Computer basic course personalized learning path recommendation method and system based on AI

Abstract

The invention discloses an AI-based computer basic course personalized learning path recommendation method and system, and belongs to the technical field of AI-based data processing. According to the system, behavior data, cognitive data and course interaction data of a learner are acquired through a multi-dimensional data acquisition module, a computer basic course knowledge point association network is established in combination with a dynamic knowledge graph construction module, and a personalized learning path is generated by using an improved deep reinforcement learning algorithm. And the path is dynamically adjusted through the real-time feedback module. The core of the method is that a learner portrait is fused with space-time correlation features of a knowledge graph, a cognitive evaluation model is updated in real time through a Bayesian network, the problems that in a traditional recommendation method, paths are solidified, and the dynamic learning state of an individual is ignored are solved, more accurate personalized learning guidance is achieved, and the learning efficiency and effect of a computer basic course are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of AI-based data processing, in particular to an AI-based computer foundation course personalized learning path recommendation method and system. BACKGROUND

[0002] In computer foundation course teaching, there are significant individual differences in learners' knowledge reserves, learning ability and cognitive style, and the traditional "one-size-fits-all" teaching path has been difficult to meet the needs of personalized learning; the existing learning path recommendation technology mainly has the following deficiencies: Most systems only generate paths based on learners' historical grades or simple interest tags, ignoring the internal relevance of the knowledge system; secondly, the existing methods are mostly static path planning, which cannot respond to the changes in learners' learning state in real time; when a learner has an understanding bias at a certain knowledge point, the system cannot adjust the path in time and still pushes forward according to the preset process, resulting in low learning efficiency; And the existing technology relies on surface data such as correct answer rate to evaluate the cognitive level of learners, without in-depth analysis of error patterns, thinking processes and other deep features; in computer foundation courses, errors in the same knowledge point may be caused by different reasons such as concept confusion, calculation errors or lack of prerequisite knowledge, and a single correct rate evaluation cannot accurately locate the problem, resulting in a lack of targetedness in the recommended path; at the same time, the knowledge graph of the existing system is mostly manually preset knowledge point association, without considering the differences in learners' perception of knowledge point difficulty, and without dynamically optimizing the association weight according to course content updates or teaching practice feedback; In addition, the existing technology has limitations in AI algorithm application: most recommendation algorithms do not combine the dynamic decision-making ability of reinforcement learning, making it difficult to balance between "exploring new knowledge points" and "consolidating known content"; at the same time, there is a lack of multi-modal data fusion mechanism, which cannot effectively utilize multi-source data such as video learning duration, code practice behavior and note keywords, resulting in inaccurate learner profile construction; therefore, there is an urgent need for an AI-based personalized learning path recommendation method and system that can dynamically perceive learner state and deeply integrate knowledge association features. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides an AI-based computer foundation course personalized learning path recommendation method and system, which solves the problems raised in the background art.

[0004] To achieve the above purpose, the present application is implemented by the following technical scheme: an AI-based computer foundation course personalized learning path recommendation method, comprising the following steps: S1, data collection and preprocessing, the behavior data, cognitive data and interaction data of the learner are obtained through a multi-dimensional data collection module, and after cleaning, standardization and feature extraction, a structured data set is formed; S2, learner portrait construction, based on the data set of step S1, the cognitive level is evaluated by Bayesian network, the learning style is classified by SVM, and the knowledge gap is located by contrast threshold, and a multi-dimensional learner portrait is constructed; S3, dynamic knowledge graph construction, the knowledge points of computer foundation course are modeled as nodes, the correlation weight is calculated by PMI, the initial knowledge graph is formed, and is updated regularly based on new data; S4, initial path generation, taking knowledge gap as starting point and course goal as end point, the improved A algorithm is used to generate the initial learning path adapted to the learner portrait; S5, real-time feedback and path optimization, learning effect data is collected through a real-time feedback module, the path is dynamically optimized based on DQN algorithm, and the feedback data is used to update the learner portrait and knowledge graph, forming a continuous iteration of personalized recommendation mechanism.

