AI-enabled quantitative teaching evaluation method and system based on multi-modal data portrait

Abstract

The application provides an AI-enabled quantitative teaching evaluation method and system based on multi-modal data portraits, relates to the technical field of teaching evaluation, and synchronously collects multi-source teaching behavior data in a teaching scene; according to a cognitive load label, feature abstraction is performed on the multi-source teaching behavior data to obtain a behavior semantic unit chain of students in the teaching scene; according to interactive description information, dynamic participation density of the students when receiving knowledge transmission is determined, and then the dynamic participation density is used to determine teaching response efficiency of the students when responding to a teaching strategy in the teaching scene; according to the behavior semantic unit chain and the teaching response efficiency, a teacher ability evaluation portrait based on teaching performance is determined, and a teaching evaluation report is generated according to the teacher ability evaluation portrait. The application can dynamically and structurally associate and evaluate the teaching behavior of teachers when the teaching evaluation result lacks strategy adaptability, so as to improve the accuracy of teaching evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of teaching evaluation, more specifically, the present application relates to an AI-enabled quantitative teaching evaluation method and system based on multi-modal data portraits. BACKGROUND

[0002] Teaching evaluation refers to a full-process evaluation activity of reflecting the teaching level of teachers and the learning effectiveness of students by measuring and analyzing the teaching process and teaching effect based on multi-dimensional data such as the teaching behavior of teachers, the learning response of students, and the use of teaching resources in real teaching scenarios. Traditional teaching evaluation mainly relies on single or subjective data such as questionnaires, grades, and lesson observation records, which is difficult to fully reflect the teaching dynamics and changes in student cognition. With the development of artificial intelligence and multi-modal sensing technology, teaching evaluation is gradually evolving towards data-driven, process-oriented, and intelligent feedback. By synchronously collecting multi-modal data such as audio, video, handwriting, eye movement, and interactive behavior in teaching, and combining semantic recognition, behavior modeling, and intelligent analysis technology, a fine-grained teacher teaching portrait and student cognitive trajectory can be constructed, thereby evaluating the teaching of teachers.

[0003] However, in the existing AI-enabled quantitative teaching evaluation based on multi-modal data portraits, there are defects such as insufficient accuracy of teaching behavior semantic modeling, single dimension of student cognitive response analysis, and insufficient dynamic quantification of teacher-student interaction, which makes the teaching evaluation results lack of strategy adaptability, making it difficult to truly reflect the actual impact of teachers' teaching strategies on students' learning effectiveness, resulting in fragmented feedback results of teaching portraits, lack of personalized guidance significance, and thus restricting the application ability of teaching evaluation in precise teaching improvement. Therefore, how to dynamically and structurally correlate model the teaching behavior of teachers when the teaching evaluation results lack of strategy adaptability to improve the accuracy of teaching evaluation is a problem faced by the industry. SUMMARY

[0004] The present application provides an AI-enabled quantitative teaching evaluation method and system based on multi-modal data portraits, which can dynamically and structurally correlate evaluate the teaching behavior of teachers when the teaching evaluation results lack of strategy adaptability to improve the accuracy of teaching evaluation.

[0005] In a first aspect, the present application provides an AI-enabled quantitative teaching evaluation method based on multi-modal data portraits, which comprises the following steps:

[0006] Synchronously collecting multi-source teaching behavior data in a teaching scenario;

[0007] Determining the cognitive load label of students participating in the classroom in the teaching scenario, and performing feature abstraction on the multi-source teaching behavior data according to the cognitive load label to obtain the behavior semantic unit chain of students in the teaching scenario.

[0008] Obtaining interaction description information of teachers and students in a teaching scene, determining a dynamic participation density of a student in receiving knowledge transmission according to the interaction description information, and further determining a teaching response effectiveness of the student in responding to a teaching strategy in the teaching scene according to the dynamic participation density;

[0009] Determining a teacher ability evaluation portrait based on teaching performance according to the behavior semantic unit chain and the teaching response effectiveness, and generating a teaching evaluation report according to the teacher ability evaluation portrait.

[0010] In the embodiment, the multi-source teaching behavior data refers to a set of original information collected by different types of sensing devices in a teaching scene, which is used to depict the teaching behavior of teachers and the learning behavior of students.

[0011] In the embodiment, determining a cognitive load label of the student participating in the classroom in the teaching scene specifically includes:

[0012] Collecting behavior performance data of the student participating in the classroom in the teaching scene;

[0013] Determining a cognitive interaction index of a cognitive state of the student according to the behavior performance data;

[0014] Determining a cognitive load label of the student participating in the classroom in the teaching scene according to the cognitive interaction index.

[0015] In the embodiment, the behavior semantic unit chain refers to a behavior sequence composed of a plurality of atomic semantic units that are continuously connected in a time dimension and have a teaching semantic logical relationship.

[0016] In the embodiment, obtaining the interaction description information of teachers and students in the teaching scene specifically includes:

[0017] Synchronously collecting a multi-modal teaching interaction stream and generating an interaction event sequence of teachers and students in the teaching scene;

[0018] Determining a teaching behavior semantic in the teaching scene according to the interaction event sequence;

[0019] Generating the interaction description information of teachers and students in the teaching scene according to the teaching behavior semantic.

[0020] In the embodiment, the interaction description information refers to structured data of interaction features of teachers and students in a teaching process.

