Method and system for generating personalized learning paths based on teaching intentions and learning profile

Abstract

This invention provides a method and system for generating personalized academic paths based on teaching intentions and student learning profiles, belonging to the field of digital teaching technology. The method collects multimodal teaching data and reconstructs hierarchical teaching intentions using a large language model; it simultaneously collects multimodal perception data from students, calculates a transmission consistency coefficient to identify intention transmission failures, and corrects false weaknesses in the student learning profile; it constructs a dynamic student learning profile based on knowledge tracking and forgetting factors, including the probability of cognitive mastery, a weak knowledge graph, and learning preferences; when calculating the difference in achievement of knowledge node goals, it superimposes a cognitive dimension gap term, a prerequisite chain effect term, and an intention transmission compensation term, and uses a psychological adaptation coefficient generated by real-time cognitive load and learning emotions to adaptively correct the difference; with the joint goal of maximizing cognitive growth and psychological recovery, it dynamically plans a personalized academic path including intention reconstruction steps and psychological adjustment nodes.

Description

Technical Field

[0001] This invention relates to the field of digital teaching technology, and in particular to a method and system for generating personalized learning paths based on teaching intentions and student profiles. Background Technology

[0002] With the deep penetration of artificial intelligence technology into education, smart education systems centered on knowledge tracking, recommendation algorithms, and learning analytics are gradually evolving from simple resource matching to personalized academic path planning. However, in actual teaching scenarios, truly effective "personalized instruction" still faces many key technological bottlenecks: In existing technologies, the generation of personalized learning paths mostly relies on static diagnostics or standardized assessment results. These methods typically analyze students' historical answer data, use knowledge tracing models (such as DKT, DKVMN, etc.) to obtain the mastery probability of each knowledge point, and then recommend learning resource sequences based on the upstream and downstream relationships of the knowledge graph, with the shortest path or maximum coverage as the goal.

[0003] For example, some solutions use collaborative filtering or content-based recommendation algorithms to push exercises and explanations related to students' weak knowledge points. These methods have obvious drawbacks: when constructing the starting point diagnosis, they are completely detached from the perception of the dynamic process of actual classroom teaching, and cannot understand the teacher's current teaching intentions and teaching focus. This results in a disconnect between the generated path and classroom teaching, and fails to effectively connect "teaching" and "learning".

[0004] To address the aforementioned disconnect, some research has begun to explore the identification of teaching intentions. Some approaches utilize natural language processing (NLP) technology to classify the intentions behind teachers' classroom speech transcriptions, for example, by using pre-trained language models to identify basic teaching behaviors such as "asking questions," "explaining," and "summarizing."

[0005] However, these solutions rely solely on single-modal text data, ignoring non-verbal signals rich in pedagogical semantics, such as blackboard emphasis, body language, and courseware switching. This results in coarse granularity of intent recognition, limited accuracy, and an inability to recreate the teacher's complex hierarchical teaching logic.

[0006] More importantly, current teaching intent recognition only treats teacher behavior as a one-way input, failing to perceive and verify students' real-time reception status. This makes the construction of student profiles and path generation based on an "ideal transmission assumption," which assumes that the teaching intent expressed by the teacher has been fully and effectively received by the students. In practice, the problem of intent transmission failure due to students' attention wandering, information overload, or confusion is common. However, existing technologies cannot distinguish between "students not learning" and "students not receiving instruction," often conflating the two as purely knowledge acquisition deficiencies, thus generating ineffective or even erroneous remedial paths.

[0007] Furthermore, current mainstream student profiling generally focuses on cognitive dimensions when depicting student status, paying attention to knowledge mastery, ability level, and distribution of weaknesses, while neglecting the dynamic and changing non-cognitive psychological factors during the learning process. Existing research shows that learners' cognitive load and learning emotions have a significant impact on knowledge reception, processing, and transfer. When students are under high cognitive load or in a negative emotional state, forcibly pushing highly difficult reinforcement tasks not only fails to achieve learning outcomes but may also exacerbate cognitive overload and learning burnout. However, most existing adaptive learning systems exclude psychological state from the core decision variables of path planning or treat it only as an auxiliary label in the system. This results in the generated learning path having theoretical rationality at the cognitive level but lacking accessibility at the psychological execution level, ultimately leading to the failure of personalized recommendations.

[0008] In summary, there is an urgent need for a method and system for generating personalized learning paths based on teaching intentions and student profiles to address the three technical problems of inaccurate analysis of teaching intentions, limited dimensions of student profiles, and lack of psychological adaptability in path generation. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for generating personalized learning paths based on teaching intentions and learning profiles. In the field of smart education, this invention solves the long-standing technical problems of inaccurate interpretation of teaching intentions, single-dimensional learning profiles, and path generation that is out of touch with actual reception status and psychological tolerance. It realizes a fully adaptive closed loop of "teaching-transmission-learning-assessment-psychology", improves the accuracy, acceptability and psychological accessibility of personalized learning paths, and promotes the transformation of classrooms from "teaching-centered" to "learning-centered".

