Teaching process evaluation and accurate intervention method and system based on ability portrait

Abstract

The invention discloses a teaching process evaluation and accurate intervention method and system based on ability portray.The method belongs to the technical field of wisdom education, and the method comprises the steps that full-dimensional teaching data containing teacher and student recessive states are collected through a multi-source heterogeneous architecture, and refined preprocessing is carried out; a cognitive load theory and a multivariate intelligence theory are fused to construct a student multi-dimensional cognitive load label library and a behavior semantic unit chain model, a Delphi method and an analytic hierarchy process are combined to construct a teacher teaching ability index system, and a teaching and learning two-way linkage mapping model is established; an improved multi-modal deep learning fusion model and a transfer learning model are used for accurate evaluation, and based on an evaluation result and real-time data, hierarchical personalized intervention suggestions are generated and dynamically adjusted through an improved decision tree and reinforcement learning. The problems that existing teaching evaluation is one-sided, intervention lags behind, linkage lacks and the data security risk is high are solved, and the objectivity of teaching evaluation and the timeliness of teaching intervention are remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of smart education data mining and artificial intelligence technology, and in particular to a method and system for process evaluation and precise intervention in teaching based on ability profiles. Background Technology

[0002] Educational evaluation is a crucial link in improving the quality of education. Traditional teaching evaluation is mostly "summative evaluation," relying on static, outcome-based indicators such as exam scores and classroom observation grades. This approach suffers from prominent problems such as a single evaluation dimension, lack of process information, strong subjectivity, and delayed feedback. This evaluation model fails to reveal the complex dynamic interactions in the teaching process and cannot provide teachers with timely, accurate, and actionable guidance for improvement.

[0003] The development of educational informatization has led to the introduction of various technologies into teaching evaluation. For example, recording systems are used to analyze student and teacher behavior, and learning management systems are used to collect online learning data. However, existing solutions still have significant limitations: First, data collection is one-sided. Most focus is on easily accessible explicit behavioral data (such as attendance and number of questions answered), severely neglecting deeper, implicit states such as students' attention, emotions, and cognitive load—the core factors affecting learning outcomes. Second, evaluation models are isolated. They typically construct independent "digital profiles" for students or teachers, failing to establish an organic link and causal relationship between the abilities of "teaching" and "learning," thus deviating from the essence of mutual learning and growth. Third, intervention strategies are static. The generated improvement suggestions are often based on general conclusions from historical data, lacking the ability to perceive and adapt to real-time changes in the teaching process, resulting in insufficient precision and timeliness of interventions. Fourth, data security and challenges in large-scale application. How to securely collect, store, transfer, and collaboratively compute multimodal data involving a large amount of sensitive information about teachers and students is a significant bottleneck faced by existing solutions when promoted across multiple campuses and regions.

[0004] Existing technologies, such as CN120542988A (classification number G06Q10), disclose a method and system for cultivating and evaluating students' comprehensive literacy based on artificial intelligence. This method includes extracting data from multiple modalities, such as audio, video, and text, from recorded videos of current classes and preprocessing them; fusing the preprocessed multimodal data; and, based on the multimodal data fusion, further calculating N levels of indicators that match the AI ​​model. These indicators reflect the comprehensive effectiveness of classroom teaching in cultivating students' literacy and its relevance to teaching. The indices of the lowest-level indicators are compared and analyzed with corresponding benchmark values ​​derived from national high-quality classrooms, pointing out the deficiencies or strengths of each indicator in the current classroom and providing targeted teaching optimization suggestions.

[0005] The above literature only assesses the appropriateness of student-teacher interactions, without evaluating knowledge mastery or multi-dimensional cognition of multidisciplinary knowledge points. It also fails to map the teaching process to the teacher's implementation, evaluating only students' theoretical knowledge without confirming whether teachers guided students or whether they truly understood the knowledge points. This leads to situations where students can answer correctly during interactions but cannot answer subsequent questions, indicating a lack of true understanding. These issues require a comprehensive assessment combining explicit and implicit observations during interactions to ensure accuracy and effectiveness. Research should address existing technologies and shortcomings, providing a method and system for process-based evaluation and precise intervention based on competency profiles, ultimately achieving greater practical value. Summary of the Invention

[0006] This invention provides a method and system for process evaluation and precise intervention in teaching based on competency profiles. It utilizes multimodal data collected in the teaching environment and performs decentralized structuring processing on the multimodal data. It uses blockchain to ensure data security and then uses a two-way linkage mapping relationship model between teacher competency profiles and student cognitive load to score teaching according to the mapping relationship between student behavior and teacher competency, thereby comprehensively and accurately evaluating teachers' teaching abilities.

