threshold. This indicates the learner's expected mastery of the knowledge points at time t; Action space: ; , indicates whether to select a knowledge point to learn. ; Reward function: in, This indicates the increase in theoretical understanding. This represents the knowledge point dependency conflict degree, which is the sum of edge weights that do not satisfy the prerequisite knowledge points. The preset threshold; Indicates the distance between the current state and the target state; Navigation strategy update formula: in, For navigation strategy network parameters; :state Take action below The probability of; The preset threshold; For the dominant function, , As a discount factor, It is a value function. ; , .

7. The system according to claim 1, characterized in that, The learning outcome data deviations include deviations in the mastery of knowledge points and deviations in the consistency of learning strategies. The knowledge point mastery deviation is the difference between the learner's actual knowledge point mastery and the expected knowledge point mastery, expressed by the following formula: in, Deviation in the mastery of knowledge points; This represents the total number of knowledge points. Indicates the first The actual probability of mastering each knowledge point; For the first The expected mastery status of each knowledge point; when When this happens, the weights of the knowledge graph edges, i.e., the weights of the edges that depend on the knowledge points, are updated. The preset threshold; The formula for updating the weight of edges that depend on knowledge points is as follows: in, This indicates the updated knowledge points. and Knowledge points depend on edge weights. The weights of the knowledge points' dependencies before the update; For the knowledge graph learning rate; Representing knowledge points The actual set of prerequisite knowledge; For indicator functions, if yes The actual prerequisite knowledge is 1 if it is 1, otherwise it is 0; The learning strategy consistency deviation is the degree of difference between the learner's actual learning path and the path recommended by the navigation strategy, reflecting the deviation in strategy execution. The formula is as follows: in, For learning strategy consistency bias, This represents the actual learning path, i.e., the sequence of knowledge points actually learned. The navigation strategy recommends a path, which is the sequence of knowledge points recommended by the navigation strategy. This indicates the number of common knowledge points between the actual and recommended paths; when At that time, the cognitive diagnostic model parameters are updated. Preset threshold; The formula for updating parameters in the cognitive diagnostic model is expressed as follows: in, The learning rate of the cognitive model; This is due to learning strategy consistency bias. This indicates that the knowledge points were not mastered before the update. The learning probability; This indicates that the knowledge points were already mastered before the update. The probability of error; This indicates the probability that the guess was correct before the update. This indicates that the knowledge points were not mastered after the update. The learning probability; This indicates that the knowledge points have been mastered after the update. The probability of error; This indicates the probability that the guess is correct after the update.

Images