Course ai-assisted question answering and learning progress tracking method

Abstract

The application discloses a course AI-assisted question answering and learning progress tracking method, comprising: constructing and maintaining a multi-dimensional dynamic knowledge graph of a target course, wherein nodes of the multi-dimensional dynamic knowledge graph comprise course knowledge points, concepts, skills and common cognitive misconceptions, edges between the nodes comprise logical relationships, dependency relationships and cognitive confusion probability relationships generated based on statistics, and the method relates to the field of course AI assistance.The application can actively predict and intervene in potential cognitive confusion by monitoring and analyzing the implicit learning behavior of a user, thereby realizing a change from passive response to active guidance; meanwhile, a deep cognitive state model is constructed by fusing multi-dimensional interactive data, thereby realizing accurate quantitative evaluation of knowledge mastery, stability and confusion relationships, and overcoming the limitations of relying on surface behavior data.

Description

Technical Field

[0001] This invention relates to the field of AI-assisted learning in courses, specifically to a method for AI-assisted Q&A and learning progress tracking in courses. Background Technology

[0002] With the popularization of online education and intelligent learning platforms, how to use artificial intelligence technology to provide students with accurate and timely assistance and answers, and effectively track their learning progress, has become a key issue in improving teaching effectiveness. Traditional learning support methods usually rely on teachers' manual answers or static FAQ databases, which are difficult to cope with large-scale, personalized, and constantly changing learning needs, and also fail to provide in-depth insights into students' true cognitive state and learning difficulties.

[0003] Currently, existing AI-assisted learning systems primarily provide answers by analyzing questions proactively raised by students and retrieving matching answers from a pre-set knowledge base. Simultaneously, based on statistics of explicit learning behavior data such as video viewing progress, homework completion, and test scores, they construct simple learning progress reports or recommend the next learning content. Some more advanced systems attempt to build subject-specific knowledge graphs to better organize learning resources and understand the intent behind questions.

[0004] However, the aforementioned existing technical solutions, on the one hand, have two main drawbacks. First, their Q&A mechanisms are essentially passive responses, only able to handle questions explicitly raised by students, and unable to proactively identify and intervene in cognitive doubts or thinking blocks that students may have not yet developed into clear questions during the learning process. Second, their learning progress tracking relies heavily on surface-level, outcome-based behavioral data, lacking in-depth mining and analysis of implicit behaviors during the learning process (such as hesitation, repetition, and revision traces). This results in a superficial and delayed assessment of students' cognitive mastery, knowledge solidity, and potential weaknesses. Consequently, the learning path suggestions provided are often not precise enough, making it difficult to achieve truly personalized and adaptive teaching. Summary of the Invention

[0005] Based on this, the purpose of this invention is to provide a course AI-assisted Q&A and learning progress tracking method to solve the technical problems in the prior art, such as delayed Q&A response, superficial characterization of learning status, and insufficient accuracy of personalized guidance, which are caused by the inability to actively identify students' potential cognitive confusion and the reliance on only surface learning data for evaluation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for AI-assisted Q&A and learning progress tracking in courses, comprising the following steps: S1: Constructing and maintaining a multi-dimensional dynamic knowledge graph of the target course, wherein the nodes of the multi-dimensional dynamic knowledge graph include course knowledge points, concepts, skills, and common cognitive misconceptions, and the edges between nodes include logical relationships, dependency relationships, and cognitive confusion probability relationships generated based on statistics; S2: Monitoring the user's full-process interaction flow in the learning environment, wherein the full-process interaction flow includes explicit Q&A interaction and implicit learning behavior; the explicit Q&A interaction is the question-and-answer process actively submitted by the user; the implicit learning behavior includes at least the text content that the user enters and then deletes on the Q&A interface, the behavior of repeatedly jumping to view specific learning resources, and the marking of answer options as uncertain and their modification traces in self-test exercises; S3: Based on the implicit learning behavior, predicting the user's potential knowledge point confusion or cognitive conflict through a pre-trained behavior-intent inference model, and generating pre-Q&A prompts to be actively pushed to the user. The pre-answer prompts are guiding questions or targeted explanation segments used to clarify inferred cognitive conflicts; S4: For explicit answer interactions and interactions that trigger pre-answer prompts, the parsed questions are mapped and matched with the multidimensional dynamic knowledge graph to generate answer content adapted to the user's current cognitive state; at the same time, the complete context information of this interaction is recorded as a cognitive state sample; S5: All cognitive state samples are integrated with the user's historical learning behavior data to construct and continuously update the user's personalized cognitive state model; the personalized cognitive state model is used to represent the user's multidimensional mastery state of each node in the multidimensional dynamic knowledge graph, and the multidimensional mastery state includes at least mastery degree, stability, and a set of easily confused related nodes; S6: Based on the personalized cognitive state model, the user's personal learning path is dynamically generated and adjusted; the personal learning path includes a sequence of knowledge nodes to be learned and reviewed, the best learning modality recommended for each node, and micro-course content tailored to make up for specific weaknesses.