[0007] To achieve the above objectives, the present invention provides a method for formative assessment and precise intervention in teaching based on competency profiles, comprising the following steps: Step S1: Synchronously collect multimodal raw data from the entire teaching scenario; the multimodal raw data includes at least: teacher and student audio and video streams, teaching platform interaction logs, and wearable device physiological signals; perform decentralized structuring processing on the multimodal raw data to generate a structured teaching process dataset; Step S2: Construct a dynamically adjustable multidimensional cognitive load label library for students; the multidimensional cognitive load label library for students includes the degree of knowledge mastery and knowledge points from multiple disciplines, establish a behavioral semantic unit chain model driven by the temporal attention mechanism, and determine students' cognitive load by associating students' explicit behaviors and implicit states; construct a two-way linkage mapping relationship model between teacher competence profiles and students' cognitive load based on the structured teaching process dataset; Step S3: Extract deep features from the structured teaching process dataset; based on the transfer learning evaluation model of the pre-trained model and the mapping relationship between teacher competence profile and student cognitive load, quantitatively score and determine the suitability level of teacher teaching ability, and generate a visual evaluation report containing teaching-learning suitability analysis; introduce a human-computer collaborative verification mechanism to correct the evaluation results.

[0008] The above settings, by collecting audio and video streams from teachers and students, as well as interaction logs and physiological signals from the teaching platform, can provide a comprehensive understanding of teacher-student interactions and their psychological states. Based on the structured teaching process dataset, student explicit behaviors are correlated with implicit states, and these are mapped to teacher competency profiles. This allows for the evaluation of the mapping relationship between student behavior and teacher teaching ability through subsequent pre-trained models. By mapping student behavior and teacher ability, teaching scores can be assigned based on this mapping relationship, resulting in more accurate evaluation scores. Furthermore, the assessment of students' multidimensional cognition includes not only the degree of knowledge mastery but also knowledge points from multiple disciplines. Therefore, the assessment of students is not limited to knowledge points within a single subject but emphasizes the interrelationships between multiple disciplines.

[0009] Furthermore, after step S3, step S4 is also included: based on the mapping relationship between the visualized assessment report, teacher competence profile, and student cognitive load, a multi-dimensional performance feature matrix of teachers is constructed; an improved decision tree algorithm combined with reinforcement learning is used to determine the priority sequence of teacher competence improvement dimensions; and improvement suggestions are given according to the priority sequence; a dynamic adaptive mechanism is designed to automatically adjust the content and intensity of intervention strategies based on real-time teaching feedback data. Step S5: Real-time and intelligent monitoring and feedback of the teaching process; adopting a hybrid architecture combining edge computing and cloud platform to realize real-time collection, preprocessing and analysis of teaching data; constructing a dynamic monitoring model based on temporal neural network to conduct real-time status evaluation, effect prediction and problem warning of the teaching process; through a multi-terminal intelligent interaction platform, realizing real-time push of intervention suggestions and instant collection of teacher and student feedback, forming a closed loop of teaching and learning.

[0010] The above settings construct a multi-dimensional performance matrix for teachers by mapping the relationship between the visual assessment report, teacher competency profile, and student cognitive load. This facilitates the assessment of teachers across multiple dimensions, prioritizes areas for improvement, and provides suggestions based on those priorities. It also enables real-time monitoring of the teaching process, real-time collection of teaching data, and dynamic evaluation of the teaching process. Furthermore, it facilitates intelligent interaction across multiple terminals and allows for the participation of feedback from teachers and students in the assessment process.

[0011] Furthermore, in step S1, collecting students' implicit state data specifically includes: collecting heart rate, skin conductance response, and body temperature physiological signals using wearable devices provided to students; analyzing the facial video streams of teachers and students using computer vision technology to identify students' attention concentration, emotional state, and body language characteristics; and identifying the teacher's teaching emotions and the emotional tendencies of student interactions using audio emotion analysis technology.

[0012] Furthermore, step S1 also includes fine preprocessing of the multimodal raw data. The fine preprocessing includes: using a combination of rule-based and cluster analysis methods to detect and filter invalid and abnormal multimodal raw data; using an imputation method based on random forest regression to repair missing data; and using min-max normalization and Z-score normalization to scale and align features of heterogeneous data sources, thereby unifying the data scale and distribution.

[0013] The above settings can easily remove abnormal data and ensure data accuracy.

[0014] Furthermore, in step S2, the student's multidimensional cognitive load includes a three-level cognitive dimension and a multiple intelligences dimension; the three-level cognitive dimension includes a basic layer, an enhancement layer, and an innovation layer. The basic layer includes the memorization and understanding of knowledge points; the enhancement layer includes the application and analysis of knowledge points; and the innovation layer includes the evaluation and creation of knowledge points. The multiple intelligences dimension includes at least linguistic, logical-mathematical, spatial, and kinesthetic intelligence. The weights of each dimension label are dynamically optimized using a gradient descent algorithm combined with real-time implicit state data.

[0015] The above setup assesses knowledge points by classifying them according to students' three cognitive dimensions, and then evaluates them from multiple disciplines, including language, logical-mathematical, spatial, and kinesthetic intelligence.