[0010] To achieve the above objectives, this application proposes a method for generating personalized learning paths based on teaching intentions and student learning profiles, including: The steps for dynamically reconstructing teaching intentions are as follows: acquire structured teaching data in real time, analyze teachers' teaching behaviors using a large language model, and dynamically identify and reconstruct the teaching intentions corresponding to the current teaching segment. Steps for constructing a multidimensional dynamic learning profile: Collect learning data generated by students in classroom interactions and after-class assessments, combine it with the teaching intentions, extract students' weak points in knowledge mastery, and generate a dynamic learning profile that includes cognitive level, learning preferences, and a weak knowledge map; Path difference calculation steps: Compare the dynamic learning profile with the current teaching objectives to calculate the difference in students' achievement of objectives at each knowledge node; Personalized academic path and resource generation steps: Based on the degree of difference in achieving the stated goals, the generation capabilities of the large language model are used to generate customized learning resources, and a personalized academic path adapted to the student is dynamically planned according to the logic of remediation-reinforcement-expansion.

[0011] As a further solution, the dynamic reconstruction step of the teaching intent further includes: Acquire multimodal teaching data, which includes teacher's voice stream, blackboard writing and courseware screenshot sequences, key points of body movements, and interactive whiteboard operation events; The multimodal teaching data within the same time slice are aligned and jointly semantically encoded to form a behavioral semantic stream; The behavioral semantic stream is segmented and identified using a first model based on a temporal classifier to obtain a sequence of basic intent units. The basic intent unit sequence is semantically reconstructed using a large language model to generate hierarchical teaching intents. The hierarchical teaching intent includes intent nodes, hierarchical relationships between nodes, and target knowledge nodes and cognitive levels corresponding to the intent nodes.

[0012] As a further solution, the steps for constructing the multi-dimensional dynamic learning profile further include: Based on the knowledge tracking model, a multidimensional cognitive mastery probability matrix covering multiple knowledge points is maintained and dynamically updated for each student; A personalized forgetting curve decay factor is introduced to correct the forgetting perception of the multidimensional cognitive mastery probability matrix. Based on the corrected mastery probability, knowledge points with mastery probabilities below a preset threshold are extracted to form weak knowledge nodes. The influence of weak associations is then calculated using a graph neural network in the subject knowledge graph to generate a weak knowledge graph. Students' learning style preferences are extracted by clustering learning behavior data, and these preferences include information reception preferences and concentration period distribution. The dynamic learning profile is composed of the cognitive mastery probability matrix, the weak knowledge graph, and the learning style preference.

[0013] As a further solution, the method also includes an intent delivery consistency detection and compensation step: Collect student-side multimodal perception data synchronized with the teaching behavior time. The student-side multimodal perception data includes at least face orientation, gaze heat area, micro-expressions, and interactive marking behavior. Using a student state recognition model, based on the multimodal perception data from the student's end, the student's perception intention state within each time window is identified in real time. The perception intention state includes at least: received, received but confused, and attention wandering. By comparing the hierarchical teaching intentions within the same time segment with the distribution of the students' perceived intentions, a transmission consistency coefficient representing the degree of matching between the teacher's teaching intentions and the students' actual perceptions is calculated. Using the aforementioned transmission consistency coefficient, the mastery probability or weakness marker of the corresponding knowledge node in the multidimensional dynamic learning profile is corrected to distinguish between weaknesses caused by cognitive deficiencies and false weaknesses caused by insufficient intention transmission.

[0014] As a further solution, the path difference calculation step further includes: When calculating the target achievement difference of the target knowledge node, an intent transmission compensation term constructed based on the transmission consistency coefficient is added; The difference calculation formula includes a cognitive dimension gap term, a prior knowledge chain influence term, and an intent transmission compensation term, so that nodes with low transmission consistency have a difference weight that is prioritized for repair.

[0015] As a further solution, the personalized academic path and resource generation steps further include: When the transmission consistency coefficient of a knowledge node is lower than the preset consistency threshold, before the logic of reinforcement-strengthening-expansion, the intention reconstruction step is planned for the node first. The intention reconstruction step forcibly pushes the intention reconstruction learning resources configured to be re-explained and multi-representation interaction. The intent reconstruction step is used as a prerequisite for entering the corresponding knowledge node reinforcement training, and a secondary evaluation is performed after the user completes the intent reconstruction in order to eliminate the impact of intent transmission failure.