[0021] In the embodiment, the dynamic participation density refers to an evaluation index of a total amount of effective feedback of a student to knowledge content transmitted by a teacher in a unit of time within a teaching cycle.

[0022] In the embodiment, the teaching response efficiency of the student in the teaching scene in response to the teaching strategy is determined according to the dynamic participation density.

[0023] The density fluctuation characteristics of the student in the teaching strategy are determined according to the dynamic participation density.

[0024] The cooperative response rule of the student in response to the teaching strategy is determined according to the density fluctuation characteristics.

[0025] The teaching response efficiency of the student in the teaching scene in response to the teaching strategy is determined according to the cooperative response rule.

[0026] In the embodiment, the teacher ability evaluation portrait based on teaching performance is determined according to the behavior semantic unit chain and the teaching response efficiency.

[0027] The knowledge transmission trajectory of the teaching strategy implementation is analyzed according to the behavior semantic unit chain, and the cognitive feedback attribute of the student in the teaching scene is determined through the teaching response efficiency.

[0028] The teacher ability evaluation portrait based on teaching performance is determined according to the knowledge transmission trajectory and the cognitive feedback attribute.

[0029] In a second aspect, the application provides an AI-enabled quantitative teaching evaluation system based on multi-modal data portraits, which is used to execute an AI-enabled quantitative teaching evaluation method based on multi-modal data portraits. The evaluation system comprises:

[0030] A data acquisition module is configured to synchronously acquire multi-source teaching behavior data in a teaching scene.

[0031] A feature extraction module is configured to determine a cognitive load label of the student participating in the classroom in the teaching scene, extract features of the multi-source teaching behavior data according to the cognitive load label, and obtain a behavior semantic unit chain of the student in the teaching scene.

[0032] An evaluation response module is configured to acquire interactive description information of teachers and students in the teaching scene, determine a dynamic participation density of the student in receiving knowledge transmission according to the interactive description information, and further determine a teaching response efficiency of the student in the teaching scene in response to a teaching strategy according to the dynamic participation density.

[0033] A report generation module is configured to determine a teacher ability evaluation portrait based on teaching performance according to the behavior semantic unit chain and the teaching response efficiency, and generate a teaching evaluation report according to the teacher ability evaluation portrait.

[0034] The technical scheme provided by the embodiments of the application has the following beneficial effects:

[0035] Multi-source teaching behavior data in the teaching scenario is collected synchronously; cognitive load labels of students participating in the classroom are determined in the teaching scenario; feature extraction is performed on the multi-source teaching behavior data based on the cognitive load labels to obtain the behavioral semantic unit chain of students in the teaching scenario; interactive description information of teachers and students in the teaching scenario is obtained; dynamic participation density of students when receiving knowledge is determined based on the interactive description information; and teaching response effectiveness of students in responding to teaching strategies in the teaching scenario is determined based on the dynamic participation density; a teacher competency assessment profile based on teaching performance is determined based on the behavioral semantic unit chain and the teaching response effectiveness; and a teaching assessment report is generated based on the teacher competency assessment profile.

[0036] Therefore, this application demonstrates that it can improve the accuracy of teaching assessments when the assessment results lack strategy adaptability. Specifically, by synchronously collecting multi-source teaching behavior data from teaching scenarios, it achieves full-dimensional coverage and time-synchronized labeling of student behavioral characteristics during the teaching process, ensuring the consistency and semantic coherence of the collected teaching behavior data. This allows for the construction of a unified behavioral data fusion framework under multi-source heterogeneous input conditions, effectively solving the problems of fragmented observation and temporal alignment difficulties in existing assessments. Furthermore, by determining the cognitive load labels of students participating in the classroom within the teaching scenario, it enables precise perception and dynamic mapping of the implicit cognitive states in student behavior, thereby constructing a behavioral semantic unit chain model based on cognitive load. This model effectively enhances the semantic correspondence between teaching behaviors and students' psychological states, improving the cognitive interpretability of behavioral analysis results. By acquiring interactive descriptions of teachers and students in teaching scenarios, it enables quantitative assessment of students' response activity under strategy guidance, effectively solving the problem of difficulty in capturing dynamic fluctuations in student participation in existing assessment methods. By constructing causal relationship paths between teachers' knowledge transfer trajectories and students' cognitive feedback attributes, it can form a teacher competency assessment profile oriented towards the teaching process. Through structured analysis of each competency dimension in the teacher competency assessment profile, it generates personalized teaching assessment reports, effectively improving the process traceability and result applicability of assessment content, thereby achieving a precise feedback mechanism oriented towards teaching effectiveness.