[0016] Furthermore, in step S2, the behavioral semantic unit chain model uses a temporal attention mechanism to associate and serialize explicit behavioral events such as raising hands, taking notes, and asking questions with implicit state events such as focus, confusion, and excitement with millisecond-level precision, forming a semantic chain that describes the cognitive and behavioral evolution of students in a specific teaching segment.

[0017] The above settings, through a temporal attention mechanism, determine the frequency of raising hands, taking notes, and asking questions, and identify hidden state events, making the correspondence between cognition and behavior more accurate.

[0018] Furthermore, in step S2, the teacher teaching ability evaluation index system includes basic ability indicators and extended ability indicators; the basic ability indicators include knowledge mastery, teaching implementation, learning guidance and evaluation feedback; the extended ability indicators include personalized adaptation, interdisciplinary integration, technology integration and reflective iteration ability; the weight of each indicator is determined by the analytic hierarchy process.

[0019] The above settings comprehensively evaluate teachers' teaching abilities from two aspects: basic ability indicators and extended ability indicators.

[0020] Furthermore, the construction of the bidirectional linkage mapping relationship model in step S2 specifically includes: S2.1, establishing a multi-level time window including millisecond and teaching segment levels, aligning the teacher behavior event sequence with the student cognitive state event sequence in time and space, and generating an aligned event pair set; S2.2, using a neural network model based on the cross-attention mechanism, with the student cognitive state as the query and the teacher's ability performance as the key, calculating the attention weight and effect intensity of the influence of various teacher abilities on various student cognitive loads, and determining the lag time of the influence; S2.3, using a neural network model based on the cross-attention mechanism, with the teacher's ability characteristics as the query and the student's cognitive state change feedback signal as the key, calculating the attention weight of the influence of student state feedback on teacher ability adjustment; S2.4, introducing moderating variables representing individual student traits and teaching context, and dynamically adjusting the effect intensity in the forward influence path through gating mechanisms or weight modulation methods.

[0021] The above settings, based on the frequency of students raising their hands, taking notes, and asking questions in the students' cognitive load, determine the hidden state events of focus and excitement, and determine the teaching implementation aspect in the teacher's teaching ability evaluation index system. When students answer incorrectly, it is also necessary to determine the teacher's confusion during the interaction with students, thereby identifying deficiencies in knowledge mastery, learning knowledge, and personalized adaptation.

[0022] In another aspect, this invention provides a teaching process evaluation and precise intervention system based on ability profiles, comprising: The multi-dimensional data acquisition and refined preprocessing module is used to deploy and manage acquisition equipment, complete the synchronous acquisition, cleaning, repair and standardization of multimodal raw data, and output structured datasets; The module for constructing a bidirectional linkage capability profile between teaching and learning is used to build and maintain a student cognitive load label library, a behavioral semantic unit chain model, a teacher capability indicator system, and a bidirectional mapping model between the two. The multimodal fusion teaching ability precision assessment module is used to run multimodal fusion models and transfer learning assessment models, complete ability quantitative assessment, generate visual reports, and manage the human-machine collaborative verification process; The personalized and dynamic improvement suggestion generation module is used to analyze the performance feature matrix, calculate the improvement priority, generate personalized suggestion packages, and dynamically adjust the suggestion strategy based on real-time feedback. The real-time and intelligent teaching process monitoring and feedback module is used to manage edge and cloud hybrid computing resources, run dynamic monitoring models, and realize real-time early warning and feedback interaction through a multi-terminal interactive platform.

[0023] Compared with the prior art, the beneficial effects of the present invention are: 1. This solution breaks through the limitation of only collecting explicit data and innovatively incorporates implicit states such as physiological signals and micro-expressions into the analysis. Combined with the fusion of multi-source heterogeneous data, it achieves a "panoramic" deep perception of the teaching process. The proposed "teaching and learning two-way linkage ability profile" model not only constructs a highly interpretable labeling system based on educational psychology theory, but more importantly, it establishes a quantitative correlation model between teacher ability and student cognitive state, so that the diagnostic conclusions truly fit the essence of teaching interaction.

[0024] 2. This invention employs multimodal deep learning and transfer learning technologies to automatically extract and evaluate complex features, significantly reducing subjective human bias. Combined with human-machine collaborative verification, it ensures the objectivity and credibility of the evaluation results. The generated improvement suggestions are based on real-time data and dynamic ranking and adjustment through reinforcement learning, exhibiting high personalization and context adaptability. This achieves a leap from "static reporting" to "dynamic navigation," significantly improving the effectiveness of intervention. Attached Figure Description

[0025] Figure 1 This is a flowchart of the present invention.

[0026] Figure 2 This is a structural block diagram of the present invention.