[0005] Optionally, the specific process of cognitive level evaluation in step S2 is: the answer data of the learner is classified according to knowledge points, the total number of answers and the number of errors of each knowledge point are counted , the error rate is calculated; based on Bayesian network, the error rate is mapped to the probability of not mastering the knowledge point , the initial prior probability is set to 0.5, and every 10 new answer data is received, the conditional probability table is updated by maximum likelihood estimation, and is recalculated; when , is determined as knowledge gap.

[0006] Optionally, in step S4, the improved A algorithm needs to meet the following constraint conditions when searching for the path: the correlation weight of any two consecutive knowledge points in the path (to ensure the coherence of knowledge); the deviation of the average difficulty coefficient of the knowledge points on the path from the current cognitive level of the learner does not exceed 1 level (to ensure the difficulty adaptation).

[0007] Optionally, the trigger conditions of path optimization in step S5 include that the error rate of the learner at a certain knowledge point is more than 0.5 for 3 times in succession; the learning effect evaluation score is less than 6 points; the learner actively marks the knowledge points "difficult to understand"; when the optimization is triggered, the DQN algorithm adjusts the first 3 nodes with the highest correlation weight with the current knowledge point in the path, increases the learning link of the preposed basic knowledge points or replaces them with the same knowledge points with lower difficulty.

[0008] An AI-based computer foundation course individualized learning path recommendation method system, comprising a multi-dimensional data acquisition module, a learner portrait construction module, a dynamic knowledge graph module, a path generation and optimization module, and a real-time feedback module; wherein, The multi-dimensional data acquisition module is used to acquire behavior data, cognitive data and course interaction data of learners in the computer foundation course learning process, and to preprocess; the learner portrait construction module constructs a multi-dimensional learner portrait containing cognitive level, learning style and knowledge gap based on the preprocessed data; the dynamic knowledge graph module is used to establish the association network of computer foundation course knowledge points, and dynamically update the knowledge point association weight according to the learner group data and individual feedback; the path generation and optimization module generates an initial learning path based on the learner portrait and the dynamic knowledge graph through an improved deep reinforcement learning algorithm, and dynamically optimizes in combination with real-time feedback; the real-time feedback module is used to collect learning effect data of learners on the recommended path, and feedback to the learner portrait construction module and the dynamic knowledge graph module, forming a closed-loop adjustment mechanism.

[0009] Optionally, the multi-dimensional data acquisition module includes a behavior data acquisition unit, a cognitive data acquisition unit, an interaction data acquisition unit and a data preprocessing unit; the behavior data acquisition unit is used to acquire the video watching time, pause / playback times, courseware download frequency, code exercise submission times and programming environment operation trajectory of learners; the cognitive data acquisition unit is used to acquire the answering data (including multiple-choice question options, fill-in-the-blank answers, programming code and error prompts) of learners, knowledge point self-test scores and learning reflection note texts; the interaction data acquisition unit is used to acquire the question content of learners in the learning community, reply frequency, like / collection behavior and dialogue record with the teaching robot; the data preprocessing unit is used to clean, standardize (map different dimensional data to the [0, 1] interval) and feature extract (use the TF-IDF algorithm to extract keywords from text data, and use the LSTM model to extract time sequence features from behavior sequences) the collected data, and the specific standardization formula is: wherein, is the original data value, , are the minimum and maximum values of the dimensional data respectively.

[0010] Optionally, the learner portrait construction module includes a cognitive level evaluation unit, a learning style classification unit and a knowledge gap positioning unit; the cognitive level evaluation unit uses a Bayesian network model to input the preprocessed cognitive data, and outputs the mastery probability of each knowledge point, and the conditional probability table of the Bayesian network is updated by maximum likelihood estimation, and the formula is: wherein, is a question error event, is a knowledge point not mastered event, is the probability of question error when the knowledge point is not mastered, is the prior probability of the knowledge point not being mastered; the learning style classification unit divides the learners into visual, auditory, reading or kinesthetic types based on the VARK model by analyzing the video / text preference, practical operation frequency and community interaction depth in the behavior data, and the classification algorithm uses a support vector machine (SVM) with a radial basis function as the kernel function ); the knowledge gap positioning unit positions the knowledge points below the threshold by comparing the knowledge point mastery probability of the learner with the course standard mastery threshold (set to 0.7), and analyzes the gap causes (concept confusion, calculation error or missing prerequisite knowledge) in combination with the error mode.