[0037] In summary, the technical solution adopted in this application can conduct dynamic, structured, and correlated assessments of teachers' teaching behaviors when the teaching assessment results lack strategy adaptability, thereby improving the accuracy of teaching assessments. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is an exemplary flowchart of an AI-empowered quantitative teaching assessment method based on multimodal data profiling provided in this application;

[0040] Figure 2 This is a flowchart illustrating the chain of semantic units for determining behavior provided in this application;

[0041] Figure 3 This is a flowchart illustrating the process for determining dynamic participation density provided in this application;

[0042] Figure 4 This is a module structure diagram of an AI-enabled quantitative teaching assessment system based on multimodal data profiling, provided in this application. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] This application provides an AI-empowered quantitative teaching assessment method and system based on multimodal data profiling. Its core is to simultaneously collect multi-source teaching behavior data in a teaching scenario; determine the cognitive load labels of students participating in the classroom in the teaching scenario; extract features from the multi-source teaching behavior data based on the cognitive load labels to obtain a behavioral semantic unit chain of students in the teaching scenario; acquire interactive description information of teachers and students in the teaching scenario; determine the dynamic participation density of students when receiving knowledge transmission based on the interactive description information; and then determine the teaching response effectiveness of students in responding to teaching strategies in the teaching scenario based on the dynamic participation density; determine a teacher competency assessment profile based on teaching performance based on the behavioral semantic unit chain and the teaching response effectiveness; and generate a teaching assessment report based on the teacher competency assessment profile.

[0045] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of an AI-empowered quantitative teaching assessment method based on multimodal data profiling according to this embodiment of the present application. The assessment method includes the following steps:

[0046] In step S1, multi-source teaching behavior data in the teaching scenario are collected synchronously.

[0047] In practical implementation, firstly, multiple sensing terminals are deployed in a real teaching scenario, including visual acquisition devices (high-definition cameras, depth cameras), audio acquisition devices (multi-channel microphone arrays), and interactive acquisition terminals (such as electronic whiteboards, student tablets, and click recording systems). Then, each device needs to be configured with a unified time synchronization protocol (such as using Network Time Protocol (NTP)) to ensure data consistency. The acquired data is initially processed and encoded by an edge computing module, and then uploaded to the teaching behavior data fusion module in a unified data format (such as a JSON structure). The data fusion module performs unified spatial and temporal alignment of visual, auditory, and operational data from different channels, and performs subject normalization based on person recognition (such as facial recognition and ID binding). Finally, a structured teaching behavior data record table is output, from which multi-source teaching behavior data can be obtained. In other embodiments, other methods can also be used to determine multi-source teaching behavior data, which will not be elaborated here.

[0048] It should be noted that, in this application, multi-source teaching behavior data refers to the collection of raw information collected from different types of sensing devices in the teaching setting, used to characterize teachers' teaching behavior and students' learning behavior.

[0049] In step S2, the cognitive load labels of students participating in the classroom in the teaching scenario are determined, and feature extraction is performed on the multi-source teaching behavior data based on the cognitive load labels to obtain the behavioral semantic unit chain of students in the teaching scenario.

[0050] In this embodiment, determining the cognitive load label of students participating in the classroom in a teaching scenario can be achieved through the following steps:

[0051] Collect data on students' behavioral performance in classroom settings.

[0052] Cognitive interaction indicators of students' cognitive state are determined based on the behavioral performance data.

[0053] The cognitive load labels of students participating in the classroom in the teaching scenario are determined based on the aforementioned cognitive interaction indicators.

[0054] In practical implementation, firstly, multimodal sensing devices are deployed in the teaching scenario, including visual acquisition devices, audio acquisition devices, and interactive acquisition devices. The visual acquisition devices include high-definition cameras and depth cameras, installed at the front and sides of the classroom to capture students' facial expressions, head orientation, and body posture in real time. The audio acquisition devices include a multi-channel microphone array to collect students' speech response frequency and speech characteristics (pitch, pauses, rhythm). Interactive acquisition devices, such as electronic tablets, classroom response systems, or smart blackboard terminals used by students, are used to record students' clicks, answers, drags, note-taking, and other operational behaviors. The data collected by the multimodal sensing devices serves as behavioral data on students' participation in the classroom. Then, the visual acquisition data can be used to extract head postures (such as the frequency of continuous head turning and head looking down) and eye movements to identify whether students are paying attention to the podium, screen, or textbook. A time window is then used to statistically analyze the continuous time periods and switching frequencies of students' gaze at the teaching content, calculating a focus score. The click paths, answer accuracy, and answer time collected from the interactive devices were then analyzed. The inverse relationship between answer time and accuracy can be used to assess cognitive load levels. If students repeatedly modify their answers or their answer time deviates significantly from the average, it indicates high cognitive expenditure during comprehension. The frequency of student participation in voice-based question-and-answer sessions, the length of their voice content, and the number of times they actively spoke were statistically analyzed. This data was converted into a behavioral initiative score, which was used as a cognitive interaction indicator. Finally, the cognitive interaction indicators of all students within a fixed time period were normalized. The processed cognitive interaction indicators were then input into a classification model, which used learned weights to classify the current student's state based on their cognitive load. The model output three levels of labels: "Low Load," "Medium Load," and "High Load," corresponding to easy comprehension, moderate challenge, and cognitive overload, respectively. These labels were bound to time periods and student identities and recorded in a cognitive load log table. This cognitive load log table provides the cognitive load labels for students participating in the classroom during the teaching scenario.

[0055] It should be noted that, in this application, behavioral performance data refers to the set of observable and quantifiable external behavioral information exhibited by students during classroom teaching; cognitive interaction indicators refer to the set of characteristic indicators that characterize students' current knowledge processing level, cognitive load status, and learning focus; and cognitive load labels refer to the graded identifiers of students' cognitive interaction during classroom participation.