[0016] As a further solution, the method also includes a psychological state adaptive adjustment step: The cognitive load and learning emotional state of students are detected in real time. The cognitive load is classified based on task completion time and frequency of operational hesitation. The learning emotional state is divided into valence-arousal quadrants based on the fusion recognition of facial expressions and interactive behaviors. Based on the cognitive load state and learning emotional state, an instantaneous psychological state vector is generated, and a psychological fit coefficient is calculated to characterize the degree of deviation from the preset ideal learning state. Using the psychological adaptation coefficient, the difference obtained from the path difference calculation step is adaptively corrected, and / or, a psychological buffer layer is dynamically inserted or activated in the personalized academic path, the psychological buffer layer containing indirect knowledge learning activities for reducing cognitive load or regulating emotions.

[0017] As a further solution, the psychological fit coefficient is used to adaptively correct the degree of difference, specifically as follows: The original difference degree is multiplicatively adjusted by using the attention effectiveness coefficient with psychological fit coefficient as the variable, and a stress compensation term generated by the difficulty of knowledge points and the history of repeated errors is added, so that the difference execution priority of high-difficulty knowledge points is dynamically reduced under high load or negative emotional state.

[0018] As a further solution, the personalized academic path and resource generation steps further include: The original cognitive goal achievement is transformed into a multi-objective optimization problem that maximizes cognitive growth and maximizes psychological resilience. Real-time psychological constraints are added during path solving. These constraints limit the proportion of nodes with high cognitive load within any continuous time window and force the insertion of an emotion regulation node when a negative high arousal state is detected. During the path execution process, the real-time psychological state is continuously monitored. When the psychological fit coefficient deviates from the fit zone, the subsequent high-load nodes to be executed are dynamically deleted or postponed, and preset cognitive load reduction or emotion regulation activities are inserted.

[0019] On the other hand, the present invention also provides a personalized academic path generation system based on teaching intentions and learning profiles, used to implement the personalized academic path generation method based on teaching intentions and learning profiles as described in any of the preceding claims, including: The module for dynamically reconstructing teaching intentions: acquires structured teaching data in real time, analyzes teachers' teaching behaviors using a large language model, and dynamically identifies and reconstructs the teaching intentions corresponding to the current teaching segment; Multidimensional dynamic learning profile construction module: Collects learning data generated by students in classroom interaction and after-class assessment, combines it with the teaching intentions, extracts students' weak points in knowledge mastery, and generates a dynamic learning profile that includes cognitive level, learning preferences and weak knowledge map; Path Difference Calculation Module: Compares the dynamic learning profile with the current teaching objectives to calculate the difference in students' achievement of objectives at each knowledge node; Personalized academic path and resource generation module: Based on the degree of difference in goal achievement, it uses the generation capabilities of the large language model to generate customized learning resources, and dynamically plans a personalized academic path that suits the student according to the logic of remediation-reinforcement-expansion.

[0020] Compared with related technologies, the personalized learning path generation method and system based on teaching intentions and learning profiles provided by this invention have the following advantages: 1. This invention collects multimodal data including teacher speech, blackboard writing, body language, and whiteboard operations. Utilizing joint semantic coding and temporal segmentation models, and incorporating a large language model for logical reconstruction of intent sequences, it generates hierarchical teaching intents labeled with target knowledge nodes and cognitive levels. This mechanism significantly overcomes the limitations of existing technologies that rely solely on coarse-grained classification of speech text, improving the accuracy of teaching intent recognition. Furthermore, it accurately recreates the complex teaching logic of "introduction—construction—clarification—verification," providing a dynamic target benchmark for path generation that resonates with real classroom teaching.

[0021] 2. This invention synchronously collects multimodal perception data from students, calculates in real time the consistency coefficient representing the matching degree between teacher intent and student perception, and uses this coefficient to correct the weakness assessment in the student learning profile, introducing an intent transmission compensation term into the difference calculation. This design breaks the default "ideal transmission assumption" of existing technologies for the first time, accurately distinguishing between "weaknesses caused by cognitive deficiencies" and "pseudo-weaknesses caused by insufficient intent transmission," and forcibly inserts an intent reconstruction step into the path planning to repair information reception gaps. This eliminates the source of ineffective remedial paths, ensuring that learning resources are accurately projected onto real needs, and significantly enhancing the path's relevance.