[0011] Optionally, the dynamic knowledge graph module includes a knowledge point modeling unit, an association weight calculation unit and a graph updating unit; the knowledge point modeling unit takes the knowledge points (including "computer composition", "operating system", "C language programming", etc.) of the computer foundation course as nodes, and the node attributes include difficulty coefficient (1-5 levels), learning time suggestion (minutes) and prerequisite knowledge point list; the association weight calculation unit calculates the association strength between knowledge points by point mutual information (PMI), and the formula is: wherein, , are two knowledge points, is the probability of the learner simultaneously learning and , , are the probabilities of learning , , respectively, and the association weight ; the graph updating unit recalculates the association weight based on the newly added learner behavior data every 7 days, and adjusts the difficulty coefficient of the key knowledge points according to the teacher feedback, so as to realize the dynamic evolution of the graph.

[0012] Optionally, the path generation and optimization module includes an initial path generation unit and a path optimization unit; the initial path generation unit takes the current knowledge gap of the learner as the starting point and the course target knowledge point as the end point, and uses an improved A algorithm to search for the optimal path, and the heuristic function is designed as: wherein, is a node​ a shortest path length to a target node, an average difficulty coefficient of knowledge points on the path, 、 a weight coefficient (a, b, c) , ), ensuring that the path takes into account the knowledge relevance and difficulty adaptability; the path optimization unit dynamically adjusts the path node order based on the DQN algorithm in deep reinforcement learning, taking the learning effect of the learner on the path as a reward signal, The value update formula is: wherein, is the current learning state, is the path adjustment action, is the reward value, is the discount factor, is the new state after adjustment.

[0013] Optionally, the real-time feedback module includes an effect evaluation unit and a feedback transmission unit; the effect evaluation unit generates an effect evaluation score (0-10) by comparing the knowledge point mastery probability change, learning time fluctuation and error rate decline trend of the learner before and after learning on the recommended path, and the evaluation formula is: wherein, is the mastery probability improvement value, is the difference between the actual learning time and the recommended time, is the recommended time, is the error rate change value; the feedback transmission unit feeds back the evaluation score and specific learning data to the learner portrait construction module (for updating the cognitive level evaluation) and the dynamic knowledge graph module (for adjusting the correlation weight of related knowledge points) respectively, to realize closed-loop updating once every 24 hours.

[0014] The application provides an AI-based computer foundation course personalized learning path recommendation method and system, which has the following beneficial effects: The AI-based computer foundation course personalized learning path recommendation method and system realizes dynamic construction of multi-dimensional learner portraits, breaking through the limitations of traditional single-dimensional evaluation; by fusing behavior data, cognitive data and interaction data, the learner characteristics are accurately described from three dimensions of cognitive level, learning style and knowledge gap, and especially the Bayesian network is used to update the cognitive evaluation model in real time, which can deeply analyze the causes behind the error mode, making the portrait more suitable for the dynamic learning state of the learner; such a multi-dimensional fusion mechanism solves the problem of rigid learner labels in the prior art, and lays a precise data foundation for personalized path recommendation; Secondly, the dynamic knowledge graph module overcomes the defects that the static knowledge graph cannot adapt to individual differences and course evolution; the knowledge point correlation weight is calculated through point mutual information, and is regularly updated combined with learner group data and teacher feedback, so that the knowledge graph can not only reflect the objective knowledge correlation of computer foundation courses, but also reflect the subjective cognitive differences of different learners; the matching degree of knowledge correlation characteristics and actual needs of learners is ensured, and the rationality of path recommendation is improved; Thirdly, the improved A algorithm and deep reinforcement learning DQN algorithm are introduced in path generation and optimization to realize the collaborative mechanism; the initial path balances the knowledge correlation and difficulty adaptability through the heuristic function, avoiding the disadvantages of the traditional A algorithm which only focuses on the path length; and the DQN algorithm dynamically optimizes the path based on the reward signal of real-time learning effect, solving the problem that the static path cannot respond to the state changes of learners. In addition, a closed-loop feedback mechanism of "data collection - portrait update - graph optimization - path adjustment" is constructed to form a continuous iteration of personalized recommendation logic; the real-time feedback module feeds back the learning effect data to the portrait and graph modules every 24 hours, so that the system can continuously optimize the recommendation strategy as the learner progresses, avoiding the limitations of "one-time recommendation" in the prior art; the recommended path is always synchronized with the real-time state of the learner, which is especially suitable for the learning scenario of computer foundation courses where knowledge points are progressively advanced. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The figure is a flowchart of the method of the application; Figure 2 The figure is a flowchart of the system of the application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be described below in conjunction with the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all embodiments.