[0056] Preferably, in this embodiment, feature extraction is performed based on the cognitive load labels to obtain a chain of behavioral semantic units for students in the teaching scenario, with reference to... Figure 2 As shown in the figure, this is a flowchart illustrating the process of determining the behavioral semantic unit chain in some embodiments of this application. In this embodiment, determining the behavioral semantic unit chain can be achieved using the following steps:

[0057] In step S21, the multi-source teaching behavior data is aligned according to the timestamps of the cognitive load labels;

[0058] In step S22, the dynamic behavior data segment of students in the teaching scenario is determined;

[0059] In step S23, behavior detection is performed based on the dynamic behavior data segment and the multi-source teaching behavior data aligned with the timestamp to obtain the atomic semantic units of students in the teaching scenario; in step S24, the chain of behavioral semantic units of students in the teaching scenario is determined through the atomic semantic units.

[0060] In practice, firstly, millisecond-level timestamps are uniformly embedded into all multi-source teaching behavior data (including video behavior sequences, voice signal records, interactive system logs, etc.). Then, using the timestamps in the student cognitive load labels as anchors, alignment windows are established based on time periods, such as 10 seconds per window. Within each window, behavior records overlapping with that time period are filtered, and video frame data, voice segments, and interactive events are merged into a unified data structure using student identification and time dimensions. For channels with different sampling frequencies (e.g., 25 frames per second for cameras, 1 semantic transcription per second for voice), interpolation mapping is used for data alignment. Finally, a unified time-axis aligned set of multi-source teaching behavior data segments is formed, with each set bound to a cognitive load label and corresponding to a teaching behavior time period. Next, based on the timestamp-aligned multi-source teaching behavior data, a preliminary analysis window is defined according to the time periods covered by cognitive load labels. Then, a sliding window mechanism is applied to extract the behavioral event sequences of each channel within each time window. For video behavior sequences, a pose estimation-based action boundary detection method is used to identify significant change points between adjacent frames, such as changes in posture or adjustments in sitting position. For the audio channel, voice activity segments are extracted as potential speaking or listening responses. For interaction logs, continuous event segments composed of operation events (such as clicking, swiping, and answering questions) are filtered. The behavioral event sequences of each channel are fused, and a time period with continuous behavioral characteristics is extracted as the dynamic behavior data segment of students using an adaptive segmentation method based on content change rate. Then, in the visual channel, a pose estimation model (such as OpenPose or MediaPipe) is used to extract keypoint coordinate sequences. The action type is determined based on the movement trajectory and spatial position changes of keypoints in the keypoint coordinate sequence, such as "looking up," "raising a hand," and "looking down to write." In the speech channel, speech recognition tools (such as Tencent Speech-to-Text or Google Speech) are used to extract transcribed text. Keyword matching and sentiment analysis are performed on the semantic content to identify behaviors such as "actively answering," "talking to oneself," and "repeatedly listening." In the interaction channel, click paths and timestamp sequences are analyzed to identify interactive behaviors such as "browsing content," "answering questions," and "page switching." "Concentrated operations" and "intermittent responses" are distinguished by key point coordinate sequence breakpoints and operation density. The action combinations identified within each behavior segment are encoded into an atomic semantic unit. Each atomic semantic unit contains the action type, executing entity, time boundary, and associated cognitive load label.Finally, the time-ordered atomic semantic units are input into the behavioral semantic construction module, including: using a sequence analysis method based on a hidden Markov model to cluster and summarize multiple atomic behavior sequences; assigning semantic labels to each behavior sequence pattern, such as "focused learning chain", "distraction interference chain", and "repeated understanding chain"; and merging atomic semantic units with continuity and semantic consistency into behavioral semantic unit chains; each behavioral semantic unit chain includes start time, end time, semantic type, corresponding cognitive load level, and triggering environment information.

[0061] It should be noted that in this application, timestamp alignment refers to matching multi-source teaching behavior data according to a unified timeline, so that data from different acquisition channels can be analyzed in a linked manner within the same time window; dynamic behavior data segment refers to teaching behavior fragments that are continuous in time and coherent in behavior; atomic semantic unit refers to a specific teaching interaction action completed by a student within a short period of time, such as "looking at the screen", "writing with head down", "raising hand to speak", etc.; behavior semantic unit chain refers to a behavior sequence composed of multiple atomic semantic units that are continuously connected in the time dimension and have teaching semantic logical relationships.

[0062] In step S3, interactive description information of teachers and students in the teaching scenario is obtained, and the dynamic participation density of students when receiving knowledge is determined based on the interactive description information. Then, the teaching response effectiveness of students in the teaching scenario when responding to teaching strategies is determined based on the dynamic participation density.

[0063] In this embodiment, obtaining the interactive description information of teachers and students in the teaching scenario can be achieved through the following steps:

[0064] Synchronously collect multimodal teaching interaction streams and generate a sequence of interaction events between teachers and students in the teaching scenario;

[0065] Determine the semantics of teaching behaviors in the teaching scenario based on the sequence of interactive events;

[0066] Based on the semantics of the teaching behaviors, interactive descriptions of teacher-student interactions in the teaching scenario are generated.