[0022] 3. This invention quantifies the real-time detected cognitive load level and learning emotion quadrant into a psychological fit coefficient. This coefficient is not only used for multiplicative adjustment of differences and stress compensation correction, but also establishes a multi-objective optimization model in path planning with cognitive growth and psychological recovery as dual objectives. This enables the generated learning path to intelligently avoid the risk of cognitive overload for students under high load or negative emotional states, and proactively inserts emotion regulation nodes for psychological empowerment, solving the path failure problem in existing technologies where students are "ability-suited but psychologically resistant." Students can continue learning within an appropriate psychological bandwidth, resulting in a significant improvement in both path completion rate and knowledge attainment rate.

[0023] 4. This invention organically integrates intent reconstruction, transmission detection, cognitive diagnosis, psychological assessment, and path generation into a collaborative closed loop, and designs a two-way feedback mechanism, enabling the system to continuously optimize the teaching intent model, learning assessment model, and resource generation strategy at both the individual and group levels. The entire method has strong interdisciplinary generalization ability and high scenario adaptability. The large language model is upgraded from a passive auxiliary tool to an intelligent agent that actively predicts, intervenes, and reflects, promoting the transformation of the classroom from "teacher-centered" to a personalized learning ecosystem where "teaching follows learning and learning is at ease." Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic diagram illustrating the steps of a method for generating a personalized academic path based on teaching intentions and student learning profiles provided by this invention; Figure 2 A schematic diagram of the structure of a personalized academic path generation system based on teaching intentions and student learning profiles provided by this invention; The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] 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 invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0028] The core concept of this invention lies in upgrading the capabilities of large language models from passive assistance to an active empowerment engine for the entire teaching process. It constructs an adaptive closed loop of "dynamic reconstruction of teaching intent - multi-dimensional learning profiling - path difference calculation - personalized path generation," and addresses the problems of neglecting teaching transmission failure and learners' non-cognitive psychological states in existing technologies by introducing intent transmission consistency compensation and adaptive adjustment of psychological states, thereby achieving high accuracy. Highly Adaptive Personalized Academic Path Planning Example 1 Please see Figure 1 This embodiment provides a method for generating academic paths that integrates multimodal teaching data analysis and cognitive diagnostic models, including the following steps S1 to S4. Step S1, Dynamic Reconstruction of Teaching Intent: Real-time acquisition of structured teaching data, analysis of teachers' teaching behavior using a large language model, dynamic identification and reconstruction of the teaching intent corresponding to the current teaching segment; Real-time acquisition of multimodal teaching data during teacher instruction, including but not limited to: teacher's voice stream acquired through a microphone array. A sequence of screenshots of whiteboard and courseware obtained through a screenshot service. Key points of teacher body movements (such as gestures) acquired through visual sensors The blackboard emphasizes actions and interactive whiteboard operation events. All data is aligned to the millisecond level using a unified timestamp. The aligned multimodal data is fed into a pre-trained joint encoding model to generate a fixed-dimensional sequence of behavioral semantic vectors, forming a behavioral semantic stream. Subsequently, a Transformer-based temporal classifier was used to segment and initially identify the behavioral semantic stream, resulting in a sequence of basic intent units, such as [“Review Questions”, “Blackboard Definition”, “Example Explanation”]. Next, the large language model is invoked, and combined with the current course knowledge graph, the basic intent unit sequence is reconstructed into hierarchical teaching intents. For example, the sequence ["Context Introduction", "Introducing the Formula", "Emphasis on the Blackboard"] can be reconstructed into the intention node "Key Points Explained", with its parent intention being "New Knowledge Instruction", the associated target knowledge node being "Pythagorean Theorem", and the expected cognitive level being "Understanding". This hierarchical teaching intention is represented by a node relationship. Target knowledge node set And the expected degree vector Structured data Step S2, Multi-dimensional Dynamic Learning Profile Construction Steps: Collect learning data generated by students in classroom interaction and after-class assessment, combine it with the teaching intentions, extract students' weak points in knowledge mastery, and generate a dynamic learning profile that includes cognitive level, learning preferences and weak knowledge map; The process of constructing a multi-dimensional dynamic learning profile for each student is as follows: First, a dynamic, multi-dimensional cognitive mastery probability matrix is ​​maintained for each student based on the knowledge tracing model. Where N is the total number of knowledge points, and K is the number of cognitive dimensions (e.g., memory). understand application analyze) Whenever students participate in class quizzes AI provides answers to questions or completes post-class assessments, and updates the mastery probability of corresponding knowledge points using a Dynamic Key-Value Memory Network (DKVMN) based on the answers. Simultaneously, a personalized forgetting curve decay factor is introduced: for each knowledge point, the time interval since the last learning is recorded, and the decay coefficient is calculated based on the Ebbinghaus forgetting curve formula to correct the mastery probability, thus obtaining the perceived mastery probability of forgetting. .