[0007] The present invention is further configured such that the step of maintaining the multidimensional dynamic knowledge graph in step S1 includes: S11: initially constructing the multidimensional dynamic knowledge graph based on the course syllabus and textbooks; S12: continuously analyzing the Q&A interaction records and implicit learning behavior data of all users, automatically identifying frequently co-occurring cognitive confusion knowledge point pairs, and dynamically updating the cognitive confusion probability relationship between corresponding nodes in the knowledge graph; S13: when a pre-Q&A prompt for a certain knowledge point is frequently triggered by the user and confirmed as effective, the cognitive misconception targeted by the pre-Q&A prompt is used as a new node or node attribute and updated to the multidimensional dynamic knowledge graph.

[0008] The present invention is further configured such that the training method of the behavior-intent inference model in step S3 includes: S31: collecting massive amounts of anonymized user learning interaction sequence data, wherein the user learning interaction sequence data contains the complete process from implicit learning behavior to finally explicitly raising a question or experiencing a turning point in understanding; S32: labeling the user learning interaction sequence data, establishing a correlation between implicit learning behavior sequences of specific patterns and the types of knowledge point confusions subsequently exposed; S33: training a temporal neural network model based on the labeled sequence data, enabling it to predict the types of cognitive difficulties that the user is about to face or already has but has not explicitly stated, and their associated knowledge points, based on the input recent implicit learning behavior sequences.

[0009] The present invention is further configured such that the construction of the personalized cognitive state model in step S5 specifically includes: S51: For each target node in the multidimensional dynamic knowledge graph, extracting a feature set related to the target node from all cognitive state samples; the feature set includes the angle and depth of the user's questions about the node, the response time after receiving the answer, the stability in subsequent related self-tests, and the frequency of association with easily confused nodes; S52: Using the feature set, calculating the multidimensional state vector of the target node through a machine learning model; the mastery represents the knowledge reproduction ability, the stability represents the reliability of knowledge transfer in different contexts, and the easily confused associated node set is generated by weighting the user's personal historical confusion records and the group cognitive confusion probability relationship in the knowledge graph.

[0010] The present invention is further configured such that the dynamic generation and adjustment of the personal learning path in step S6 specifically includes: S61: Based on the personalized cognitive state model, identifying nodes to be intervened that meet any of the following conditions: mastery is lower than a first preset threshold, or stability is lower than a second preset threshold; S62: For each node to be intervened, combining the user's metacognitive feature tags, matching the best learning modality and micro-course content from the learning resource library; the metacognitive feature tags are obtained by analyzing the correlation data between the user's historical learning performance and different learning modalities, and are used to identify the user's preferred information reception and processing style; S63: Based on the dependency relationship and cognitive confusion probability relationship in the multidimensional dynamic knowledge graph, performing topological sorting and conflict resolution on the selected nodes to be intervened and the matched learning resources, generating a logically coherent and cognitively optimized linear learning path and network review path.