[0017] In the description of the application, unless otherwise specified, the meaning of "multiple" is two or more; the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail" and the like indicate the orientation or positional relationship shown in the drawings, which is only for the purpose of facilitating the description of the application and simplifying the description, and does not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation on the application. In addition, the terms "first", "second", "third" and the like are only for descriptive purposes, and cannot be understood as indicating or implying relative importance.

[0018] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0019] Please see Figure 1 This invention provides a technical solution: a personalized learning path recommendation method for computer fundamentals courses based on AI, comprising the following steps: S1. Data Acquisition and Preprocessing: Learners' behavioral data, cognitive data, and interaction data are acquired through a multi-dimensional data acquisition module. After cleaning, standardization, and feature extraction, a structured dataset is formed. S2. Learner profile construction: Based on the dataset from step S1, a multi-dimensional learner profile is constructed by using a Bayesian network to assess cognitive level, an SVM to classify learning style, and a comparison threshold to locate knowledge gaps. The specific process of cognitive level assessment is as follows: Learners' answer data is categorized by knowledge point, and the total number of answers for each knowledge point is counted. and number of errors Calculate the error rate Based on Bayesian networks, the error rate is... Mapped to the probability of not mastering a knowledge point Initial prior probability Set to 0.5, update the conditional probability table using maximum likelihood estimation and recalculate every 10 new responses received. ,when At that time, it was determined to be a knowledge gap; S3. Dynamic knowledge graph construction: Computer fundamentals course knowledge points are modeled as nodes, association weights are calculated through PMI to form an initial knowledge graph, and it is updated regularly based on new data. S4. Initial path generation: Starting with knowledge gaps and ending with course objectives, an improved A algorithm is used to generate an initial learning path that is adapted to the learner profile. The improved A algorithm must satisfy the following constraint during path search: the association weight between any two consecutive knowledge points in the path. The average difficulty level of the knowledge points along the learning path deviates from the learner's current cognitive level by no more than one level. S5. Real-time feedback and path optimization: The learning effect data is collected through the real-time feedback module, the path is dynamically optimized based on the DQN algorithm, and the feedback data is used to update the learner profile and knowledge graph to form a continuously iterative personalized recommendation mechanism. The triggering condition for path optimization includes a learner's error rate on a certain knowledge point exceeding 0.5 for three consecutive times; If the learning outcome assessment score is below 6, the learner actively marks the knowledge points as "difficult to understand". When optimization is triggered, the DQN algorithm prioritizes adjusting the top 3 nodes with the highest weight associated with the current knowledge point in the path, adding learning links for prerequisite basic knowledge points or replacing them with similar knowledge points of lower difficulty.