[0067] In practice, the process involves several steps. First, during synchronous data acquisition, cameras are deployed in the teaching environment to capture the physical behaviors and spatial position changes of teachers and students. Microphone arrays are installed to acquire audio streams for identifying speakers and semantic content. Teaching system logs are accessed, including records of clicks on answer questions, raising hands, and electronic blackboard operations. If touch devices are present, touch trajectories and interactive gestures are integrated. Millisecond-level unified timestamps are added to all acquisition channels. Window-level matching (e.g., a 1-second sliding window) is performed on timestamped data segments from different channels. Within the same time window, events are merged into a single interactive event record based on priority and hierarchy. Finally, an interactive event sequence is constructed chronologically, such as "teacher explains → student nods → student answers → teacher provides feedback." Then, using the sequence of interactive events as input, local event fragments are extracted through a sliding window approach. A sequence labeling model based on conditional random fields is applied to semantically label the events within these fragments, such as "questioning behavior fragment," "answering response fragment," and "error correction feedback fragment." The labeling results are categorized to identify semantic tags for teaching behaviors, and contextual information, such as "current knowledge point," "interaction location (opening, middle, closing)," and "number of teacher-student interactions," is added to the identified semantic behaviors. This forms a semantic structure for teaching behaviors that includes semantic categories, initiators, response chains, and teaching environment labels. This structure is then used as the semantic framework for teaching behaviors within the teaching scenario. Finally, the semantics of teaching behaviors over a specific time period, such as 5 minutes per segment, are summarized. The frequency of each type of interactive behavior is then statistically analyzed (e.g., number of teacher questions, number of student responses). The interaction direction (teacher-initiated, student-initiated), whether it is a closed loop (with or without feedback), and the duration of each type of behavior are also labeled. Next, we introduce an interaction density indicator: the number of effective interactive behaviors occurring per unit of time; an interaction completeness indicator: whether the effective interaction chain includes initiation, response, and feedback; and an interaction diversity indicator: the coverage of different types of interactive behaviors (such as asking questions, answering questions, commenting, and collaborating). These indicators are then structured to generate interactive description information. For example, based on this interactive description information, interactive description statements can be automatically written using rule templates or natural language generation models. In this lesson, the teacher initiated 16 questions about knowledge points, 87% of which received effective responses from students; students initiated 4 proactive inquiries, resulting in a high classroom interaction density and a feedback loop closure rate of 92%.

[0068] It should be noted that, in this application, multimodal teaching interaction flow refers to a continuous information flow that expresses the interactive behavior between teachers and students, composed of multiple information carriers (such as images, voice, click logs, touch behavior, screen projection data, etc.) during the teaching process; interactive event sequence refers to a time series set formed by recording, sorting, and classifying various interactive behaviors with teaching significance contained in the multimodal teaching interaction flow as event units; teaching behavior semantics refers to the interactive semantic units of teaching intentions and teaching process nodes; and interactive description information refers to structured data of the interactive characteristics of teachers and students during the teaching process.

[0069] Preferably, in this embodiment, the dynamic participation density of students when receiving knowledge is determined based on the interactive description information, with reference to... Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining dynamic participation density in some embodiments of this application. In this embodiment, determining dynamic participation density can be achieved using the following steps:

[0070] In step S31, response time markers in the teaching scenario are extracted from the interactive description information;

[0071] In step S32, the dynamic participation range of students when receiving knowledge is determined;

[0072] In step S33, the student's active response index when receiving knowledge is determined based on the response time marker and the dynamic participation interval;

[0073] In step S34, the dynamic participation density of students when receiving knowledge is determined based on the active response index.

[0074] In practical implementation, firstly, "knowledge transfer behaviors" are identified in the interactive description information, including lecturing, example explanation, and task assignment. For each knowledge transfer event initiated by the teacher, its start time is extracted as the "transfer starting point." Then, it is calculated whether students exhibit interactive behaviors such as voice response, answering questions, eye contact, or note-taking within a limited time window (e.g., 0 to 15 seconds) after the knowledge transfer event. Events with valid responses are labeled "valid response," and the response delay time is recorded to form a response timeliness marker. Next, each lecturing behavior of the teacher during the teaching process is used as an anchor behavior node and labeled as a "knowledge input event." Then, based on the interaction model, the effective response range for each input event is set, for example, extending it by 10 to 30 seconds as a dynamic participation range. All student behavior flows are filtered by timestamp, retaining only behavior events appearing within each dynamic participation range, and binding them with the corresponding knowledge input events. After binding, the dynamic participation range of students when receiving knowledge transfer is obtained. Then, student behavioral events are extracted from the dynamic participation interval and categorized according to behavior type: verbal (answering, asking questions), action (raising hands, nodding), operational (answering questions, clicking), and perceptual (eye following, taking notes). Each type of response is assigned a weight, for example, 2 points for actively answering a question and 1 point for nodding. A delay correction factor is constructed based on response delay time, for example, a score is multiplied by 1.2 for responses within 3 seconds and by 0.8 for responses beyond 10 seconds. The "response score" corresponding to each knowledge transfer event is calculated, and each student's response behavior is mapped to an active response index. Finally, the entire teaching process is divided into several time periods (e.g., one unit every 5 minutes). Within each time period, the total score of students' active response index is calculated. The response score density value for that time period is calculated, i.e., the score divided by the time length. A standard deviation analysis is performed on the response score density values ​​of all students in the class to establish a baseline. The response score density value of each student is normalized to the class average to obtain the dynamic participation density.