[0029] Secondly, knowledge nodes with a current mastery probability below a threshold (e.g., 0.6) are extracted as weak nodes, and a graph neural network is used in the subject knowledge graph to calculate the weak influence coefficient between each node. Generate a weak knowledge graph containing directly weak nodes and related weak impact subgraphs. Finally, the students' learning behavior data were clustered to extract learning style preferences, including information reception preferences (e.g., visual / textual) and attention span distribution, and compared with the aforementioned mastery probability. Weak knowledge graphs together constitute a dynamic learning profile. Step S3, Path Difference Calculation Step: Compare the dynamic learning profile with the current teaching objectives to calculate the difference in students' achievement of objectives at each knowledge node; The teaching objective status parsed in step S1 is compared with the student dynamic learning profile generated in step S2, and the objective achievement difference of each knowledge node is calculated. The following basic difference calculation formula is used in this embodiment: in, To achieve the expected level of mastery in teaching, For students to currently master probability, As preset cognitive dimension weights, Let k be the set of prerequisite knowledge nodes. Here, λ is the influence coefficient, and λ is the prior influence balance coefficient. This difference comprehensively reflects the cascading effects of cognitive gaps and weak prior knowledge. Based on the calculated degree of difference, knowledge nodes are divided into the basic reinforcement layer. The capability enhancement layer and the mindset expansion layer provide a hierarchical basis for path generation. Step S4, Personalized Academic Path and Resource Generation Step: Based on the difference in the achievement of the goals, use the generation capability of the large language model to generate customized learning resources, and dynamically plan a personalized academic path that suits the student according to the logic of remediation-reinforcement-expansion. The difference information of each node is converted into a natural language requirement description, which is then input into a large language model to generate customized learning resources, including targeted analysis micro-lessons. Variation exercises and inquiry tasks, etc. At the same time, semantic matching is used to retrieve and integrate existing high-quality resources from the school-based resource database. The generated / retrieved resources are scored using a difficulty assessment model based on item response theory, and candidate resources that match the students' zone of proximal development are selected. Finally, with the goal of "making up for the most weaknesses and achieving the expansion target in the shortest time", the focus is on "making up for weaknesses". strengthen Using the order of "expansion" as a constraint, the optimal resource sequence is solved through integer programming to form a personalized academic path and generate a corresponding learning guide. Example 2 In the scheme of Example 1, although the four links of dynamic reconstruction of teaching intention, multi-dimensional learning profile, difference calculation and path generation form a closed loop, there is a key problem: the reconstruction of teaching intention is based only on the analysis of teacher behavior, while ignoring the students' actual reception of these intentions, which leads to the learning profile and path generation being based on an "ideal transmission assumption".

[0030] Specifically, in step S1, the large model reconstructs the teacher's intention to "clarify key points," but if students fail to effectively receive this intention due to inattentiveness, unresolved confusion, or information overload, their weaknesses in subsequent assessments may be misjudged as cognitive deficiencies rather than a failure to convey the intention. This causes the difference calculation in step S3 and the path generation in step S4 to fail to distinguish between "students not learning" and "students not receiving instruction," resulting in generated remedial paths that may repeat ineffective and inefficient stimuli or ignore the "intent breakpoints" that need to be prioritized for repair.

[0031] Therefore, this embodiment further improves upon embodiment 1 by incorporating a method for detecting and compensating for consistency in intent transmission. It adds an intent transmission consistency detection and compensation mechanism to solve the problem of "false weakness" caused by the mismatch between the teacher's intent and the student's receiving state, making the learning profile and path generation closer to the real learning process.

[0032] While step S1 is being executed, multimodal perception data from the student's device is being collected simultaneously, including at least facial orientation. visual hotspots Micro-expressions (such as nodding) Frowning and actively marking points of doubt Using a pre-trained CNN-Transformer-based attention state recognition model, the system identifies the student's perceived intention state in real time within each time window, classifying the state into at least "received". "Received but confused" Categories such as "Caution: Freezing" For each time segment, the hierarchical teaching intentions of teachers are compared with the distribution of perceived intentions of the student group, and the transitivity coefficient η, representing the degree of matching, is calculated. For example, when the teacher's intention is "to clarify key points", if more than 40% of the students are in the state of "receiving but confused", then the value of η for this intention node is 0.3 (range 0-1).