[0027] Figure 3 This is a diagram of the teaching assessment interface in one embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figure 1 As shown, this invention provides a method for formative assessment and precise intervention in teaching based on competency profiles, comprising the following steps: Step S1: Synchronously collect multimodal raw data from the entire teaching scenario; the multimodal raw data includes at least: teacher and student audio and video streams, teaching platform interaction logs, and wearable device physiological signals; perform decentralized structuring processing on the multimodal raw data to generate a structured teaching process dataset.

[0030] Employing a "multi-source heterogeneous acquisition architecture," high-definition visual sensors, array microphones, interactive teaching terminals, wearable physiological sensing devices, and environmental sensors are deployed in a natural teaching environment to collect students' implicit state data. Specifically, this includes: collecting heart rate, skin conductance response, and body temperature physiological signals through wearable devices provided to students; analyzing facial video streams of teachers and students using computer vision technology to identify students' attention concentration, emotional state, and body language characteristics; identifying teachers' teaching emotions and students' emotional tendencies through audio emotion analysis technology; and using decentralized structured processing to process the multimodal raw data. Blockchain is used to ensure the security and reliability of the multimodal raw data.

[0031] Then, refined preprocessing is performed: first, anomaly detection and filtering are carried out based on rules and K-means clustering; second, to address the problem of missing data, a strategy combining linear interpolation and random forest regression is used for intelligent repair; finally, min-max normalization and Z-score normalization are used to perform feature alignment and scaling on data of different dimensions and distributions, outputting a high-quality structured teaching process dataset.

[0032] Step S2: Construct a dynamically adjustable multidimensional cognitive load label library for students; the multidimensional cognitive load label library for students includes the degree of knowledge mastery and knowledge points from multiple disciplines, establish a behavioral semantic unit chain model driven by the temporal attention mechanism, and determine the cognitive load of students by associating their explicit behaviors with implicit states; construct a bidirectional linkage mapping relationship model between teacher competence profiles and student cognitive load based on a structured teaching process dataset.

[0033] Construction of a bidirectional teaching and learning ability profile. This step involves building a core cognitive model. For the student, integrating cognitive load theory and multiple intelligences theory, a label library is constructed that includes three levels of cognitive dimensions: "basic layer (memory, understanding) - advanced layer (application, analysis) - innovative layer (evaluation, creation)" and multiple non-intellectual dimensions such as language and logic-mathematics. The weights are dynamically adjusted based on the student's real-time physiological and facial expression data using a gradient descent algorithm.

[0034] Explicit behaviors include raising hands, taking notes, and asking questions; implicit behaviors include latent state events such as focus, confusion, and excitement. A behavioral semantic unit chain model driven by cognitive load and latent states is proposed, introducing a temporal attention mechanism to associate students' "behavioral atoms" (such as raising hands) with "state atoms" (such as confusion) on a millisecond-level timeline, forming a dynamic semantic chain describing their cognitive process. For teachers, the Delphi method is used to gather expert wisdom, and the Analytic Hierarchy Process (AHP) is combined to quantify and determine the weights of evaluation indicators, constructing a multi-dimensional teacher teaching ability evaluation indicator system that includes "basic ability indicators" (knowledge mastery, teaching implementation, etc.) and "core extension indicators" (personalized adaptation, technology integration, etc.).

[0035] Ultimately, a two-way linkage mapping model for teaching and learning is established to quantitatively characterize the influence weight between a teacher's ability (such as clarity of explanation) and a student's specific cognitive state (such as comprehension). For example, if a student's overt behavior shows frequent questioning and their covert behavior shows focus, it indicates that the student has a higher level of cognition. The model maps students' specific cognitive states to teachers' knowledge mastery and teaching implementation, with the weight of teachers' knowledge mastery being greater than the weight of teaching implementation. This facilitates the subsequent calculation of corresponding evaluation scores based on the mapping relationship.

[0036] The student's multidimensional cognitive load includes a three-level cognitive dimension and a multiple intelligences dimension. The three-level cognitive dimension includes a basic layer, an enhancement layer, and an innovation layer. The basic layer includes the memorization and understanding of knowledge points; the enhancement layer includes the application and analysis of knowledge points; and the innovation layer includes the evaluation and creation of knowledge points. The multiple intelligences dimension includes at least linguistic, logical-mathematical, spatial, and kinesthetic intelligence. The weights of each dimension label are dynamically optimized using a gradient descent algorithm combined with real-time implicit state data.

[0037] The behavioral semantic unit chain model uses a temporal attention mechanism to associate and serialize explicit behavioral events such as raising hands, taking notes, and asking questions with implicit state events such as focus, confusion, and excitement with millisecond-level precision, forming a semantic chain that describes the cognitive and behavioral evolution of students in a specific teaching session.

[0038] The teacher teaching ability evaluation index system includes basic ability indicators and extended ability indicators; the basic ability indicators include knowledge mastery, teaching implementation, learning guidance and evaluation feedback; the extended ability indicators include personalized adaptation, interdisciplinary integration, technology integration and reflective iteration ability; the weight of each indicator is determined by the analytic hierarchy process.