[0020] Please see Figure 2 The present invention provides a technical solution: a system for recommending personalized learning paths for computer basic courses based on AI, including a multi-dimensional data acquisition module, a learner profile construction module, a dynamic knowledge graph module, a path generation and optimization module, and a real-time feedback module; in, The multi-dimensional data acquisition module is used to collect learners' behavioral data, cognitive data, and course interaction data during the computer basics course, and to preprocess them. The multi-dimensional data acquisition module includes a behavioral data acquisition unit, a cognitive data acquisition unit, an interaction data acquisition unit, and a data preprocessing unit. The behavioral data acquisition unit collects learners' video viewing time, pause / replay times, courseware download frequency, code practice submissions, and programming environment operation patterns. The cognitive data acquisition unit collects learners' answer data (including multiple-choice options, fill-in-the-blank answers, programming code, and error messages), self-test scores for knowledge points, and learning reflection notes. The interaction data acquisition unit collects learners' questions in the learning community, reply frequency, likes / favorites, and dialogue records with the teaching robot. The data preprocessing unit cleans the collected data (removing outliers, such as learning time exceeding a reasonable range), standardizes (mapping different dimensions of data to the [0,1] interval), and extracts features (using the TF-IDF algorithm to extract keywords from text data and the LSTM model to extract temporal features from behavioral sequences). The specific standardization formula is as follows: in, The original data values, , These are the minimum and maximum values ​​of the data in this dimension, respectively. The learner profile building module constructs a multi-dimensional learner profile based on preprocessed data, including cognitive level, learning style, and knowledge gaps. The learner profile construction module includes a cognitive level assessment unit, a learning style classification unit, and a knowledge gap location unit. The cognitive level assessment unit uses a Bayesian network model, taking preprocessed cognitive data as input and outputting the mastery probability of each knowledge point. The conditional probability table of the Bayesian network is updated through maximum likelihood estimation, with the following formula: wherein, is the answer error event, is the knowledge point not mastered event, is the probability of answer error when the knowledge point is not mastered, is the prior probability of knowledge point not mastered; the learning style classification unit divides the learners into visual, auditory, reading or kinesthetic types based on the video / text preference, practical operation frequency and community interaction depth in the behavior data, and the classification algorithm uses support vector machine (SVM) and radial basis function as the kernel function ); the knowledge gap positioning unit positions the knowledge points below the threshold by comparing the knowledge point mastery probability of the learner with the course standard mastery threshold (set as 0.7), and analyzes the gap causes (concept confusion, calculation error or missing prerequisite knowledge) in combination with the error mode; The dynamic knowledge graph module is used to establish the association network of the knowledge points of the computer foundation course, and dynamically update the association weight of the knowledge points according to the learner group data and individual feedback; The dynamic knowledge graph module includes a knowledge point modeling unit, an association weight calculation unit and a graph updating unit; the knowledge point modeling unit takes the knowledge points of the computer foundation course as nodes, and the node attributes include difficulty coefficient (1-5 levels), learning time suggestion (minutes) and prerequisite knowledge point list; the association weight calculation unit calculates the association strength between knowledge points by point mutual information (PMI), and the formula is: wherein, , are two knowledge points, is the probability that the learner simultaneously learns and , , are the probabilities of learning , respectively, and the association weight ; the graph updating unit recalculates the association weight based on the newly added learner behavior data every 7 days, and adjusts the difficulty coefficient of the key knowledge points according to the teacher feedback, so as to realize the dynamic evolution of the graph; The path generation and optimization module generates the initial learning path based on the learner portrait and the dynamic knowledge graph by using the improved deep reinforcement learning algorithm, and dynamically optimizes it in combination with the real-time feedback; the path generation and optimization module includes an initial path generation unit and a path optimization unit; the initial path generation unit takes the current knowledge gap of the learner as the starting point and the target knowledge point of the course as the end point, and uses the improved A algorithm to search the optimal path, and the heuristic function is designed as: ​ wherein, is the node the shortest path length to the target node, is the average difficulty coefficient of the knowledge points on the path, , is the weight coefficient (wherein, , ), ensuring that the path takes into account the knowledge relevance and difficulty adaptability; the path optimization unit dynamically adjusts the path node order based on the DQN algorithm in deep reinforcement learning, taking the learning effect of the learner on the path as the reward signal, The value update formula is: wherein, is the current learning state, is the path adjustment action, is the reward value, is the discount factor, is the adjusted new state; The real-time feedback module is used to collect the learning effect data of the learner on the recommended path, and is fed back to the learner portrait construction module and the dynamic knowledge graph module, forming a closed-loop adjustment mechanism. The real-time feedback module includes an effect evaluation unit and a feedback conduction unit; the effect evaluation unit generates an effect evaluation score (0-10) by comparing the knowledge point mastery probability change, learning time fluctuation and error rate decline trend of the learner before and after learning on the recommended path, and the evaluation formula is: wherein, is the mastery probability improvement value, is the difference between the actual learning time and the recommended time, is the recommended time, is the error rate change value; the feedback conduction unit feeds back the evaluation score and specific learning data to the learner portrait construction module (for updating the cognitive level evaluation) and the dynamic knowledge graph module (for adjusting the correlation weight of related knowledge points) respectively, realizing closed-loop update once every 24 hours.