[0075] It should be noted that in this application, the response time marker is a time delay characteristic information indicating the student's response to the teacher's knowledge transmission behavior within a specific period of the teaching process; the dynamic participation interval refers to the range of time during which students can respond, extending backward from the teacher's knowledge transmission behavior on the classroom timeline; the active response indicator is a quality evaluation quantity that measures whether students can produce meaningful output after cognitive input; and the dynamic participation density is an evaluation indicator of the total amount of effective feedback that students give to the knowledge content transmitted by the teacher per unit time within a teaching cycle.

[0076] In this embodiment, determining the teaching response effectiveness of students in a teaching scenario based on the dynamic participation density can be achieved through the following steps:

[0077] The density fluctuation characteristics of students in teaching strategies are determined based on the dynamic participation density.

[0078] Based on the density fluctuation characteristics, determine the collaborative response rules when students respond to teaching strategies;

[0079] The teaching response effectiveness of students responding to teaching strategies in a teaching scenario is determined based on the collaborative response rules.

[0080] In practice, firstly, teaching strategy labels are preset during the teaching process, such as lecturing, interactive Q&A, group discussion, and scenario simulation. Then, time labels are added to each strategy activity on the time axis, and the dynamic participation density is mapped to the time period corresponding to each teaching strategy, forming a sequence of "strategy type-time segment-density value". This sequence of "strategy type-time segment-density value" is plotted as a curve, and the maximum slope of the curve is obtained as the density fluctuation characteristic of students in the teaching strategy. Then, students are divided into multiple subgroups based on optional dimensions such as academic performance, personality preferences, or historical participation behavior models. The density fluctuation characteristics of all students within each subgroup under the same teaching strategy are time-aligned and normalized to ensure that participation density is compared under the same standard dimension. The similarity between fluctuation curves is evaluated using Dynamic Time Warping or multidimensional Euclidean distance calculation. It is determined whether there are obvious group response patterns with synchronous peaks, synchronous troughs, or reverse trends in participation. If such patterns exist, they are used as collaborative response rules, such as "scenario simulation strategies easily trigger overall participation peaks, but the duration is short." Finally, all teaching strategies in the teaching scenario are mapped one-to-one with the corresponding collaborative response rules for students, constructing a data structure of "strategy name - response rule - student group - density response." Three response effectiveness dimensions are defined: strong response (high density, consistent behavior), medium response (moderate fluctuation, partially consistent), and weak response (low density or scattered behavior). Each teaching strategy is scored in each group based on the degree of matching between the actual fluctuation characteristics and the coordination rules. The overall response of the teaching strategy is comprehensively scored to obtain the teaching response effectiveness of the teaching strategy in the teaching scenario.

[0081] It should be noted that, in this application, density fluctuation characteristics refer to the trend and degree of fluctuation of students' dynamic participation density over time under different teaching strategy interventions; collaborative response rules refer to indicators of whether students produce a consistent response pattern under specific teaching strategy interventions; and teaching response efficacy refers to indicators of students' participation quality, response density, and behavioral consistency in response to specific teaching strategies applied by teachers during the teaching process.

[0082] In step S4, a teacher competency assessment profile based on teaching performance is determined according to the behavioral semantic unit chain and the teaching response effectiveness, and a teaching assessment report is generated based on the teacher competency assessment profile.

[0083] In this embodiment, determining the teacher competency assessment profile based on teaching performance according to the behavioral semantic unit chain and the teaching response effectiveness can be achieved through the following steps:

[0084] The knowledge transfer trajectory during the implementation of the teaching strategy is analyzed based on the behavioral semantic unit chain; the cognitive feedback attributes of students in the teaching scenario are determined through the teaching response effectiveness.

[0085] A teacher competency assessment profile based on teaching performance is determined based on the knowledge transfer trajectory and the cognitive feedback attributes.

[0086] In practical implementation, firstly, based on the student's behavioral semantic unit chain, behavioral segments that are triggered by teacher behavior and significantly responded to by students are extracted, such as raising hands after asking a question, taking notes after explanation, and operating after demonstration. Each explicit behavioral segment is back-mapped to the corresponding teacher teaching strategy type, such as lecturing, guiding, discussion, and heuristic questioning. Rule matching is performed using a teaching behavior coding table. Each teaching strategy node is labeled with the name and category of the knowledge point being transmitted, such as concept, principle, skill, and problem-solving, forming a teaching content unit. Then, the teaching strategy nodes and student behavioral response nodes are linked together according to the timestamp sequence to construct a path sequence of "strategy behavior - response behavior - knowledge point," thus obtaining the knowledge transmission trajectory. Then, relevant indicators such as participation density, behavioral delay, and interaction depth are extracted from the teaching response effectiveness. High density, rapid response, and rich behavior are marked as "high cognitive feedback"; medium density and intermittent response are marked as "medium cognitive feedback"; and no response or significant lag is marked as "low cognitive feedback." The feedback annotations of students during the transmission of different knowledge points are summarized, and the proportions of high, medium, and low feedback are statistically analyzed according to the knowledge point dimension, forming the cognitive feedback attributes of students in the teaching scenario. Finally, assessment dimensions for teacher competency profiles are set, including: teaching strategy selection ability: the ability to match appropriate teaching strategies to different knowledge points; knowledge organization and expression ability: the coherence and logic of knowledge paths; student cognitive stimulation ability: the proportion of high cognitive feedback; and teaching control ability: the frequency and effectiveness of strategy adjustments in the face of low cognitive feedback. Then, a weighted scoring mechanism is used to score the teaching strategy selection ability, knowledge organization and expression ability, student cognitive stimulation ability, and teaching control ability. At the same time, a visual profile map is constructed, including radar charts, strategy distribution maps, feedback heat maps, etc., forming a three-level mapping model of "behavioral path-student feedback-teacher ability", and outputting the final teacher ability assessment profile.