[0011] The present invention is further configured such that the method for generating the micro-course content in step S62 is as follows: for the specific weak dimension of the node to be intervened, relevant knowledge point fragments are extracted from the standard course resources, and reorganized and connected according to the preset cognitive correction logic framework. At the same time, positive and negative cases adapted from the user's high-frequency errors or points of confusion are embedded, and targeted explanation units are synthesized in real time by programming.

[0012] The present invention is further configured to include step S7: establishing an anonymous cognitive state network across users, analyzing student groups with similar personalized cognitive state models or similar personal learning path adjustment patterns, discovering potential effective learning strategies or common teaching difficulties, and feeding back the analysis results to course designers to optimize course content and multidimensional dynamic knowledge graph structure.

[0013] The present invention is further configured such that when generating Q&A content in step S4, a multi-turn dialogue format is adopted, and the real-time changes in the user's cognitive state are dynamically evaluated during the dialogue, and the depth, breadth and expression of the subsequent dialogue content are adjusted according to the evaluation results.

[0014] In summary, the present invention has the following main beneficial effects: This invention monitors and analyzes users' implicit learning behaviors, enabling proactive prediction and intervention of their potential cognitive confusion, thus shifting from passive response to proactive guidance. Simultaneously, by integrating multi-dimensional interactive data to construct a deep cognitive state model, it achieves precise quantitative assessment of knowledge mastery, stability, and confusion relationships, overcoming the limitations of relying on surface-level behavioral data. Finally, personalized learning paths and micro-courses dynamically generated based on this model can adaptively adjust to users' specific weaknesses, significantly improving the accuracy of personalized learning support and the effectiveness of teaching intervention. Attached Figure Description

[0015] Fig. 1 This is a flowchart of the method of the present invention; Fig. 2 This is a schematic diagram illustrating the construction and dynamic maintenance of the multidimensional dynamic knowledge graph of the present invention; Fig. 3 This is a diagram illustrating the process of constructing the personalized cognitive state model of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] like Figs. 1-3As shown, this invention provides a method for AI-assisted Q&A and learning progress tracking in courses, aiming to achieve proactive and precise teaching intervention through in-depth analysis of students' learning behaviors. The following detailed description of the invention is provided in conjunction with specific steps.

[0018] Step S1: Construct and maintain a multidimensional dynamic knowledge graph of the target course.

[0019] This knowledge graph is the core data structure of the entire system. It is not only a static map of course knowledge, but also a cognitive model that can dynamically evolve with teaching feedback. The specific construction and maintenance process is as follows: First, during the system initialization phase, based on the official syllabus, authoritative textbooks, and lesson plans of the target course, subject matter experts and engineers collaborated to extract core course knowledge points, key concepts, essential skills, and common cognitive misconceptions summarized from teaching experience, which served as the initial nodes of the knowledge graph. Then, various relationships between nodes were defined as "edges," primarily including: 1) logical relationships, such as "belongs to" or "prerequisite is"; 2) dependency relationships, such as "knowledge of chapter A is the foundation for learning chapter B"; 3) cognitive confusion probability relationships, which are quantified edges, with initial values ​​set based on expert experience. For example, "knowledge point X and knowledge point Y are easily confused by students" can be assigned an initial probability value. This graph is stored in a graph database for efficient relational queries and traversal.