[0033] In step S2, when updating the learning profile, the weakness marking rules are corrected using the transmission consistency coefficient: for knowledge nodes where η is below a threshold (e.g., 0.5), if a student answers incorrectly in a subsequent assessment, the confidence level for that knowledge node as purely "insufficient ability" is reduced, and it is instead marked as "false weakness caused by insufficient intention transmission," and the duration of confusion is correlated. In step S3, when calculating the path difference, an intent delivery compensation term is introduced, and the difference formula is modified as follows: in, Let k be the transitive consistency coefficient. μ is the preset compensation amount for the underestimation of mastery that may result from insufficient transmission, and μ is the transmission compensation weight. This formula automatically increases the difference degree of nodes with low transmission consistency, prioritizing the triggering of repair. When generating a personalized academic path in step S4, when a certain node... If the value is below a threshold, a pre-defined reconstruction step is forced to be inserted before "weakness repair". This step is configured to force a re-explanation. Use multiple representation methods (such as animation) The metaphorical approach aims to reconstruct resources and makes completing this step a prerequisite for entering the corresponding knowledge node remedial training. After completing the intent reconstruction, students need to pass a quick confirmation assessment. If the recovery is accepted, the compensation item will be removed from the subsequent path. Furthermore, in the closed-loop feedback of teaching, when multiple high-intent breakpoints appear in the class as a whole, the system automatically pushes teaching warnings to teachers and generates improvement suggestions such as "adding instant voting to confirm understanding" and "replacing abstract expressions with real-life examples," thereby achieving continuous optimization of the teaching intent model. Example 3 Based on the innovative improvement scheme of Example 1, after a thorough examination of its technical details, another key issue affecting the accuracy of personalized paths can be found: the learning profile and path generation overemphasize cognitive ability shortcomings and ignore students' real-time non-cognitive psychological states during the learning process, especially dynamic cognitive load and learning emotions, resulting in blind spots in path adaptability at the "executability" level.

[0034] Specifically, in step S2, although the system collects emotional signals such as students' facial expressions and focus during class, this information is only stored in the profile as static preferences and is not involved in the difference calculation in step S3 or the path planning in step S4. In step S4, the optimization goal of the path is only to "make up for the most weaknesses and achieve the extension goal in the shortest time." This easily generates high-density and high-difficulty reinforcement paths, ignoring that students may already be under high cognitive load or negative emotions. Forcing the path forward will only cause thinking inhibition and learning fatigue, forming a failed path that is "suitable in ability but rejected psychologically." This split between cognition and emotion prevents the adaptive system from truly achieving "achievable" personalization.

[0035] Therefore, it is necessary to introduce cognitive load perception and emotional adaptive adjustment mechanisms into the existing technical framework, so that the learning profile dimension can be extended from cognition to psychology, and these can be injected into the path calculation in real time as dynamic constraints to achieve "cognition-psychology" dual-driven adaptation.

[0036] This embodiment, based on embodiment 1 or 2, integrates an adaptive adjustment method that combines cognitive load and learning emotion. It adds a psychological state adaptive adjustment step that integrates cognitive load and learning emotion, further addressing the problem of path planning neglecting students' non-cognitive psychological states. During the student's learning process, their cognitive load and learning emotional state are monitored in real time. Cognitive load status is based on the current task completion time. Frequency of operational hesitation Error types and other behavioral indicators are categorized into three levels—"Low-load scalable," "Medium-load adaptability," and "High-load requiring channeling"—using a pre-trained SVM classifier. The learning of emotional states is based on the fusion of facial expressions and interactive behaviors, and is divided into four quadrants: "positive high arousal," "negative high arousal," "positive low arousal," and "negative low arousal." Based on both, an instantaneous mental state vector is generated, and the mental fit coefficient is calculated. (0 to 1, where 1 represents the optimal state) After obtaining the basic difference Δk in step S3, the psychological fit coefficient is used for dynamic correction to obtain the psychological fit difference. : In the formula, For example, take the attention efficiency coefficient. This temporarily reduces the urgency of the cognitive gap under adverse conditions; β is the stress compensation value for node k, mapped based on the node's historical error count; β is the stress penalty weight. At the same time, a psychological buffer layer has been added to the recent developmental differentiation layer. When high load or negative emotions are detected, some reinforcement nodes will be moved to this layer and temporarily not pushed. In the path planning stage of step S4, the original single-objective optimization is upgraded to cognitive-psychological dual-objective optimization. The objective function for path solving is: Where P represents the candidate path. For node knowledge increment, For task cognitive load cost, For the psychological recovery benefits of the emotional regulation node, , , Preset weights The optimization constraints include: the proportion of high-load nodes does not exceed 30% within any continuous time window, and a forced insertion of a breathing-style cleanup is required when a negative high-wake-up state is detected. Emotional regulation points such as achievement review During path execution, psychological state is monitored in real time; if the psychological fit coefficient... If the learning curve deviates from the appropriate range, subsequent high-load nodes will be dynamically deleted or postponed, and pre-set cognitive load-reducing activities (such as mind map organization) will be automatically inserted to achieve flexible and softened learning paths, ensuring that students are always within a manageable range. Willing to participate in learning within the psychological bandwidth It should be noted that Examples 2 and 3 can be implemented in combination, that is, simultaneously including consistency detection of intent transmission and cognitive load / emotion regulation, forming a full-link adaptive system of "teaching-transmission-learning-assessment-psychology", with all calculation factors uniformly integrated in the difference formula. Example 4 Please see Figure 2 Corresponding to the system embodiment of Embodiment 1 above, the present invention also provides a personalized academic path generation system based on teaching intention recognition and dynamic learning profile, comprising: Module 10 for Dynamic Reconstruction of Teaching Intent: Real-time acquisition of structured teaching data, analysis of teachers' teaching behavior using a large language model, dynamic identification and reconstruction of the teaching intent corresponding to the current teaching segment; Multidimensional dynamic learning profile construction module 20: Collects learning data generated by students in classroom interaction and after-class assessment, combines it with the teaching intention, extracts students' weak points in knowledge mastery, and generates a dynamic learning profile that includes cognitive level, learning preferences and weak knowledge map; Path Difference Calculation Module 30: Compares the dynamic learning profile with the current teaching objectives and calculates the difference in students' achievement of objectives at each knowledge node; Personalized academic path and resource generation module 40: Based on the difference in the achievement of the stated goals, it uses the generation capabilities of the large language model to generate customized learning resources, and dynamically plans a personalized academic path that suits the student according to the logic of remediation-reinforcement-expansion.