[0039] Step S3: Precise Assessment of Multimodal Integrated Teaching Ability. Deep features are extracted from the structured teaching process dataset; based on a pre-trained transfer learning assessment model and the mapping relationship between teacher competence profiles and student cognitive load, the teacher's teaching ability is quantitatively scored and its suitability level is determined, generating a visual assessment report containing teaching-learning suitability analysis; a human-machine collaborative verification mechanism is introduced to correct the assessment results.

[0040] This step enables the intelligent transformation from data to evaluation. A multimodal deep learning fusion model is designed, which combines an improved ResNet-50 (for extracting visual spatiotemporal features), Bi-LSTM (for extracting audio sequences and semantic features), and GCN (for analyzing knowledge graph structural features) in parallel, and performs feature weighted fusion through an attention layer.

[0041] Building upon this foundation, a transfer learning evaluation model based on a pre-trained model (such as BERT) is constructed. This model initializes feature layers with knowledge obtained through pre-training on a large-scale general corpus, and then builds a network fine-tuned for the teaching evaluation task. Finally, through dual output layers, it provides a quantitative score of the teacher's ability and a teaching-learning fit level (high / medium / low). The evaluation results are presented in a dynamic, visual report format, and a human-machine collaborative verification mechanism is introduced. Educational experts sample, review, and correct the model's output to ensure the fairness and credibility of the evaluation.

[0042] Step S4: Generation of Personalized and Dynamic Improvement Suggestions. This step translates assessment into action. First, based on the assessment report from S3 and the linkage mapping model from S2, a multi-dimensional performance feature matrix reflecting teachers' specific shortcomings is constructed. Using an improved decision tree (such as LightGBM) combined with reinforcement learning, improvements in teaching effectiveness are simulated as reward signals, automatically prioritizing the most critical capability improvement dimensions for overall effectiveness. Next, based on teachers' current capability level and improvement priorities, a three-tiered (basic improvement, advanced optimization, and innovative breakthrough) personalized improvement suggestion package is generated. Each suggestion includes specific operational methods, similar successful cases, and links to relevant teaching resources. Finally, a dynamic adaptive mechanism is designed. This mechanism continuously monitors fluctuations in teaching effectiveness and changes in student status after teachers adopt suggestions, and uses reinforcement learning strategies to adjust the direction and intensity of subsequent intervention suggestions in real time.

[0043] Step S5: Real-time and Intelligent Monitoring and Feedback of the Teaching Process. This step ensures the timeliness of intervention. A hybrid architecture of edge computing and cloud platform is adopted: edge computing nodes are deployed locally in the classroom to perform real-time preprocessing and lightweight analysis of audio and video streams, greatly reducing latency; the cloud platform is responsible for massive data aggregation, complex model operation, and long-term storage. On this architecture, a dynamic monitoring model for teaching effectiveness based on a temporal attention mechanism (such as Bi-GRU with Attention) is built to achieve multi-task parallel output of real-time student status assessment, instantaneous prediction of teaching effectiveness, and early warning of potential problems (response latency <500ms). Through a multi-terminal intelligent interactive platform (intelligent podium for teachers, tablet / mobile APP for students), early warning information and personalized suggestions are pushed to teachers and students in real time, and their feedback is received, forming an instant teaching closed loop of "monitoring-assessment-intervention-feedback".

[0044] The construction of the bidirectional linkage mapping relationship model in step S2 specifically includes: S2.1 Establish a multi-level time window that includes millisecond-level and teaching segment-level time windows, and align the teacher behavior event sequence with the student cognitive state event sequence in time and space to generate an aligned event pair set; S2.2. Using a neural network model based on the cross-attention mechanism, with students' cognitive state as the query and teachers' performance as the key, calculate the attention weight and effect strength of the influence of each teacher's ability on each student's cognitive load, and determine the lag time of the influence. S2.3. Using a neural network model based on the cross-attention mechanism, with teacher ability characteristics as the query and student cognitive state change feedback signals as the key, calculate the attention weight of the influence of student state feedback on teacher ability adjustment. S2.4 Introduce moderating variables that represent individual student traits and teaching context, and dynamically adjust the effect intensity in the forward influence path through gating mechanisms or weight modulation.

[0045] In step S2.2, the attention weight of the teacher's ability on the student's cognitive load is calculated through the following cross-attention mechanism: Let the teacher's ability feature vector at time t be... The student's cognitive state feature vector is First, the query is obtained through a linear transformation. :

[0046] in, This is a learnable weight matrix.

[0047] Then the forward attention weight matrix Calculated using the following formula:

[0048] in, This is the dimension of the key vector; the Softmax function normalizes it row-wise. Weight matrix. In This indicates the degree of attention paid to the teacher's j-th ability characteristic when predicting changes in the student's i-th cognitive load.