[0020] The "dynamic" nature of this knowledge graph is reflected in its continuous self-updating capability. During operation, the system continuously collects and analyzes collective learning data from all users. For example, through data mining algorithms (such as association rule analysis), it is discovered that many students, after asking a question about node A, will soon review related materials for node B, or frequently choose the wrong answers for A and B during practice. The system can then automatically identify A and B as a high-frequency "cognitive confusion pair" and dynamically increase the weight of the "cognitive confusion probability relationship" edge between these two nodes in the knowledge graph. Furthermore, if the system finds that a "pre-answer hint" (see subsequent steps) generated for a certain knowledge point (such as node C) is frequently triggered by a large number of users and verified as effective (e.g., after a user clicks to view it, the accuracy rate of subsequent related answers increases), the system can determine that the hint targets a common cognitive misconception not initially included in the knowledge graph. At this point, the system can suggest that the course administrator add this cognitive misconception as a new attribute of node C (such as "Common Misconceptions:...") or directly as a new related node to the knowledge graph, thereby making the graph increasingly complete and intelligent.

[0021] Step S2: Monitor the user's entire interaction flow in the learning environment.

[0022] The system records all user interactions with the system throughout the entire process via monitoring modules embedded in various interfaces of the learning platform (such as video players, exercise pages, and Q&A dialog boxes). These interactions are divided into two categories: The first category is explicit Q&A interaction, which is the data stream generated by students actively entering questions in the Q&A box, asking questions by voice, and other explicit requests for help.

[0023] The second category is implicit learning behavior. Students may not be aware of these behaviors, but they profoundly reflect hesitation and conflict in their cognitive processes. Specifically, this includes: 1) entering text in a Q&A box or notes area and then deleting it entirely; the text content before deletion is temporarily recorded and analyzed; 2) repeatedly dragging, jumping, and pausing to view a specific segment while watching a video or reading a document; 3) when taking online self-tests, users not only submit their final answers, but their behavior during the answering process is also recorded, such as selecting and then deselecting an option, marking a question as "uncertain," and making multiple modifications to the answer before submission. All these behaviors, along with timestamps and context (the location of the learning resource), are completely recorded, forming a continuous "behavioral flow."

[0024] Step S3: Generate "pre-answer" prompts based on implicit learning behavior.

[0025] This step is crucial for achieving proactive intervention. The system incorporates a pre-trained behavior-intent inference model. The training method for this model is as follows: First, massive amounts of anonymized, coherent user learning interaction sequence data are collected. This data completely records the entire process of a user's behavior within a learning segment, from exhibiting various implicit behaviors (such as repeated viewing or hesitant marking) to ultimately triggering an explicit behavior (such as asking a specific question or revealing a knowledge error in a subsequent test). Then, educational experts or algorithms annotate these sequence data, establishing association labels between "specific implicit behavior patterns" and "the ultimately exposed cognitive problem type." For example, the behavioral sequence "repeatedly dragging 3 times near time segment T of video P + selecting / deselecting option B twice on a related multiple-choice question" ultimately corresponds to the cognitive problem "confusion regarding concept M." Using this annotated data, a temporal neural network model (such as LSTM or Transformer) is trained, enabling it to predict the type of cognitive difficulty the user may currently face but has not yet explicitly expressed, and its associated knowledge graph nodes, based on the user's recent (e.g., the previous 10 minutes) implicit behavior sequences.

[0026] In practical applications, when the system's real-time monitoring of a user's implicit behavioral sequence triggers the model's prediction threshold—that is, when the model determines with high confidence that the user has a specific confusion—the system will proactively generate a pre-answer prompt. This prompt is not a complete answer, but rather a guiding question (such as "Are you thinking about the difference between X and Y?") or a very concise, targeted explanation (such as a 30-second animation clarifying a common misunderstanding), and is pushed to the user in the form of non-interrupted speech bubbles, sidebar reminders, etc. This is equivalent to the system proactively handing over a "key" when the user is "stuck" but has not yet formed a clear question.

[0027] Step S4: Conduct Q&A interaction and record cognitive state samples.