[0037] The data flow and signal transmission relationships between the various modules, as described above, constitute a complete technical system. The system can be deployed on cloud servers or edge computing nodes, enabling real-time interaction through sensing devices in the classroom and student terminals. The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for generating an individual academic path based on teaching intentions and a profile of learning conditions, characterized by, include: The steps for dynamically reconstructing teaching intentions are as follows: acquire structured teaching data in real time, analyze teachers' teaching behaviors using a large language model, and dynamically identify and reconstruct the teaching intentions corresponding to the current teaching segment. Steps for constructing a multidimensional dynamic learning profile: Collect learning data generated by students in classroom interactions and after-class assessments, combine it with the teaching intentions, extract students' weak points in knowledge mastery, and generate a dynamic learning profile that includes cognitive level, learning preferences, and a weak knowledge map; Path difference calculation steps: Compare the dynamic learning profile with the current teaching objectives to calculate the difference in students' achievement of objectives at each knowledge node; Personalized academic path and resource generation steps: Based on the degree of difference in achieving the stated goals, the generation capabilities of the large language model are used to generate customized learning resources, and a personalized academic path adapted to the student is dynamically planned according to the logic of remediation-reinforcement-expansion. 2.The method of claim 1, wherein, The dynamic reconstruction step of the teaching intent further includes: Acquire multimodal teaching data, which includes teacher's voice stream, blackboard writing and courseware screenshot sequences, key points of body movements, and interactive whiteboard operation events; The multimodal teaching data within the same time slice are aligned and jointly semantically encoded to form a behavioral semantic stream; The behavioral semantic stream is segmented and identified using a first model based on a temporal classifier to obtain a sequence of basic intent units. The basic intent unit sequence is semantically reconstructed using a large language model to generate hierarchical teaching intents. The hierarchical teaching intent includes intent nodes, hierarchical relationships between nodes, and target knowledge nodes and cognitive levels corresponding to the intent nodes. 3.The method of claim 1, wherein, The steps for constructing the multi-dimensional dynamic learning profile further include: Based on the knowledge tracking model, a multidimensional cognitive mastery probability matrix covering multiple knowledge points is maintained and dynamically updated for each student; A personalized forgetting curve decay factor is introduced to correct the forgetting perception of the multidimensional cognitive mastery probability matrix. Based on the corrected mastery probability, knowledge points with mastery probabilities below a preset threshold are extracted to form weak knowledge nodes. The influence of weak associations is then calculated using a graph neural network in the subject knowledge graph to generate a weak knowledge graph. Students' learning style preferences are extracted by clustering learning behavior data, and these preferences include information reception preferences and concentration period distribution. The dynamic learning profile is composed of the cognitive mastery probability matrix, the weak knowledge graph, and the learning style preference.

4. A method for generating personalized academic paths based on teaching intentions and student learning profiles according to claim 2 or 3, characterized in that, The method also includes an intent transmission consistency detection and compensation step: Collect student-side multimodal perception data synchronized with the teaching behavior time. The student-side multimodal perception data includes at least face orientation, gaze heat area, micro-expressions, and interactive marking behavior. Using a student state recognition model, based on the multimodal perception data from the student's end, the student's perception intention state within each time window is identified in real time. The perception intention state includes at least: received, received but confused, and attention wandering. By comparing the hierarchical teaching intentions within the same time segment with the distribution of the students' perceived intentions, a transmission consistency coefficient representing the degree of matching between the teacher's teaching intentions and the students' actual perceptions is calculated. Using the aforementioned transmission consistency coefficient, the mastery probability or weakness marker of the corresponding knowledge node in the multidimensional dynamic learning profile is corrected to distinguish between weaknesses caused by cognitive deficiencies and false weaknesses caused by insufficient intention transmission.