[0049] In step S2.2, the strength of the effect of teacher competence on student cognitive load. The prediction is achieved through the following steps: First, using the attention weights For teacher eigenvectors Perform weighted aggregation to obtain the context vector. :

[0050] Then, the student's current state vector With context vector The data is fused and then passed through a multilayer perceptron regression network. Predicting the future Changes in students' cognitive load over a period of time :

[0051] Where [;] denotes vector concatenation. Effect intensity From the predicted change The absolute value of the attention weight A{ts}(i,j) is determined by the attention weight A{ts}(i,j), specifically:

[0052] in, It is a monotonically increasing scalar function used to synthesize two pieces of information.

[0053] In step S2.3, the attention weight of the influence of student status feedback on teacher ability adjustment is calculated through the following cross-attention mechanism: Let the feedback signal for changes in student cognitive state be... Teacher competency characteristics are The query is obtained through linear transformation. :

[0054] in, This is a learnable weight matrix.

[0055] Then the reverse attention weight matrix Calculated using the following formula:

[0056] weight matrix elements in This indicates the degree of attention the teacher pays to the student's cognitive state feedback signal when adjusting the teacher's k-th ability characteristic.

[0057] In step S2.4, the integration of dynamic adjustment factors is achieved through a gating mechanism: Let the vector of adjustment variables be... Through a gating network Generate modulation gate g:

[0058] in, It is the Sigmoid activation function. For learnable parameters, [;] indicates vector concatenation.

[0059] The gate signal g is used to modulate the context vector in the forward influence path. or final effect intensity Modulated output for:

[0060] in, This indicates element-wise multiplication.

[0061] The output of the bidirectional linkage mapping model is a dynamic linkage mapping matrix R(t), where the matrix elements... Defined as:

[0062] in, Indicates the teacher's ability near time t Cognitive load on students The intensity of the influence effect; Feedback on teacher competence Modulation gate value.

[0063] This determines the teacher's cognitive load on students. Influence effect strength, student cognitive load Feedback on teacher competence The relationship between the adjusted attention intensity and the typical lag time of the effect determines the evaluation value of the teaching process.

[0064] In step S3, the transfer learning evaluation model based on the pre-trained model and the mapping relationship between the teacher's competence profile and the student's cognitive load are used to quantitatively score and determine the suitability level of the teacher's teaching ability. This includes: in the student's cognitive load, if the frequency of students raising their hands, taking notes, and asking questions is greater than a preset frequency, it is determined that the student's focus and excitement are latent events, which are related to the teaching implementation aspect in the teacher's teaching ability evaluation index system; if the frequency of students raising their hands, taking notes, and asking questions is less than a preset frequency, it is determined that the student is confused and has a deficiency in knowledge mastery in the teacher's teaching ability evaluation index system; if the frequency of students raising their hands, taking notes, and asking questions is less than a preset frequency, and the frequency of incorrect answers to knowledge points during student-teacher interaction is greater than a preset value, and the teacher shows confusion during the interaction with students, then it is determined that the student is confused and has a deficiency in knowledge mastery, learning knowledge, and personalized adaptation in the teacher's teaching ability evaluation index system.

[0065] Following step S5, the process also includes: obtaining student suggestions for the lesson through a multi-terminal intelligent interactive platform. For example, if the teacher's explanation of a certain knowledge point is inadequate, semantic analysis is used to determine whether it pertains to knowledge mastery in the basic ability indicators of the teacher's teaching ability evaluation system. Then, the knowledge mastery aspect in the teacher's ability profile is determined. Based on the mapping relationship between the teacher's ability profile and the student's cognitive load, the most frequently asked questions are identified, and optimization suggestions are given. The most frequently asked questions are reinforced with knowledge points. In the knowledge point reinforcement class, the teacher's teaching is re-evaluated through steps S1-S4. The evaluation score determines whether the improvement plan is qualified. For example, if the teacher's teaching evaluation score increases significantly after the fixed knowledge point class, the improvement plan is deemed effective.

[0066] In one embodiment, for example, when a teacher is explaining knowledge point A, multimodal raw data such as audio and video data of classroom interactions between students and teachers are collected. Then, image recognition algorithms are used to determine implicit state events of students and teachers regarding classroom emotions, frequency of raising hands, frequency of asking questions, and frequency of note-taking. This determines whether the student is in a state of focus, confusion, or excitement. For example, if a student frequently raises their hand to answer a question from the teacher, it indicates that they are in a state of focus and excitement, indicating that the student has a good grasp of the knowledge point. Furthermore, the logic behind the student's answers is extracted based on the language used in the answers, such as whether the question involves integration of other subjects. This is specifically implemented through semantic recognition algorithms. Then, a mapping is made between the student's cognitive load and the teacher's competency profile. For example, a student's good grasp of the knowledge point is mapped to a teacher's strong explanation ability and good teaching implementation. If the student's answers have poor correlation with other subjects, it indicates that the teacher's interdisciplinary integration is poor and is associated with it, so that the weight of interdisciplinary integration is the highest. Then, based on the weight, the corresponding evaluation scores between different student cognitive loads and different teacher competency profiles are calculated, thereby generating a reliability evaluation report.