[0021] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. An AI-based computer foundation course individualized learning path recommendation method, characterized in that, Comprise the following steps: S1, data acquisition and pretreatment, through multi-dimensional data acquisition module to obtain the behavior data, cognitive data and interactive data of learners, after cleaning, standardization and feature extraction, form structured data set; S2, learner portrait construction, based on the data set of step S1, through the Bayesian network to evaluate the cognitive level, SVM classification learning style, and compare the threshold to locate the knowledge gap, and construct the multi-dimensional learner portrait; S3, dynamic knowledge graph construction, modeling the knowledge points of computer foundation course as nodes, calculating the correlation weight through PMI, forming the initial knowledge graph, and updating regularly based on new data; S4, initial path generation, taking knowledge gap as starting point and course goal as end point, using improved A algorithm to generate initial learning path adapted to learner portrait; S5, real-time feedback and path optimization, collecting learning effect data through real-time feedback module, dynamically optimizing the path based on DQN algorithm, and using feedback data to update learner portrait and knowledge graph, forming continuous iteration of personalized recommendation mechanism.

2. The method of claim 1, wherein, The specific process of cognitive level assessment in step S2 is as follows: The learner's answer data is categorized by knowledge point, and the total number of answers for each knowledge point is counted. and number of errors Calculate the error rate Based on Bayesian networks, the error rate is... Mapped to the probability of not mastering a knowledge point Initial prior probability Set to 0.5, update the conditional probability table using maximum likelihood estimation and recalculate every 10 new responses received. ,when At that time, it was determined to be a knowledge gap.

3. The method of claim 1, wherein, The improved A algorithm in step S4 needs to meet the following constraint conditions when searching for a path: the correlation weight of any two consecutive knowledge points in the path The average difficulty coefficient of the knowledge points on the path and the deviation of the learner's current cognitive level are not more than 1 level.

4. The method of claim 1, wherein, The trigger conditions of path optimization in step S5 include that the error rate of learners at a certain knowledge point is more than 0.5 for 3 times in succession; The learning effect evaluation score is lower than 6 points; the learners actively mark the knowledge points "difficult to understand"; when the optimization is triggered, the DQN algorithm adjusts the first 3 nodes with the highest correlation weight associated with the current knowledge point in the path, increases the learning link of the preposed basic knowledge points or replaces them with similar knowledge points with lower difficulty.

5. A system for the AI-based computer foundation course individualized learning path recommendation method of claims 1 to 4, characterized by, Comprise multi-dimensional data acquisition module, learner portrait construction module, dynamic knowledge graph module, path generation and optimization module, real-time feedback module; Among them, The multi-dimensional data acquisition module is used for collecting the behavior data, cognitive data and course interaction data of learners in the process of learning computer foundation course, and pretreating; the learner portrait construction module constructs a multi-dimensional learner portrait containing cognitive level, learning style and knowledge gap based on the pretreated data; the dynamic knowledge graph module is used for establishing the correlation network of computer foundation course knowledge points, and dynamically updating the knowledge point correlation weight according to the learner group data and individual feedback; the path generation and optimization module generates the initial learning path based on the learner portrait and dynamic knowledge graph through the improved deep reinforcement learning algorithm, and dynamically optimizes combined with real-time feedback; the real-time feedback module is used for collecting the learning effect data of learners on the recommended path, and feeding back to the learner portrait construction module and dynamic knowledge graph module, forming a closed loop adjustment mechanism.