[0028] When a user engages in explicit Q&A interaction (actively asking a question) or interacts with a system-push "pre-Q&A prompt" (such as clicking to view), the system initiates the formal Q&A process. First, natural language processing technology is used to parse the user's input question or the system's inferred points of confusion, mapping them to one or more specific nodes in a multi-dimensional dynamic knowledge graph. Then, the system retrieves information from the Q&A knowledge base and, combined with the user's current cognitive state (available after model initialization in step S5), generates a Q&A with appropriate difficulty and matching expression. For example, for the same question, the system may provide different levels of explanation and examples of difficulty to a beginner and an advanced user.

[0029] More importantly, the system packages and records the complete context of this interaction as a cognitive state sample. This sample is a structured data package that includes not only the question, answer, and relevant knowledge points, but also details of the user's interaction with the answer (such as which explanation they spent the most time on), the user's subsequent feedback (such as clicking the "Understood" button or asking follow-up questions), and the total time spent from asking the question to understanding. This rich contextual information is a key raw material for building deep cognitive models.

[0030] Step S5: Build and update the user's personalized cognitive state model.

[0031] The goal of this step is to create a dynamic, digital "cognitive mirror" for each student. This model uses a multidimensional dynamic knowledge graph as its framework, but instead of simply labeling things as "learned / unlearned," it assesses the user's multidimensional mastery status for each node in the graph. This status includes at least three dimensions: Mastery: Represents the ability to reproduce knowledge, similar to traditional accuracy, but more complex to calculate.

[0032] Stability: Characterizes the reliability of knowledge transfer and application. A knowledge point with low stability means that students sometimes get it right and sometimes make mistakes, resulting in large fluctuations in performance.

[0033] Easily Confused Related Nodes Set: A personalized list that records the set of other knowledge points that are easily confused with the current node for the user.

[0034] The specific construction method is as follows: For a target node in the knowledge graph, the system extracts a set of features related to it from all historically accumulated cognitive state samples and learning behavior data. For example, features may include: the average depth of all questions a user has asked about that node in the past (whether they are definitional or application questions?), the "feedback response time" between the user receiving the system's answer and the next related action, the variance of scores in different times and different forms of tests involving that node (used to calculate stability), and the frequency with which that node is co-associated with other nodes in the graph in the user's historical errors.

[0035] These features are then fed into a machine learning model (such as a gradient boosting decision tree or neural network), which outputs a multi-dimensional state vector for the node, quantifying the values ​​of the three dimensions mentioned above. The generation of the "easily confused associated node set" integrates the user's individual confusion history and behavioral data, and weights the global "cognitive confusion probability relationship" within the knowledge graph, making predictions more personalized and forward-looking. This model is updated in real-time or periodically with each new learning activity of the user.

[0036] Step S6: Dynamically generate and adjust the personal learning path.

[0037] Based on the latest personalized cognitive state model, the system can plan truly customized learning programs for users. The specific process is as follows: First, the system scans the model to identify nodes that require intervention. Intervention conditions are not singular; for example: 1) mastery level is below a certain threshold (weak knowledge); 2) stability is below a certain threshold (unreliable knowledge, potential risks).

[0038] Next, for each node to be intervened in, the system not only recommends learning content but also the optimal learning modality (such as videos, diagrams, text, and interactive experiments). This recommendation is based on the user's metacognitive characteristic tags. These tags are derived by analyzing the user's long-term historical learning data. For example, if the analysis reveals that watching animated explanations significantly improves the accuracy of related exercises, while reading pure text materials is less effective, the system can then label the user as having a "visual learning preference."

[0039] Finally, the system performs global planning based on the strong dependencies and cognitive confusion relationships between nodes in the multidimensional dynamic knowledge graph. For example, if it needs to remedy node D, but finds that the stability of its prerequisite knowledge node C is also low, the system will package C and D, sort them according to their dependencies, and generate a coherent "linear learning path" from C to D. Simultaneously, for node E, which is easily confused with D, the system may recommend a "networked review path" comparing D and E after the user has learned D. Each link in the path is equipped with micro-course content best suited to the user's modality. These micro-courses are not pre-recorded, but are dynamically generated in real time by the system based on the weak dimensions of the node (such as unclear concepts or unfamiliarity with application), intelligently extracting relevant segments from the standard course resource library, reassembling them according to a cognitive correction logic framework of "presenting misunderstandings -> clarifying concepts -> comparing positive and negative examples," and embedding case studies adapted from the user's own historical mistakes. Therefore, they are highly targeted.