[0087] It should be noted that, in this application, the knowledge transfer trajectory refers to the behavioral path by which teachers gradually transfer knowledge to students' cognitive structures through specific teaching behaviors in classroom teaching; the cognitive feedback attribute refers to the response characteristics of students in cognitive understanding, knowledge absorption, and behavioral expression under the stimulation of specific teaching strategies and knowledge content; and the teacher competence assessment profile refers to a comprehensive assessment model that reflects the teacher's competence level in dimensions such as strategy application, teaching control, and cognitive stimulation, extracted through a holistic analysis of the knowledge transfer path and student response quality in the teaching process.

[0088] Furthermore, in practical implementation, generating a teaching evaluation report based on the teacher competency assessment profile can be achieved in the following way: First, the data for each competency dimension in the teacher competency assessment profile is standardized and compared horizontally with historical teaching performance data to identify the teacher's strengths and areas for improvement in dimensions such as teaching strategy application, knowledge transfer logic, and cognitive stimulation efficiency. Then, based on a predefined multi-dimensional teaching evaluation report template, competency indicators, student feedback, typical teaching segments, and strategy suitability are dynamically embedded into the report structure, automatically generating a teaching evaluation report combining text and graphics.

[0089] It should be noted that, in this application, the teaching evaluation report refers to a comprehensive document that presents the teacher's teaching behavior, student cognitive feedback, and the results of the teaching effectiveness analysis within a specific teaching cycle.

[0090] Therefore, this application demonstrates that it can improve the accuracy of teaching assessments when the assessment results lack strategy adaptability. Specifically, by synchronously collecting multi-source teaching behavior data from teaching scenarios, it achieves full-dimensional coverage and time-synchronized labeling of student behavioral characteristics during the teaching process, ensuring the consistency and semantic coherence of the collected teaching behavior data. This allows for the construction of a unified behavioral data fusion framework under multi-source heterogeneous input conditions, effectively solving the problems of fragmented observation and temporal alignment difficulties in existing assessments. Furthermore, by determining the cognitive load labels of students participating in the classroom within the teaching scenario, it enables precise perception and dynamic mapping of the implicit cognitive states in student behavior, thereby constructing a behavioral semantic unit chain model based on cognitive load. This model effectively enhances the semantic correspondence between teaching behaviors and students' psychological states, improving the cognitive interpretability of behavioral analysis results. By acquiring interactive descriptions of teachers and students in teaching scenarios, it enables quantitative assessment of students' response activity under strategy guidance, effectively solving the problem of difficulty in capturing dynamic fluctuations in student participation in existing assessment methods. By constructing causal relationship paths between teachers' knowledge transfer trajectories and students' cognitive feedback attributes, it can form a teacher competency assessment profile oriented towards the teaching process. Through structured analysis of each competency dimension in the teacher competency assessment profile, it generates personalized teaching assessment reports, effectively improving the process traceability and result applicability of assessment content, thereby achieving a precise feedback mechanism oriented towards teaching effectiveness.

[0091] In summary, the technical solution adopted in this application can conduct dynamic, structured, and correlated assessments of teachers' teaching behaviors when the teaching assessment results lack strategy adaptability, thereby improving the accuracy of teaching assessments.

[0092] Example 2: This application provides an AI-enabled quantitative teaching assessment system based on multimodal data profiling, referencing... Figure 4 As shown in the figure, this is a modular structure diagram of an AI-empowered quantitative teaching assessment system based on multimodal data profiling, according to this embodiment of the present application. The assessment system includes:

[0093] Data acquisition module 100 is used to synchronously collect multi-source teaching behavior data in the teaching scenario;

[0094] The feature extraction module 200 is used to determine the cognitive load labels of students participating in the classroom in the teaching scenario, and to extract features from the multi-source teaching behavior data based on the cognitive load labels to obtain the behavioral semantic unit chain of students in the teaching scenario.

[0095] The evaluation response module 300 is used to acquire interactive description information of teachers and students in the teaching scenario, determine the dynamic participation density of students when receiving knowledge transmission based on the interactive description information, and then determine the teaching response effectiveness of students when responding to teaching strategies in the teaching scenario based on the dynamic participation density.

[0096] The report generation module 400 is used to determine a teacher competency assessment profile based on teaching performance according to the behavioral semantic unit chain and the teaching response effectiveness, and to generate a teaching assessment report based on the teacher competency assessment profile.