[0067] This invention provides a teaching process evaluation and precise intervention system based on ability profiles, including: a multi-dimensional data acquisition and refined preprocessing module, used to deploy and manage acquisition equipment, complete the synchronous acquisition, cleaning, repair and standardization of multimodal raw data, and output structured datasets; The module for constructing a bidirectional linkage capability profile between teaching and learning is used to build and maintain a student cognitive load label library, a behavioral semantic unit chain model, a teacher capability indicator system, and a bidirectional mapping model between the two. The multimodal fusion teaching ability precision assessment module is used to run multimodal fusion models and transfer learning assessment models, complete ability quantitative assessment, generate visual reports, and manage the human-machine collaborative verification process; The personalized and dynamic improvement suggestion generation module is used to analyze the performance feature matrix, calculate the improvement priority, generate personalized suggestion packages, and dynamically adjust the suggestion strategy based on real-time feedback. The real-time and intelligent teaching process monitoring and feedback module is used to manage edge and cloud hybrid computing resources, run dynamic monitoring models, and realize real-time early warning and feedback interaction through a multi-terminal interactive platform.

[0068] The working principle of this invention is as follows: By collecting audio and video streams from teachers and students, as well as interaction logs and physiological signals from the teaching platform, a comprehensive understanding of teacher-student interactions and their psychological states can be achieved. Based on a structured teaching process dataset, students' explicit behaviors are correlated with their implicit states, and these behaviors are mapped to teacher competency profiles. This allows for the evaluation of the mapping relationship between student behavior and teacher teaching ability through a pre-trained model. The mapping between student behavior and teacher ability enables subsequent teaching scoring based on this relationship, resulting in more accurate assessment scores. Furthermore, the multidimensional cognition of students includes not only the degree of knowledge mastery but also knowledge points from multiple disciplines. Therefore, the assessment of students is not limited to knowledge points within a single subject but emphasizes the interrelationships between multiple disciplines.

Claims

1. A method for formative assessment and precise intervention in teaching based on competency profiles, characterized in that: Includes the following steps: Step S1: Synchronously collect multimodal raw data across the entire teaching scenario; The multimodal raw data includes at least: audio and video streams from teachers and students, interactive logs from the teaching platform, and physiological signals from wearable devices; the multimodal raw data is then processed in a decentralized and structured manner to generate a structured teaching process dataset; Step S2: Construct a dynamically adjustable multidimensional cognitive load label library for students; the multidimensional cognitive load label library for students includes the degree of knowledge mastery and knowledge points from multiple disciplines, establish a behavioral semantic unit chain model driven by the temporal attention mechanism, and determine students' cognitive load by associating students' explicit behaviors and implicit states; construct a two-way linkage mapping relationship model between teacher competence profiles and students' cognitive load based on the structured teaching process dataset; Step S3: Extract deep features from the structured teaching process dataset; based on the transfer learning evaluation model of the pre-trained model and the mapping relationship between teacher competence profile and student cognitive load, quantitatively score and determine the suitability level of teacher teaching ability, and generate a visual evaluation report containing teaching-learning suitability analysis; introduce a human-computer collaborative verification mechanism to correct the evaluation results.

2. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 1, characterized in that, Step S3 is followed by step S4, which involves constructing a multi-dimensional performance characteristic matrix of teachers based on the mapping relationship between the visualized assessment report, teacher competence profile, and student cognitive load. An improved decision tree algorithm incorporating reinforcement learning is used to determine the priority sequence of teacher competence improvement dimensions; and improvement suggestions are given based on the priority sequence. Design a dynamic adaptive mechanism to automatically adjust the content and intensity of intervention strategies based on real-time teaching feedback data; Step S5: Real-time and intelligent monitoring and feedback of the teaching process; adopting a hybrid architecture combining edge computing and cloud platform to realize real-time acquisition, preprocessing and analysis of teaching data; constructing a dynamic monitoring model based on temporal neural network to conduct real-time status evaluation, effect prediction and problem early warning of the teaching process; Through a multi-terminal intelligent interactive platform, intervention suggestions can be pushed out in real time and feedback from teachers and students can be collected instantly, forming a closed loop of teaching and learning.

3. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 1, characterized in that, In step S1, collecting students' implicit state data specifically includes: collecting heart rate, skin conductance response, and body temperature physiological signals using wearable devices provided to students; analyzing the facial video streams of teachers and students using computer vision technology to identify students' attention concentration, emotional state, and body language characteristics; and identifying the teacher's teaching emotions and the emotional tendencies of student interactions using audio emotion analysis technology.

4. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 1, characterized in that, Step S1 further includes fine preprocessing of the multimodal raw data. Fine preprocessing includes: using a combination of rule-based and cluster analysis methods to detect and filter invalid and abnormal multimodal raw data; using an imputation method based on random forest regression to repair missing data; and using min-max normalization and Z-score normalization to scale and align features of heterogeneous data sources, thereby unifying the data scale and distribution.

5. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 1, characterized in that, In step S2, the student's multidimensional cognitive load includes a three-level cognitive dimension and a multiple intelligences dimension. The three-level cognitive dimensions include: a basic layer, an enhancement layer, and an innovation layer. The basic layer includes the memorization and understanding of knowledge points; the enhancement layer includes the application and analysis of knowledge points; and the innovation layer includes the evaluation and creation of knowledge points. The multiple intelligences dimensions include at least linguistic, logical-mathematical, spatial, and kinesthetic intelligence. The weights of each dimension label are dynamically optimized using a gradient descent algorithm combined with real-time implicit state data.

6. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 1, characterized in that, In step S2, the behavioral semantic unit chain model uses a temporal attention mechanism to associate and serialize explicit behavioral events such as raising hands, taking notes, and asking questions with implicit state events such as focus, confusion, and excitement with millisecond-level precision, forming a semantic chain that describes the cognitive and behavioral evolution of students in a specific teaching segment.

7. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 6, characterized in that, In step S2, the teacher teaching ability evaluation index system includes basic ability indicators and extended ability indicators; the basic ability indicators include knowledge mastery, teaching implementation, learning guidance and evaluation feedback; the extended ability indicators include personalized adaptation, interdisciplinary integration, technology integration and reflective iteration ability. The weights of each indicator were determined using the analytic hierarchy process (AHP).

8. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 2, characterized in that, The construction of the bidirectional linkage mapping model in step S2 specifically includes: S2.1, establishing multi-level time windows including millisecond and teaching segment levels, aligning the teacher behavior event sequence with the student cognitive state event sequence in time and space, and generating an aligned event pair set; S2.2, using a neural network model based on the cross-attention mechanism, with the student cognitive state as the query and the teacher's ability performance as the key, calculating the attention weight and effect intensity of the influence of various teacher abilities on various student cognitive loads, and determining the lag time of the influence; S2.3, using a neural network model based on the cross-attention mechanism, with the teacher's ability characteristics as the query and the student's cognitive state change feedback signal as the key, calculating the attention weight of the influence of student state feedback on teacher ability adjustment; S2.4, introducing moderating variables representing individual student traits and teaching situations, and dynamically adjusting the effect intensity in the forward influence path through gating mechanisms or weight modulation methods.

9. The method for formative assessment and precise intervention in teaching based on competency profiles according to claim 1, characterized in that, In step S3, the transfer learning evaluation model based on the pre-trained model and the mapping relationship between the teacher's competence profile and the student's cognitive load are used to quantitatively score and determine the suitability level of the teacher's teaching ability. This includes: in the student's cognitive load, if the frequency of students raising their hands, taking notes, and asking questions is greater than a preset frequency, it is determined that the student's focus and excitement are latent events, which are related to the teaching implementation aspect in the teacher's teaching ability evaluation index system; if the frequency of students raising their hands, taking notes, and asking questions is less than a preset frequency, it is determined that the student is confused and has a deficiency in knowledge mastery in the teacher's teaching ability evaluation index system; if the frequency of students raising their hands, taking notes, and asking questions is less than a preset frequency, and the frequency of incorrect answers to knowledge points during student-teacher interaction is greater than a preset value, and the teacher shows confusion during the interaction with students, then it is determined that the student is confused and has a deficiency in knowledge mastery, learning knowledge, and personalized adaptation in the teacher's teaching ability evaluation index system.

10. A teaching process evaluation and precise intervention system based on competency profiles, used in any one of the teaching process evaluation and precise intervention methods based on competency profiles according to claims 1-9, characterized in that, include: The multi-dimensional data acquisition and refined preprocessing module is used to deploy and manage acquisition equipment, complete the synchronous acquisition, cleaning, repair and standardization of multimodal raw data, and output structured datasets; The module for constructing a bidirectional linkage capability profile between teaching and learning is used to build and maintain a student cognitive load label library, a behavioral semantic unit chain model, a teacher capability indicator system, and a bidirectional mapping model between the two. The multimodal fusion teaching ability precision assessment module is used to run multimodal fusion models and transfer learning assessment models, complete ability quantitative assessment, generate visual reports, and manage the human-machine collaborative verification process; The personalized and dynamic improvement suggestion generation module is used to analyze the performance feature matrix, calculate the improvement priority, generate personalized suggestion packages, and dynamically adjust the suggestion strategy based on real-time feedback. The real-time and intelligent teaching process monitoring and feedback module is used to manage edge and cloud hybrid computing resources, run dynamic monitoring models, and realize real-time early warning and feedback interaction through a multi-terminal interactive platform.

Images