[0040] Step S7: Group analysis and feedback to optimize course design.

[0041] The system can anonymize and analyze the personalized cognitive state models and individual learning path adjustment patterns of all users. Through cluster analysis, it can identify student groups with specific cognitive state characteristics (such as high confusion levels at certain nodes) that have made significant progress after adopting a particular learning path. This path can then be distilled into an effective "learning strategy package." Conversely, if a large number of students consistently exhibit low stability and high confusion at a particular knowledge node, it likely indicates a design challenge in that area of ​​the course. The system can feed these group analysis conclusions back to teachers or course designers, providing data-driven decision support for optimizing teaching content, adjusting knowledge graph structures, and improving teaching methods, thus forming a complete closed loop of "teaching-learning-assessment-optimization."

[0042] Although embodiments of the present invention have been shown and described, these specific embodiments are merely explanations of the invention and are not intended to limit it. The specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. After reading this specification, those skilled in the art may make modifications, substitutions, and variations to the embodiments as needed without departing from the principles and spirit of the invention, but such modifications, substitutions, and variations are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for AI-assisted Q&A and learning progress tracking in courses, characterized in that, Includes the following steps: S1: Construct and maintain a multidimensional dynamic knowledge graph of the target course. The nodes of the multidimensional dynamic knowledge graph include course knowledge points, concepts, skills and common cognitive misconceptions. The edges between nodes include logical relationships, dependency relationships and cognitive confusion probability relationships generated based on statistics. S2: Monitor the user's full-process interaction flow in the learning environment, which includes explicit Q&A interaction and implicit learning behavior; the explicit Q&A interaction is the question-and-answer process actively submitted by the user; the implicit learning behavior includes at least the text content that the user enters and then deletes on the Q&A interface, the behavior of repeatedly jumping to view specific learning resources, and the marking of answer options as uncertain and their modification traces in self-test exercises; S3: Based on the implicit learning behavior, a pre-trained behavior-intention inference model is used to predict the user's potential knowledge confusion or cognitive conflict, and pre-answer prompts are generated and proactively pushed to the user; The pre-answer prompts are guiding questions or targeted explanation segments used to clarify inferred cognitive conflicts; S4: For explicit Q&A interactions and interactions that trigger pre-Q&A prompts, the parsed questions are mapped and matched with the multi-dimensional dynamic knowledge graph to generate Q&A content that is adapted to the user's current cognitive state; at the same time, the complete context information of this interaction is recorded as a cognitive state sample. S5: Integrate all cognitive state samples with the user's historical learning behavior data to construct and continuously update the user's personalized cognitive state model; the personalized cognitive state model is used to characterize the user's multidimensional mastery state of each node in the multidimensional dynamic knowledge graph, and the multidimensional mastery state includes at least mastery degree, stability, and easily confused related node set; S6: Based on the personalized cognitive state model, dynamically generate and adjust the user's personal learning path; the personal learning path includes a sequence of knowledge nodes to be learned and reviewed, the best learning mode recommended for each node, and micro-course content tailored to make up for specific weaknesses.

2. The course AI-assisted Q&A and learning progress tracking method according to claim 1, characterized in that, The steps for maintaining the multidimensional dynamic knowledge graph in step S1 include: S11: Initially, the multidimensional dynamic knowledge graph is constructed based on the course syllabus and textbooks; S12: Continuously analyze the Q&A interaction records and implicit learning behavior data of all users, automatically identify frequently co-occurring cognitive confusion knowledge point pairs, and dynamically update the cognitive confusion probability relationship between corresponding nodes in the knowledge graph; S13: When a pre-answer prompt for a certain knowledge point is frequently triggered by the user and confirmed to be effective, the cognitive misconception targeted by the pre-answer prompt is updated to the multidimensional dynamic knowledge graph as a new node or node attribute.