5. The method for generating personalized learning paths based on teaching intentions and learning profiles according to claim 4, characterized in that, The path difference calculation step further includes: When calculating the target achievement difference of the target knowledge node, an intent transmission compensation term constructed based on the transmission consistency coefficient is added; The difference calculation formula includes a cognitive dimension gap term, a prior knowledge chain influence term, and an intent transmission compensation term, so that nodes with low transmission consistency have a difference weight that is prioritized for repair.

6. The method for generating personalized learning paths based on teaching intentions and learning profiles according to claim 4, characterized in that, The personalized academic path and resource generation steps further include: When the transmission consistency coefficient of a knowledge node is lower than the preset consistency threshold, before the logic of reinforcement-strengthening-expansion, the intention reconstruction step is planned for the node first. The intention reconstruction step forcibly pushes the intention reconstruction learning resources configured to be re-explained and multi-representation interaction. The intent reconstruction step is used as a prerequisite for entering the corresponding knowledge node reinforcement training, and a secondary evaluation is performed after the user completes the intent reconstruction in order to eliminate the impact of intent transmission failure.

7. A method for generating personalized learning paths based on teaching intentions and learning profiles according to any one of claims 1, characterized in that, The method also includes a psychological state adaptive adjustment step: The cognitive load and learning emotional state of students are detected in real time. The cognitive load is classified based on task completion time and frequency of operational hesitation. The learning emotional state is divided into valence-arousal quadrants based on the fusion recognition of facial expressions and interactive behaviors. Based on the cognitive load state and learning emotional state, an instantaneous psychological state vector is generated, and a psychological fit coefficient is calculated to characterize the degree of deviation from the preset ideal learning state. Using the psychological adaptation coefficient, the difference obtained from the path difference calculation step is adaptively corrected, and / or, a psychological buffer layer is dynamically inserted or activated in the personalized academic path, the psychological buffer layer containing indirect knowledge learning activities for reducing cognitive load or regulating emotions.

8. The method for generating personalized learning paths based on teaching intentions and learning profiles according to claim 7, characterized in that, The psychological fit coefficient is used to adaptively correct the degree of difference, specifically as follows: The original difference degree is multiplicatively adjusted by using the attention effectiveness coefficient with psychological fit coefficient as the variable, and a stress compensation term generated by the difficulty of knowledge points and the history of repeated errors is added, so that the difference execution priority of high-difficulty knowledge points is dynamically reduced under high load or negative emotional state.

9. The method for generating personalized learning paths based on teaching intentions and learning profiles according to claim 7, characterized in that, The personalized academic path and resource generation steps further include: The original cognitive goal achievement is transformed into a multi-objective optimization problem that maximizes cognitive growth and maximizes psychological resilience. Real-time psychological constraints are added during path solving. These constraints limit the proportion of nodes with high cognitive load within any continuous time window and force the insertion of an emotion regulation node when a negative high arousal state is detected. During the path execution process, the real-time psychological state is continuously monitored. When the psychological fit coefficient deviates from the fit zone, the subsequent high-load nodes to be executed are dynamically deleted or postponed, and preset cognitive load reduction or emotion regulation activities are inserted.

10. A personalized learning path generation system based on teaching intentions and student learning profiles, characterized in that, A method for generating personalized academic paths based on teaching intentions and learning profiles as described in any one of claims 1 to 9, comprising: The module for dynamically reconstructing teaching intentions: acquires structured teaching data in real time, analyzes teachers' teaching behaviors using a large language model, and dynamically identifies and reconstructs the teaching intentions corresponding to the current teaching segment; Multidimensional dynamic learning profile construction module: Collects learning data generated by students in classroom interaction and after-class assessment, combines it with the teaching intentions, extracts students' weak points in knowledge mastery, and generates a dynamic learning profile that includes cognitive level, learning preferences and weak knowledge map; Path Difference Calculation Module: Compares the dynamic learning profile with the current teaching objectives to calculate the difference in students' achievement of objectives at each knowledge node; Personalized academic path and resource generation module: Based on the degree of difference in goal achievement, it uses the generation capabilities of the large language model to generate customized learning resources, and dynamically plans a personalized academic path that suits the student according to the logic of remediation-reinforcement-expansion.

Images