[0097] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0099] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. An AI-enabled quantitative teaching evaluation method based on multi-modal data profiling, characterized in that, The evaluation method comprises the following steps: Synchronously collecting multi-source teaching behavior data in a teaching scene; Determining a cognitive load label of student participation in a classroom in the teaching scene, and performing feature abstraction on the multi-source teaching behavior data according to the cognitive load label to obtain a behavior semantic unit chain of the student in the teaching scene; Wherein, the feature abstraction according to the cognitive load label to obtain the behavior semantic unit chain of the student in the teaching scene specifically comprises: Aligning the multi-source teaching behavior data according to the time stamp of the cognitive load label; Determining a dynamic behavior data segment of the student in the teaching scene; Performing behavior detection on the dynamic behavior data segment and the multi-source teaching behavior data aligned by the time stamp to obtain an atomic semantic unit of the student in the teaching scene; Determining a behavior semantic unit chain of the student in the teaching scene through the atomic semantic unit, wherein a sequence analysis method based on a hidden Markov model is used to cluster and induce multiple atomic behavior sequences, a semantic label is assigned to each behavior sequence pattern, and atomic semantic units with continuity and semantic consistency are merged into a behavior semantic unit chain; each behavior semantic unit chain contains a start time, an end time, a semantic type, a corresponding cognitive load level and trigger environment information; Obtaining interaction description information of teachers and students in the teaching scene, determining a dynamic participation density of the student in accepting knowledge transmission according to the interaction description information, and further determining a teaching response efficiency of the student in responding to a teaching strategy in the teaching scene according to the dynamic participation density; Wherein, the determination of the teaching response efficiency of the student in responding to the teaching strategy in the teaching scene according to the dynamic participation density specifically comprises: Determining a density fluctuation feature of the student in the teaching strategy according to the dynamic participation density, wherein a teaching strategy label is preset in the teaching process, and a time label is marked for each strategy activity on a time axis, the dynamic participation density is mapped to each time segment corresponding to the teaching strategy to form a sequence of "strategy type-time segment-density value", and the sequence of "strategy type-time segment-density value" is drawn into a curve to obtain a maximum slope value in the curve as the density fluctuation feature of the student in the teaching strategy; Determining a collaborative response rule of the student in responding to the teaching strategy according to the density fluctuation feature; Determining the teaching response efficiency of the student in responding to the teaching strategy in the teaching scene according to the collaborative response rule; Wherein, the teaching response efficiency refers to an index of participation quality, reaction density and behavior coherence of the student in the teaching process in response to the teaching strategy applied by the teacher; Determining a teacher ability evaluation portrait based on teaching performance according to the behavior semantic unit chain and the teaching response efficiency, and generating a teaching evaluation report according to the teacher ability evaluation portrait; wherein, the determination of the teacher ability evaluation portrait based on teaching performance according to the behavior semantic unit chain and the teaching response efficiency specifically comprises: Analyzing a knowledge transmission trajectory when the teaching strategy is implemented according to the behavior semantic unit chain; determining a cognitive feedback attribute of the student in the teaching scene through the teaching response efficiency; Determining the teacher ability evaluation portrait based on teaching performance according to the knowledge transmission trajectory and the cognitive feedback attribute.

2. The AI-enabled quantitative teaching evaluation method based on multi-modal data portraits of claim 1, wherein, The multi-source teaching behavior data refers to a set of original information collected by different types of sensing devices in a teaching scene, which is used to depict the teaching behavior of teachers and the learning behavior of students.

3. The AI-enabled quantitative teaching evaluation method based on multi-modal data portraits of claim 1, wherein, The cognitive load label of the student participating in the classroom in the teaching scene is determined, and specifically includes: Collecting behavior performance data of the student participating in the classroom in the teaching scene; According to the behavior performance data, the cognitive interaction index of the cognitive state of the student is determined; According to the cognitive interaction index, the cognitive load label of the student participating in the classroom in the teaching scene is determined.

4. The AI-enabled quantitative teaching evaluation method based on multi-modal data portraits of claim 1, wherein, The interactive description information of teachers and students in the teaching scene is obtained, specifically including: Synchronously collecting multi-modal teaching interaction flow and generating an interactive event sequence of teachers and students in the teaching scene; According to the interactive event sequence, the teaching behavior semantics in the teaching scene is determined; According to the teaching behavior semantics, the interactive description information of teachers and students in the teaching scene is generated.

5. The AI-enabled quantitative teaching evaluation method based on multi-modal data portraits of claim 1, wherein, The interactive description information refers to the structured data of the interaction characteristics of teachers and students in the teaching process.

6. The AI-enabled quantitative teaching evaluation method based on multi-modal data portraits of claim 1, wherein, The dynamic participation density refers to an evaluation index of the total amount of effective feedback of students to the knowledge content transmitted by teachers in a unit of time within a teaching cycle.

7. An AI-enabled quantitative teaching evaluation system based on multi-modal data portraits, configured to perform an AI-enabled quantitative teaching evaluation method based on multi-modal data portraits according to any one of claims 1 to 6. The evaluation system includes: A data collection module for synchronously collecting multi-source teaching behavior data in a teaching scene; A feature extraction module for determining the cognitive load label of the student participating in the classroom in the teaching scene, and performing feature extraction on the multi-source teaching behavior data according to the cognitive load label to obtain a behavior semantics unit chain of the student in the teaching scene; An evaluation response module for obtaining the interactive description information of teachers and students in the teaching scene, determining the dynamic participation density of the student when receiving knowledge transmission according to the interactive description information, and further determining the teaching response efficiency of the student when responding to the teaching strategy in the teaching scene according to the dynamic participation density; A report generation module for determining a teacher ability evaluation portrait based on teaching performance according to the behavior semantics unit chain and the teaching response efficiency, and generating a teaching evaluation report according to the teacher ability evaluation portrait.

Images