3. The course AI-assisted Q&A and learning progress tracking method according to claim 1, characterized in that, The training method for the behavior-intent inference model described in step S3 includes: S31: Collect massive amounts of anonymized user learning interaction sequence data, which includes the complete process from implicit learning behavior to finally explicitly asking questions or experiencing a shift in understanding; S32: Label the user learning interaction sequence data and establish a correlation between the implicit learning behavior sequence of a specific pattern and the knowledge point confusion type that is subsequently exposed explicitly; S33: Based on labeled sequence data, train a temporal neural network model so that it can predict the types of cognitive difficulties that users will face or already have but have not explicitly stated, and their related knowledge points, according to the input sequence of recent implicit learning behaviors.

4. The course AI-assisted Q&A and learning progress tracking method according to claim 1, characterized in that, The personalized cognitive state model mentioned in step S5 specifically includes: S51: For each target node in the multidimensional dynamic knowledge graph, extract a feature set related to the target node from all cognitive state samples; the feature set includes the angle and depth of the user's questions about the node, the response time after receiving the answer, the stability of the performance in subsequent related self-tests, and the frequency of association with easily confused nodes; S52: Using the feature set, calculate the multidimensional state vector of the target node through a machine learning model; the mastery represents the knowledge reproduction ability, the stability represents the reliability of knowledge transfer in different contexts, and the easily confused associated node set is generated by weighting the user's personal historical confusion records and the group cognitive confusion probability relationship in the knowledge graph.

5. The course AI-assisted Q&A and learning progress tracking method according to claim 1, characterized in that, The dynamic generation and adjustment of the personal learning path in step S6 specifically includes: S61: Based on the personalized cognitive state model, identify the nodes to be intervened that meet any of the following conditions: mastery is lower than a first preset threshold, or stability is lower than a second preset threshold. S62: For each node to be intervened, the best learning modality and micro-course content are matched from the learning resource library by combining the user's metacognitive feature tags; the metacognitive feature tags are obtained by analyzing the correlation data between the user's historical learning performance and different learning modalities, and are used to identify the user's preferred information reception and processing style. S63: Based on the dependency relationships and cognitive confusion probability relationships in the multidimensional dynamic knowledge graph, the selected nodes to be intervened and the matching learning resources are topologically sorted and conflict-resolved to generate logically coherent linear learning paths and network review paths with optimized cognitive load.

6. The course AI-assisted Q&A and learning progress tracking method according to claim 5, characterized in that, The method for generating micro-course content in step S62 is as follows: for the specific weak dimensions of the node to be intervened, relevant knowledge point fragments are extracted from standard course resources, and reorganized and connected according to the preset cognitive correction logic framework. At the same time, positive and negative cases adapted from the user's high-frequency errors or points of confusion are embedded, and targeted explanation units are synthesized in real time by programming.

7. The course AI-assisted Q&A and learning progress tracking method according to claim 1, characterized in that, It also includes step S7: An anonymous cognitive state network is established across users. By analyzing student groups with similar personalized cognitive state models or similar personal learning path adjustment patterns, potential effective learning strategies or common teaching difficulties can be discovered. The analysis results are then fed back to curriculum designers to optimize course content and multidimensional dynamic knowledge graph structures.

8. The course AI-assisted Q&A and learning progress tracking method according to claim 1, characterized in that, When generating Q&A content in step S4, a multi-turn dialogue format is adopted, and the real-time changes in the user's cognitive state are dynamically evaluated during the dialogue. The depth, breadth, and expression of subsequent dialogue content are adjusted based on the evaluation results.

Images