6. The system of claim 1, wherein, The multi-dimensional data acquisition module comprises a behavior data acquisition unit, a cognitive data acquisition unit, an interactive data acquisition unit and a data preprocessing unit; The behavior data collection unit is used for collecting video watching time, pause / playback times, courseware downloading frequency, code practice submission times and programming environment operation trajectory of the learner; the cognitive data collection unit is used for collecting answering data, knowledge point self-test scores and learning reflection note texts of the learner; the interaction data collection unit is used for collecting question contents, reply frequency, like / collection behaviors and dialogue records with the teaching robot of the learner in the learning community; the data preprocessing unit is used for cleaning, standardizing and feature extracting the collected data, and a specific standardization formula is as follows: wherein, is the original data value, , are the minimum and maximum values of the dimension data, respectively.

7. The system of claim 1, wherein, The learner portrait construction module comprises a cognitive level evaluation unit, a learning style classification unit and a knowledge gap positioning unit; the cognitive level evaluation unit adopts a Bayesian network model, takes the preprocessed cognitive data as input, and outputs a mastery probability of each knowledge point; a conditional probability table of the Bayesian network is updated through maximum likelihood estimation, and a formula is as follows: wherein, is a question error event, is a knowledge point not mastered event, is the probability of question error when the knowledge point is not mastered, is the prior probability of the knowledge point not being mastered; the learning style classification unit divides the learners into visual, auditory, reading or kinesthetic types based on the VARK model by analyzing the video / text preference, practical operation frequency and community interaction depth in the behavior data, and the classification algorithm uses a support vector machine (SVM) with a radial basis function as the kernel function ; the knowledge gap positioning unit positions the knowledge points below the threshold by comparing the knowledge point mastery probability of the learner with the course standard mastery threshold (set to 0.7), and analyzes the gap causes in combination with the error mode.​ 8. The system of claim 1, wherein, The dynamic knowledge graph module comprises a knowledge point modeling unit, an association weight calculation unit and a graph updating unit; the knowledge point modeling unit takes knowledge points of a computer foundation course as nodes, and node attributes comprise a difficulty coefficient, a learning time length suggestion and a pre-knowledge point list; The association weight calculation unit calculates an association strength between knowledge points through point mutual information, and a formula is as follows: wherein, , are two knowledge points, are simultaneously learned by the learner and the probability of learning , the probability of learning , the probability of learning ; the graph updating unit re-computes the correlation weight based on the newly added learner behavior data every 7 days, and adjusts the difficulty coefficient of the key knowledge points according to the teacher feedback, realizing the dynamic evolution of the graph.

9. The system of claim 1, wherein, The path generation and optimization module comprises an initial path generation unit and a path optimization unit; the initial path generation unit takes a current knowledge gap of the learner as a starting point and a course target knowledge point as an end point, adopts an improved A algorithm to search an optimal path, and a heuristic function is designed as follows: wherein, is a node the shortest path length to the target node, is the average difficulty coefficient of the knowledge points on the path, , is a weight coefficient (wherein, , ), ensuring that the path takes into account both knowledge relevance and difficulty adaptability; the path optimization unit dynamically adjusts the path node order based on the DQN algorithm in deep reinforcement learning, taking the learning effect of the learner on the path as a reward signal, the value update formula is: wherein, is the current learning state, is the path adjustment action, is the reward value, is the discount factor, is the adjusted new state.

10. The system of claim 1, wherein, The real-time feedback module comprises an effect evaluation unit and a feedback conduction unit; the effect evaluation unit generates an effect evaluation score by comparing a knowledge point mastery probability change, a learning time length fluctuation and an error rate downward trend of the learner before and after learning on a recommended path, and an evaluation formula is as follows: wherein, is a probability promotion value, is a difference between an actual learning duration and a recommended duration, is a recommended duration, is an error rate change value; the feedback conduction unit feeds back the evaluation score and the specific learning data to the learner portrait construction module and the dynamic knowledge graph module, respectively, to realize closed-loop updating once every 24 hours.

Images